Welcome everybody. My name's Alan Stewart, I'm the CFO of ShotSpotter. Thank you for joining us today in our day two of investor presentations. Hopefully, you'll find the information interesting. Before we go too far, I do need to remind everybody that we are going to be sharing some forward-looking statements. This presentation does contain forward-looking statements within the meaning of the safe harbor provisions of the SEC documents. Please, listen to what we say, but know that some of this does have information that includes the future. Before we get started, I'll go ahead give you a brief summary of the agenda. Contrary to what we did in day one, day two, we're going to talk about our sales, our pipeline review. We're going to go through technology, operations, and support. We're also going to give you a product demonstration of our Connect product.
I will give an update of the financials, and then we will also have, at the very end, and hopefully time left over to do a Q&A session. Hopefully that gives us at least 15 to 20 minutes, where if you ask questions during the day or during the presentation, we will answer them at the end. We may not actually answer the presentation or answer the questions when you ask them, we'll hold them to the end. Before we get started, I'll go ahead and turn it over to Gary.
Good afternoon. Thank you, Alan. I appreciate the opportunity to present to you this afternoon. Over the next 30 minutes or so, I'm going to give you an update on ShotSpotter's sales program, including a very brief reflection on the company's accomplishments from 2020. I'm going to summarize the company's sales objectives for 2021 in the context of some of the market forces that are shaping our sales program here this year. I'm going to talk briefly about the transformation that is underway to shift our sales model and organization from one that was aligned around being a domestic gunshot detection vendor to one that is aligned to grow our business as a global Precision Policing solutions provider.
I'm going to wrap up with a profile of our sales pipeline and talk about each of the sales funnels that are being developed to drive bookings and revenue across this expanding product suite. Getting started here. Looking back at 2020. 2020 was no doubt one of the most challenging years we've faced from a sales and business development standpoint. I'm not going to spend a lot of time talking about those challenges, we did want to spend a few minutes talking about some of the key accomplishments that we achieved in spite of some of those challenges. Working from left to right on this chart, I'm pleased to report that we finished the year at 107% of plan for Respond bookings and with 68 new Respond miles booked for the year.
This chart tells much of the story as agencies shuttered, sent command staff to work from home, and put many technology contracts on hold to deal with more pressing or urgent matters dealing with the COVID pandemic. The result was very few deals were actually canceled, but many of the opportunities we were working slid to the right, many into Q4 and many into 2021. We finished the year with Q4 being a record quarter with respect to Respond miles. In addition, we signed 12 brand new accounts, including several very strategic wins for the company. Most notably, City of Detroit, establishing our first anchor customer in the state of Michigan.
Harris County, Texas, which is the county surrounding the city of Houston, the largest county in Texas, the third-largest county in the U.S., and Broward County, Florida, the second-largest county in Florida, and the 17th largest county in the U.S. We launched a new program focused on smaller, lower-tier agencies in 2020, and in doing so, we brought in six new Tier four accounts. Tier four accounts to us are agencies that have a gun violence problem, but are agencies with less than 100 sworn officers. In spite of COVID and the distractions and uncertainty that came with the pandemic, we also helped nine existing customers enter into contracts to expand their ShotSpotter coverage. We entered into two very strategic pilots, one with the City of Houston, the fourth-largest city in the U.S., the other with the City of Ferguson, Missouri, a very high-profile Tier four agency.
These two pilots targeting two key strategic accounts, each at opposite ends of our domestic target addressable market. The combination of Harris County and the City of Houston represent a major step forward for ShotSpotter establishing a foothold in the Lone Star State. A new state for the company, a state where we had no active ShotSpotter service prior to these contracts, a state that represents a huge growth opportunity for ShotSpotter with cities like Dallas, Fort Worth, Austin, San Antonio, in a state that has a total of 41 municipalities with populations greater than 100,000 people. We also saw growing interest from a number of retail, distribution, transportation enterprises as more and more companies are seeing a need to take responsibility for protecting their employees, their customers, their profits, and their brand.
We wrapped up 2020 with a sales pipeline of 374 opportunities totaling over $66 million in estimated value. We're going to talk more about that in a bit. Looking over at customer retention, very proud of the customer commitment and customer loyalty that was tested and proven in 2020. Again, in spite of COVID and the distractions and uncertainty and budget crisis that came with this pandemic, we saw many customers get creative, get aggressive, bend over backwards to continue their ShotSpotter service. We heard on a number of occasions, especially on the heels of the growing violence that unfolded in the summer of last year, we had a number of customers basically express that they were facing a huge budget crisis, but they needed ShotSpotter more than ever.
By launching a proactive program to identify customers that were at risk, collaborating with those customers, getting creative in some cases, and helping them work through their budget crisis by driving modest price increases wherever we could and securing multi-year renewals wherever possible, we drove renewals to just over $22 million in bookings against a plan of just over $16 million, or finished the year at 134% of renewal bookings for the year. At the same time, by sitting down and saying, "We know you're facing some big challenges, but we're here for the long haul, and we're prepared to roll up our sleeves and work through ways to navigate this temporary budget crisis together," we also forged some lasting partnerships with a number of agencies who truly appreciated our commitment to them and the communities they are serving.
In the end, we only lost three accounts over the course of the year. When you look at the net impact on 2020 GAAP revenue, and you weigh the positive adjustments, the negative adjustments in those few lost accounts, we finished the year with a net attrition of about $358,000. We feel really good about how we finished this very challenging year. Let's now look forward. Moving forward. As Ralph and Sam referenced both on Tuesday, the company's made a great investment to position ShotSpotter with a new and expanded role in the public safety marketplace, and that is a role of a global Precision Policing solutions provider. As you know, we've traditionally offered Respond to alert law enforcement to gunfire and thereby assist in protecting citizens from gun violence, enhancing officer safety, improving evidence recovery, rendering aid to victims, and enhancing community trust in law enforcement.
That has been our sweet spot, our center of gravity for years. Naturally, we're very proud of what we've done to help law enforcement protect the communities we serve. In 2019, we started expanding or augmenting that solution with Connect, allowing agencies to use historical data and other data sources to determine where their limited resources can best be deployed to deter crime. Deployed using a data-driven model and not traditional gut-based deployment models that can often be unknowingly biased. Now we're in the process of adding Investigate to this product suite, an investigative case management solution bringing data, analytical tools, collaboration tools, and robust reporting tools to improve case closure rates. Not just gun crime cases, all case types. Sales is on the pointy end of the arrow here.
We have a lot of wood behind that arrow. We have to transform sales from a domestic gunfire detection vendor to this global Precision Policing solutions provider. That means we have to be skilled, credible, and competent at representing multiple products. It means we have to be comfortable stepping into a new account and leading with gunshot detection or patrol management or case management, or all of the above. A number of wheels are in motion to transform the ShotSpotter sales engine. Historically, and still to this day, we're organized around geographic regions with regional sales directors who own everything we do in each region. We worked for years in largely a distributed or geographic sales model with a couple of key team members available to support these regions. Starting late in 2019 with the introduction of Connect, we began building a more structured sales model.
We're still working through the regions, but we started building a much broader sales support infrastructure to support these regions and drive sales through the regions. We hired a new VP of regional sales to provide leadership, strategy, and process across these regions. We created a new role to exclusively pursue lower-tier Respond sales opportunities. We formalized the sales operations team with proposal support, analytic support, demo support, and ownership for many of the administrative tasks required to drive renewals. Most importantly, we established the product solutions group with subject matter experts or sales consultants for each product offering in the suite, and created positions for Respond, Security, and Connect. We're now taking the next step in building that sales organization to represent ShotSpotter as a global precision policing solution provider by expanding or building upon that infrastructure we laid just over a year ago.
We've moved Latin American sales into this organization, we're expanding sales operations, solution sales in the regions with the resources, some new hires, and some traditional or transition from LEEDs that will be dedicated to driving Investigate contracts through these regions. There are many steps we're taking here in early 2021 to drive these different products that now make up the ShotSpotter Precision Policing suite. In the limited time we have together, we just wanted to highlight what we felt were some of the higher priority steps that we're taking in each product area. Starting with Respond, with the impact that COVID has had on many municipal budgets, more and more agencies are exploring federal grant programs. We have engaged a company that does nothing but write grant applications.
We're going to bring them into select opportunities where we know the customer wants ShotSpotter, but the customer cannot secure the required local funding, and needs assistance to properly apply for the appropriate federal grant program. As Sam mentioned on Tuesday, the American Rescue Plan Act is making over $130 billion available to municipalities across the country. We're already working aggressively with a number of agencies, educating them on how to secure their fair share of that funding, and of course, for ShotSpotter. We're building a pipeline of lower-tier Respond opportunities with the expectation that we will build upon the success achieved last year one of this program, to drive more logos, more new accounts every year moving forward.
We're working a number of very exciting security opportunities that really represent new markets like retail, distribution, sports, and entertainment venues that represent significant growth opportunities in the months and years ahead. Looking to the right at Connect, one of our highest priorities on the Connect side is to capitalize on the early adopters that have already been signed, and work with those customers to obtain and document Connect successes and agency testimonials for use in our prospective sales efforts. We're seeing more and more communities take a more active voice in the technology decisions being made by local police departments, and many of the opposing positions are the result of being uninformed or misinformed regarding what technologies like Connect can do to help law enforcement deploy resources in ways that are more fair and equitable for the communities they serve.
We're putting a lot of attention into shaping our sales presentations and messaging to ensure that we're not only speaking to law enforcement and elected officials, but we're also speaking to local communities as well. The most significant step we're taking with respect to Connect sales in 2021 is for the first time, we're stepping outside of the existing ShotSpotter customer base and proactively reaching out to new agencies that are not ShotSpotter users today. Up to this point, we've really restricted our Connect sales efforts to existing ShotSpotter customers. This step expands our playing field for Connect by orders of magnitude moving forward. Looking at Investigate, our sales model for Investigate is really in its infancy. We're just beginning to define a sales program, recruit dedicated regional sales resources, define pricing, establish diagnostic processes, demo strategies, and presentation material.
We're currently evaluating options to establish two Investigate pilots, one within the ShotSpotter customer base, and one outside of the ShotSpotter customer base. Immediately following the product launch in Q2, we'll begin building an Investigate sales pipeline. Most of our sales efforts in the field this year will be about filling that sales funnel with strong leads, and we'll talk more about that in a bit. With that background, talk a little bit about sales objectives for 2021. No doubt there have been many forces, or there are many forces that are shaping or reshaping our market today.
With many local municipal budgets in crisis, with literally dozens of law enforcement leadership changes around the country, and growing opposition and a growing voice for many community groups and civilian oversight groups, we do face a number of challenges in the public safety market today. At the same time, increases in gun violence over the past year in recent active shooter events or tragedies are causing more and more agencies and more and more enterprises in the private sector to bring gunshot detection to the front burner. As we mentioned earlier, the American Rescue Plan Act is pushing $ billions to local municipalities, and many are already looking at how to take some of that money to protect their citizens from gun violence.
Staffing shortages of patrol resources, coupled with growing scrutiny or accountability from communities, are prompting more and more agencies to evaluate technologies like Connect to help them deploy their limited resources in ways that rely upon data and avoid the biases that can often undermine gut-level deployment models. There are many forces at play in the market today. Some are working for us, and some are working against us. In 2021, we set the goal of 90 new Respond miles for domestic sales. That is 80 traditional Respond miles and 10 from the lower tiers of the market. That equates to new Respond bookings totaling just under $7 million in new subscriptions for the year. We have over $33 million in contracts coming due for renewal in 2021.
With help from customer success, customer support, and others across the company, we'll be laser-focused on securing, hopefully, every one of those pending renewals. We've set goals of $600,000 for new Security bookings. That equates to approximately eight new Security contracts. We've set goals of $800,000 for new Connect bookings, and that too equates to approximately eight new Connect contracts. We have no sales bookings objectives for Investigate for 2021. As previously mentioned, we're really going to be focusing on building a sales pipeline for Investigate in 2021 and launching a sales program for Investigate. With those objectives in mind, I'd like to take a look at the ShotSpotter sales pipeline that has been developed, created to support these objectives. First, look at U.S. domestic Respond. We have a very healthy and robust Respond sales funnel. We have over 250 identified opportunities totaling over $50 million in estimated value.
I would point out that this funnel represents about 10 x the Respond sales bookings objectives for 2021. About 80% of these opportunities are new customers, 20% are expansions of existing customer systems. We currently have nine opportunities where we are in some stage of contract negotiations, meaning we have been selected and we're moving forward in creating the necessary paperwork to enter into a contract with that agency. In 2021, you will see us focus on expanding on the strong start we generated in the lower Tier four, Tier five market last year. You'll see us working to capitalize on the early successes we're having in states like Texas with the initial anchor tenants that we referenced a bit earlier to drive more tipping point scenarios like we've done in states like Ohio and Florida and some others.
You'll also see us focus more on select counties like Harris County and Broward County, where the county sheriff's office has responsibility for public safety in unincorporated areas around the county. Turning to U.S. domestic security opportunities. Here we're developing a much more diverse sales funnel. I say that because our security funnel is changing very quickly from what was largely a list of colleges and universities to a list that now contains some corporate campuses, big box retailers, some state capitals, and some distribution opportunities. We have over 25 identified opportunities totaling about $2.2 million in estimated value. This funnel represents about four times the Respond sales bookings objectives for 2021.
When you look at Connect, we have a very solid Connect sales funnel, but we expect to see this expand dramatically in 2021 as we venture outside our ShotSpotter customer base and open the aperture to the entire U.S. public safety market. We have over 60 identified opportunities totaling about $5.8 million in estimated value, making this funnel about seven times the Connect sales booking objectives for 2021. Again, that is before we've taken any steps to start selling outside of the ShotSpotter customer base. In 2021, you'll see us focus on building upon the successes of our early adopters to drive deeper into our existing customer base and begin selling Connect to agencies that have never done business with ShotSpotter before. There's a lot of upside here.
We're just really getting started with Connect and very confident that you're going to see this sales funnel grow over the course of the year. That takes us to Latin America, Caribbean sales. Here, we're talking about Central and South America as well as the Caribbean. We have a number of opportunities we're working, primarily in Brazil, Mexico, Colombia, and several Caribbean nations. We have over a dozen active opportunities identified totaling about $6 million in estimated value. This represents about six times the Respond sales booking objectives for this geographic region. Our big challenge here is COVID even still. Brazil, to a slightly lesser extent Mexico, is still reeling from the COVID pandemic, and it's been very difficult to capture and retain the attention of many government officials.
We expect that challenge to remain until many of these countries can roll out vaccines and start defining a clear path to normalcy for their citizens. Last but not least, our domestic Investigate opportunities. As mentioned before, here's where we're really just getting started, but it's all upside. We have a small handful of opportunities that were really being worked by the LEED s team prior to the acquisition. We have about a half a dozen opportunities we're working that total somewhere around $3.6 million in estimated value. These include several really exciting opportunities in the federal space, really a new market for ShotSpotter. As previously referenced, we've just begun the process of identifying two pilot customers, one existing ShotSpotter customer and one new. As we previously mentioned, this year is going to be all about building the sales funnel for Investigate.
We will start by approaching select ShotSpotter customers, but we'll quickly go beyond that to pursue leads across the entire U.S. public safety TAM, as well as some select state and federal agencies. This is a real exciting new opportunity for the sales organization. All new and all upside. That is a quick summary. I hope you find it helpful. Thank you for your time and your interest. I would like to close by letting you know that I've been selling technology solutions to public safety agencies for almost 30 years now. I don't recall ever being armed with a unique blend of new, exciting, and unique products that make up our Precision Policing suite today. We are uniquely positioned with products that no other company is providing, with sales and marketing teams that are truly the best in the public safety market.
We're selling in a market that is growing through tremendous change, but a market that is looking more and more for new and innovative ways, meaning innovative technologies, to protect and serve. I think 2021, and the next few years are going to be very exciting for all of us at ShotSpotter. I appreciate your time. With that, I'm going to turn the floor over to Mr. Paul Ames, our favorite Senior VP of Engineering.
Thanks, Gary. I'm going to spend the next 25 minutes or so walking you through our ShotSpotter Respond and Connect products from a technology perspective. I have with me a key member of my team, Simen Oestmo who's the data scientist primarily focused on Connect. You'll be hearing from Simen a little later. I'm going to start with ShotSpotter Respond and take you through a couple of slides that describe some really cool projects focused on driving up the quality of Respond service delivery. First, I want to take you through a quick recap of how wide area gunshot detection actually works. To do this, I'm going to use the Insight app that Sam mentioned on Tuesday that's delivered as part of the Respond service. Rather than attempting a high wire act in this virtual format, I pre-recorded a short video of a demo.
Take it away, Paul. This map view shows a fictional ShotSpotter Respond coverage area of about two square miles for a city that we internally call Metropolis West. This red line here shows you the boundary of the coverage area, and these squares on the map are the individual ShotSpotter acoustic sensor locations. As an internal user, I have access to a lot of data that allows me to assist in management of the delivery of our service, and specifically, these ShotSpotter locations are not divulged to the police department as dictated by our privacy policy. This yellow pin here shows the location of an individual gunshot incident that comprises of five individual shots. You'll notice that some of the sensors here are colored green. What this is indicating is that these sensors heard this particular gunshot incident.
The primary role of a sensor, an acoustic sensor, is to precisely timestamp the arrival of the acoustic event. The propagation time of the audio from the shooter location to this near sensor that's hiding under the pin is going to be almost zero. The propagation time from here down to here, this sensor that's about 2 km away, on the other hand, is going to be about three seconds. Let's take a look at some detail. What we're looking at here is a summary of the individual sensors that heard this particular incident, along with their distance away from where the incident was located. There were 20 sensors in all. It is the difference in time of arrival of the acoustic event at these different sensors at varying distances away from the shooter that allows us to precisely locate this particular incident.
After we've located the incident, it's time to classify. The first stage of classification is machine classification. We use a deep learning neural network classifier to take the metadata associated with the incident, run it through the classifier, and come up with a classification that allows us primarily to filter out incidents that we feel really confident are not gunshots. In this particular case, this particular incident is classified as 98% likely to be gunfire. Let's take a look at this section down here. What you see here are four waveforms from four sensors that heard the gunshot. It's these four waveforms, plus the audio, plus the metadata, plus the machine classification, that the human reviewer considers in their final classification step. If the reviewer believes that the acoustic event is a gunshot, they hit the publish button.
The incident is pushed to the patrol officers in the field that are using our application, and we're at the end of the process. This process from beginning to end takes typically less than 45 seconds. We'll just take a listen to this incident just to see if the machine classifier actually got it right. I think it did. Okay. Great demo, Paul. Thank you. Okay. Let's turn our attention to driving up performance. When Joe and Nasim talk later this afternoon, you'll see that we have a bit of a preoccupation with driving continuous improvement. With gunshots, whenever we miss or mislocate, we always investigate the root cause. In general, when we miss, it's because not enough sensors heard the sound. In general, when an incident is mislocated, there were enough sensors that heard the gunshot, but we didn't do a good enough job of calculating the result.
Calculating locations is no easy task, as Rob Calhoun, one of our co-founders, illustrated during last year's investor meeting. Motivated by the challenge and desire to take a big step forward, at the beginning of 2020, we started a project to rewrite some of the core algorithms, specifically to resolve two situations where our root cause analysis told us we could really move the needle. The first area of improvement was in situations where the calculated location was significantly off from the ground truth. These errors were often introduced by coincident, but unrelated sounds. Imagine a gunshot occurring in one area of the city, and a backfire occurring at the other end of the city. If the backfire was inadvertently included in the location calculation, the location would be significantly off.
At a high level, the solution there was to consider the physical locations of the sensors rather than relying solely on mathematical consistency. The second area of improvement required us to reassess our priorities when calculating locations. The old algorithms were optimized to calculate a location in the shortest possible time. In some instances, late-arriving data from sensors, perhaps because of network latency, were not considered as part of the solution, reducing the level of accuracy. The change that we made was to trade a second or two increase in publication time for improved accuracy, which it turns out better aligns with police department priorities. The new algorithms were completed midway through 2020, when the new and old algorithms were run in parallel in multiple cities, and the results were consistently good. The picture on the slide captures a multiple shooter incident.
The yellow pins represent the locations calculated by the old algorithms and the blue, the new. The ground truth from this particular agency confirmed that the blue pins were on the nose. Although the improvements you see here in this particular incident are fairly modest, they're actually less than 20 m, as Regan described on Tuesday, accurate locations drive evidence collection, recovery of weapons, and locating gunshot victims. Location accuracy is critically important. These new algorithms were rolled out to all customers during the second half of 2020. The technology team invested a fair amount of time in 2020 to surface KPIs through an internal dashboard. When an agency reports ground truth, this is added to our incident database. It can be used to assess some aspects of human reviewer performance. These on-the-job classification accuracy KPIs can, of course, be used to drive targeted training plans.
To execute these training plans, my team is currently working on the first version of a new training service. At a high level, the training service will assign challenges to reviewers. The reviewer uses the standard reviewer application they use in production, but are presented with a set of incidents curated for a specific training purpose. For example, differentiating between fireworks and gunfire. Ultimately, the training service gives us another set of KPIs. Our plan is to use both of these KPIs to make review workflow decisions in real time. We're piloting a feature called peer review, whereby policy or on-demand, a reviewer can request a peer to validate a classification decision. The assignment of this incident will be made by considering the KPIs associated with the individual reviewers on shift.
The goal is to assign the peer review task to someone who is more skilled at classification for this particular situation. All of these approaches drive up the quality of Respond service delivery. Let's talk about Connect. The primary goal of Connect is to help drive down crime by directing officers to patrol in the parts of the city where Connect assesses the highest risk of crime occurring. Wendy Ethridge will be showing a Connect demo later this afternoon that will give you a user perspective of the product. For now, let's focus on the core of Connect, the crime risk assessment model. Connect is a machine learning application. We use a machine learner to discover relationships between crime data and publicly available data that describes features of when and where the crime occurred.
The diagram on the slide shows at a high level the data we use and what we do with it. Let's start with the data. Unsurprisingly, there are multiple crime theories that attempt to explain when and where crime occurs. Risk terrain modeling suggests that proximity to physical features of the city, such as the location of schools, bars, overpasses, et cetera, drive the likelihood of crime. Temporal cycles suggest that day of week, time of day, season of the year, et cetera, are correlated with crime. Near repeat patterns suggest that crime occurring in one location is likely to be repeated subsequently in adjacent locations. Think, for example, about a rash of residential burglaries in a neighborhood. All of these theories have some value, and are used to varying degrees by police agencies.
For Connect, rather than selecting a specific crime theory, we consider all of them. The data we use for modeling are a superset of all the data that underpin the competing crime theories. This is a novel approach. We're effectively allowing the different crime theories to compete for the best prediction result. Let's look at the process, starting on the left-hand side of the diagram. First, we source and ingest a minimum of three years of historical data to train the crime risk assessment model. We test the model against six months of holdout data to confirm that it performs as we expect. Prior to each shift, we ingest updated data, such as the weather forecast for the next eight hours, and run them through the model to come up with crime predictions.
Finally, the crime predictions are weighted, converted to risk scores, and run through an allocation engine to create a patrol plan that you see on the right-hand side of the diagram. I'm going to hand it over to my colleague, Simen, who will contrast how Connect compares to other approaches for creating patrol plans, and take a look at how we mitigate against bias. Over to you, Simen.
Thank you, Paul. Okay, let's talk about bias mitigation. First, I want to provide some background. Generally, a significant amount of patrol time is dominated by responding to calls for service, which are calls initiated by the community for help related to violent crimes such as shootings, assaults, robberies, or non-violent crimes such as burglaries, or theft, or motor vehicle theft. When not engaged in responding to calls for service, officers have something called uncommitted time, which is when they have some degree of discretion on what to do or where to go, and the intent is to use this time proactively to deter crime or suspicious activity and gather input from the community. There are several formal methods by which police have traditionally directed their resources during this uncommitted time, they fall into three main buckets. You can see that on the slide here.
Gut-based policing, hotspot policing, and predictive. First two approaches are most frequently used by police today, and agencies with at least one crime analyst tend to do hotspot policing. While well-intended, these options have flaws that do not take into account potential harm to the community. This may leave citizens feeling overpoliced and discriminated against, and not actually maximizing the crime deterrent effect. These three methods mainly suffer from what I here call spatial bias. Spatial bias are spatial because they all have to do with location. In gut-based policing, the spatial bias comes in three forms. Bias because some communities call in crime more than others, and thus police attention is focused on those areas where people call the most or at all.
Number two is bias from police patrolling some areas more than others due to, quote-unquote, "knowing where crime is happening." This knowing can be due to more subjective bias or discriminatory behavior, unintentionally or not. The third one is reinforcement bias, which happens when police discover crime while on patrol, then that data gets used in the decision-making the next time the officer is patrolling. Discovered crimes are usually crime types such as drug crime, prostitution, trespassing, vandalism. In the second approach, the hotspot policing, aggregated data on where crime happens and how it clusters is used to determine where to go. This approach alleviates much of the subjective policing bias, but it still suffers from community calls for service bias and from reinforcement bias, since all types of crime, both discovered and called in by citizens, are used to model where crime is happening.
In the newer approach, the term predictive policing, subjective police knowledge is countered by relying on crime data with precise times and locations to predict where crimes will occur. However, if you are predicting crime types that are predominantly discovered by police, such as drug crimes, you are creating reinforcement bias where the police will loop back to the areas where they discover crimes themselves. The use of only crime events as variables to predict crime do not alleviate the spatial bias due to some communities calling about crime more than others. Now, in Connect, we are focused on limiting all three forms that the spatial bias takes. First, we model with Part I crimes. That is more serious crimes such as burglary, robbery, motor vehicle theft, aggravated assault, which limits reinforcement bias because these crime types are rarely discovered by the police out on patrol.
This means that we are limiting the effect that the discovered crimes have on police coming back or looping back to the places that they themselves have found the crime. We also use our own ShotSpotter Respond gunshot data to model the risk of gunfire inside of our coverage zone. This data source is much more objective in nature as it sidesteps the issue of people calling in gunshots and when and where police respond. Secondly, we consider multiple crime theories and patterns in our modeling, which Paul just described, such as risk terrain modeling, weather patterns, social cohesion, and seasonality. This helps mitigate spatial bias due to different calls for service rates among communities.
In other words, the use of these variables, such as weather, city features, and census data, helps us model the context of the crime, which ensures we consider risks in areas where no crime has been called in or is rarely called in. Finally, we use what we call an allocation engine in our post-modeling to increase fairness in patrolling. On the next slide, I will talk about that. The goal of the allocation engine is to help reduce the policing gap. What I mean by that is that we want to meter patrols in high-risk areas that often get visited by patrols and allocate more of the patrol time to underserved areas. The allocation engine uses risk scores that consist of past crime values in addition to the prediction of a crime occurring.
The past crime weights allows for an appropriate weighting scale between crimes of different seriousness to be considered when assessing the risk to the community. This helps prioritize areas with more serious violent crime, for example. The allocation engine considers all areas with risk above a set threshold and then randomizes the patrols' assignments among those areas to avoid oversaturation and looping of any particular area during a shift. The spread of patrol boxes among a range of high-risk areas helps reduce the policing gap, thus increasing fairness in police patrolling. Over to you, Paul.
Thanks, Simen. In the interest of time, I'll finish by quickly covering two of the five areas listed here that we're considering for Connect. First, the team is currently focused on launching a new Connect mobile app with a consistent user experience with our other ShotSpotter apps. The screenshot on the slide is the Connect app that's under development. This will drive consistency in user experience across our Precision Policing platform, including user authentication, how to reach out to support, map styling, et cetera. This app will be launched over the coming months. Second, Sam outlined yesterday two efficacy studies performed in Greensboro and Philadelphia. Both of these studies were designed and executed out of band from the product, meaning they required a tailored test design, execution, and analysis of results by data scientists. This kind of efficacy test is going to continue to be extremely important.
A lighter version of efficacy reporting would be extremely valuable as part of the Connect product. I believe that ShotSpotter is uniquely positioned to drive this kind of product forward because of what's in our DNA: helping police provide equal protection for all. I see it myself every day in team conversations that revolve around community impact, police trust, and bias mitigation. This really is an exciting time for Connect. That wraps up our technical look at ShotSpotter Respond and ShotSpotter Connect. Let me hand it over to my colleague, Joe Hawkins.
Thank you, Paul. As you all saw from Paul's and Simen's presentation, and as many of you know very well, ShotSpotter technology is pretty amazing in terms of what it does, how it works, and what it's capable of doing for our customers. Capability powered by great technology in and of itself does not translate automatically to performance out of the box when exposed to the real world, which is a chaotic and unpredictable system itself. Every city and every community, even different neighborhoods or districts within the same city, is different geographically, topographically, environmentally, even acoustically. Because of those differences, an acoustic gunshot location system must be implemented, managed, and maintained with careful consideration of the environment in which it's being deployed.
We often describe this as a blend of art and science, and the fusion of those two elements is a product of expert knowledge born of deep and broad experience. This, in a nutshell, is what operations does, deploying new ShotSpotter service to customers, and keeping it running reliably from that point forward. Now, service implementation and service assurance may sound like two different things, and they are. They are united by a common thread, and that is the mission that drives operations. We are a mission-driven organization, and that manifests in a few ways that I'll touch upon over the next few minutes. First and foremost, though, is what you see in red italics right there. We ensure the service our customers want and pay for works as promised or better.
To support that mission, operations is comprised of three somewhat autonomous functions, a three-legged stool, if you will, that nevertheless have some important synergies. Put together, they form a virtuous circle of continuous improvement, as you see illustrated in my corny little graphic there on the right. The first leg is service implementation, leading the project teams in the design, configuration, and deployment of new or expanded service to our public safety and security customers. The second leg is service assurance, monitoring, managing, and maintaining our rapidly expanding network of IoT-enabled acoustic sensors that are the backbone of our gunshot location service. The third leg is a commitment to continuous improvement. This is what completes the virtuous circle I mentioned. That perpetual machine, if you will, is propelled by the overarching principles of measure and learn from everything and apply what we learn to make things better.
I'll give you a few examples of how we put these principles into practice as we go. First, let's take a closer look at each of these three components. ShotSpotter service implementation for Respond, Connect, and our Security offerings is led by our project management organization. Project management is, of course, a professional discipline practiced by virtually every business in every industry in the world, whether that happens to be in technology, financial services, construction, research and higher education, or government. In fact, some of the most common project management tools and techniques, like the image you see there on the upper right, originated in the U.S. military decades ago. ShotSpotter's project management team is a little different in that our project managers wear two very different hats. The first is classic project management.
They lead the cross-functional, interdisciplinary project teams charged with planning, implementing, and activating service for our customers, including installation of sensors, provisioning and configuring our cloud-based compute and application services, and the onboarding and training of those customers so that they're ready to hit the ground, taking full advantage from day one at go live. Project managers own the master project plan, and they manage schedules, resources, risks, and change to make sure that high-quality service is delivered on time and on budget. The second hat is truly unique to ShotSpotter. Our project managers also design our acoustic sensor arrays, determining how many sensors we want and where we want them based on the unique characteristics of the planned coverage area, and identify the sites that we'd like to place a sensor, usually tall buildings, but sometimes utility poles or streetlights, again, depending on the physical topography of the neighborhood.
They have the responsibility of finding the owners or managers of these properties, selling them on the altruistic value of volunteering the use of their rooftop for us, and securing written permission for us to install. This is essentially a sales and community relationship job, and they're very good at it. Lastly, our project managers plan, organize, and conduct live fire tests with our customers. We call the plays, the cops shoot the guns, all very safely under carefully controlled conditions, but outdoors in public when it's advisable to do so. The purpose of these tests is generally to benchmark performance because we know where the shooter is standing, though they can be used for diagnostic purposes as well. It's the second hat the project manager wears where expertise born of experience really matters. Allow me to introduce you to our project management team.
That's the six gentlemen in the upper right there. Well, six of the eight in the picture anyway. The other two are former ShotSpotter project managers who moved on to other roles within the business. Once a project manager, always a project manager, and I really like that picture. Our project management team as it stands today has been designing and implementing ShotSpotter's acoustic gunshot location services for a long time, longer than anyone anywhere has ever done, in fact. 40 years collective experience working directly with law enforcement and the communities they serve. Approaching 200 deployment projects. There's over 275 separate ShotSpotter systems covering almost 800 sq mi now. From over 100 mi of Chicago and close to that across the five boroughs of New York City, to the smaller Tier four or five cities Gary described, like Monroe, Louisiana or Kankakee, Illinois.
From the densest and noisiest urban centers to the broader, flatter residential neighborhoods of smaller towns and semi-rural communities, and everything in between. By now, we've tackled just about every urban challenge that's been thrown at us many times over, and each new project we undertake today looks like many others we've already done. This harkens back to the third leg of the stool I mentioned earlier, continuous improvement. We have accumulated the world's largest knowledge base on the design and performance of acoustic gunshot location services. We know how big our coverage areas are. We know their acoustic environmental properties. We know how many sensors are there and how our service performs.
With every new project, we apply that knowledge to design and implementation of the next systems, ensuring we have the right number of sensors we want for and reliable service quality, but also optimizing for efficiency in terms of capital investment and time to go live. Speaking of optimizing for efficiency and time to go live, I described a little bit about the who behind our service implementation projects and what makes them special. Now I'll briefly show you what it looks like from a what and when perspective, otherwise known as our standard deployment project life cycle. As much as the design of our sensor arrays are customized to each community, like a bespoke suit, our approach to executing these projects is standardized, consistent, and repeatable. This lends itself to highly effective, high-performing project teams.
After a couple of 100 projects, team members know what to expect, what they need to do, and when. The pie chart you see on the right shows the basic elements at play in all projects, roughly approximating the relative work effort for each element. Permissions acquisition is generally the most labor-intensive and time-consuming, although onboarding customers, especially new customers who haven't put a lot of thought into policies and practices yet, takes more energy. The size of the permissions and onboarding effort can shrink or expand depending on the size of the contracted coverage, 3 mi versus 10 mi, or the size of the law enforcement agency. Laid out over time in a schedule, we use concurrent engineering practices to parallelize as much work as possible to achieve the shortest possible duration from start to finish or from contract to go live.
In fact, whenever we can, with the agency's approval, we like to start our work before the contract is executed. We've found that this is a risk well worth taking, as permissions acquisition is almost always the long pole in the tent, and the lion's share of the project cost comes with sensor installation a bit downstream. The shortened overall time of the deployment cycle means the customer gets their service faster, a big win for them, and we recognize revenue sooner, a big win for us. While this particular model is specific to ShotSpotter Respond, Secure Campus, and Site Secure, a very similar project model exists for our ShotSpotter Connect. Though there, the critical path or long pole the tent is the data acquisition and creating, training, and validating of the crime risk prediction model for directed patrols.
Switching gears now to the second leg of the stool, service assurance is the principal domain of network and field operations. This small but mighty team is responsible for everything to do with the installation, monitoring, management, repair, upgrade, and replacement of acoustic sensors, and the other ancillary hardware like radios, power supplies, mounts, and so forth in the field. There are two distinct functions within this team, each staffed by specialists dedicated to those roles: field systems and field service. The field systems teams are the eyes and ears overseeing the health and performance of more than 18,000 sensors in the field today, and that number's expected to grow to 20,000 or more this year.
They remotely monitor and manage our network of sensors 24/7, 365 days a year, with the help of a lot of automated systems, provide second-level diagnostics anytime we see devices or sensor arrays showing any indication of potential trouble. Keep in mind, our sensors, while built and ruggedized for long-term outdoor environments and are generally extremely reliable, still rely on electrical power that we don't directly control. It's not uncommon for sensors to occasionally drop off the network. Again, that's why we overbuild our sensor arrays in the first place. If a sensor goes offline and it doesn't come back in a reasonable amount of time, as they usually do in the event of a temporary power outage, our field systems engineers troubleshoot and attempt to restore the sensor remotely. If they cannot, they hand it off to the field service team.
The field service team are the eyes and ears. If the field systems team are the eyes and ears, you could say the field service team are the hands and feet of the team. Our field service managers are responsible for scheduling, dispatch, and management of field service technicians for both sensor installation and maintenance. We employ a team of our own field service techs and electricians, and a cadre of third-party service partners to provide the regional coverage we need. Our internal techs serve as installation team leads on large projects. They train our third-party partner techs and provide dedicated coverage in the largest, most densely populated metro areas like New York City, Chicago, South Florida, and the Bay Area in San Francisco.
Now, you noticed that I said we expect to grow our footprint from 18,000 to 20,000 sensors this year, yet I've got 5,000 new installs noted on this slide. The difference is a very large maintenance upgrade project that we've just kicked off, necessitated by both AT&T and Verizon's announced sunsets of the 3G cellular band in 2022. AT&T 3G service will be discontinued next February. Verizon will follow suit by the end of next year. Our field service team will be systematically replacing all of our 3G sensors at a clip of about 250 to 300 a month, with a goal of finishing a month or two ahead of their respective sunsets. Our network and field operations organization plays a big role in the third leg of our stool, the continuous improvement program.
Before I dive into that, I need to tell you about one last vital part of the organization, and that's the service provisioning and engineering function. Service provisioning engineering is a bit of a bridge between service implementation and service assurance, having a solid foothold at both camps. Just as you see here, it is principally responsible for the allocation, setup, configuration of the cloud-based database network and application services that our sensors connect to for gunshot location services, and that our customers connect to when they log in to get access to the ShotSpotter alerts and historical gunshot data. In the case of ShotSpotter Connect, they are comparable services for connecting customers with the directed patrols and associated reports. This team manages our ShotSpotter Connect implementation projects as well.
Lastly, they manage one of our most important tools, ServiceMax, that I'll get into in just a moment. We use a number of technologies and operations to supplement and amplify our human capabilities. The three that I want to tell you about just briefly today are network monitoring, ServiceMax, and something we call Sapphire. We use two primary tools for network monitoring, diagnostics, and performance management: Prometheus from Grafana Labs and our own homegrown Status tool. Prometheus provides fully automated remote monitoring of sensors and services aggregated at the coverage area level and alerts us to events worth looking at 24/7.
In the event a pre-established service threshold is exceeded, say an outage on a local electrical grid that kills power to a large number of sensors, alerts are sent via email, text, or even a Slack channel to watchers who can begin to investigate and take corrective action and to notify customers who may be affected by a temporary disruption of their service. One of the features we really like about Prometheus is a powerful graphical user interface for examining and conducting root cause analysis for historical events, whether that might be a network or a systemic issue, or individual incidents of undetected gunfire. That historical data is the foundational building block to our continuous improvement program in that it provides visibility into patterns and trends that might suggest a chronic problem or gradual degradation of service that can be isolated and solved at its root cause.
Status, on the other hand, is an internally developed tool going back to the earliest days of ShotSpotter, and it does just what its name suggests: device-level operational state right now and real-time telemetry on several health indicators to enable health and performance diagnostics on individual sensors and a bird's eye view of sensor array health, both of which inform our immediate field service plans and priorities. ServiceMax is a tool with even greater strategic importance for us. It's an application that lives in the salesforce.com ecosystem and is designed for field service organizations like ours. We use it to track and manage service on our sensors and the sites that are installed, as well as the vendors and technicians who service them. Project managers use ServiceMax to track permissions on their new deployment projects, which lead to our field service team to schedule new sensor installs.
ServiceMax work orders make tracking all sensor touches, whether it's installation or maintenance in the field, easy. Most importantly, though, is that tracking the history of all of our assets from cradle to grave, including the people that work on them, in one system, provides the essential foundation for long-term sustainable quality improvement. It gives us the rich historical data we need to identify the most common causes of maintenance and repair trips, whether it's a failure-prone component or the way a particular vendor connects to power, so we can solve those problems and forever eliminate them as drivers of maintenance costs. Because field service maintenance is the largest driver of operating expense, that systemic quality improvement, in turn, will lead to long-term sustainable growth and profitability and gross margins. Lastly, Sapphire is the code name for a new interactive modeling tool we've been working on internally.
You've heard me say in the past that sensor array design is a blend of art and science. Well, Sapphire aims to move us a little away from the art and more towards science, leveraging the massive amounts of data we have accumulated over the years about the performance of the many ShotSpotter systems we have implemented and been operating for years. We've identified two primary use cases so far. First is a sensor array design aid for our project managers, allowing them to inspect the as-built design and performance of existing coverage areas that are comparable to the new service we're implementing, so they can design for the redundancy, resilience, and performance we want, while also optimizing for efficiencies, again, in terms of capital investment and speed to market.
Taken together, these technologies are the keys to our progress towards our long-term objective of delivering the highest quality service while reducing the cost to do so. With that, I would like to introduce Nasim, who will take you through the customer support and professional service organization.
Thank you very much, Joe. Good afternoon. My name is Nasim Golzadeh, I lead our customer support and professional services organization. Well, under the umbrella of customer support and professional services, we really have five different teams that work together and provide service to our customers. Those teams are Incident Review Center, customer support, forensic services and litigation support, integration services, and training services. Today, we're going to briefly focus on the first three. Let me begin with IRC, which is the heart and the human engine behind incident classification and review. Many of you are already familiar with the IRC. IRC's operations are designed around 24/7, 365 service to our customers, and again, that is just a continuous operation. In addition to their primary role for incident review and classification, they are also in charge of providing customer support, Tier one customer support to our customers.
Staffing is always extremely important for the IRC because it is important for them to be exactly staffed and appropriately staffed. We use a smart way of managing that staffing by analyzing our historical trends and coming up with models that design our staffing based on the anticipated volume on different days of the week and hours of the day. That model has served us really well and has made us extremely cost-effective and efficient. Historically, our IRC has been located in our headquarters in Newark, California. In the past couple of years, we had the vision of having another center of gravity, another center of excellence on the East Coast, which would provide us with two benefits. First, it is having resiliency of locating one of the most important teams within our organization in two separate physical locations.
Also being able to host guests and visitors, prospects, customers, elected officials, and others to come and see IRC in practice and firsthand, because we know that experience is extremely powerful. Well, in 2020, we achieved that goal, and we have a really nice, impressive facility and office in Washington, D.C., within walking distance from U.S. Capitol. As soon as it is safe and it's practical, we're going to have a grand opening event. When we think about incident review and classification, there are two really important factors that go hand in hand and are equally important. The first one, of course, is accuracy because in our business, there is very little tolerance for false negatives and false positives. The second factor is speed because every second counts. Our team is extremely capable.
They perform this function on average in about 45 seconds from trigger pull to when the incident is published to the police department. The reason they can do that is twofold. The first set of reasons for that has to do with our technology. Paul covered some of this. For example, the engine behind the machine classification is extremely sophisticated, and every time that an incident is presented to human for review, it goes with a machine classification and a confidence level tied to that. This is exactly the information we use for our suppression tool to control the incoming incidents when there are unusual activity, for example, extreme firework activities, wind, or helicopters. That has made us very effective and very efficient.
Another example of technology is proximal incidents, which paints a very nice context around the incident that is being reviewed by showing the prior incidents within a specific proximity and a time prior to the incident that is being reviewed. Peer review is another example. Peer review, in simple terms, is asking for a second opinion on a classification before classification is finalized. That is a more recent capability that we have added to our toolbox. Aside from that, the second group of factors have to do with the Incident Review Center training and their experience. They go through extensive training and a certification process to teach them how to perform the auditory review of the incident or how to use the waveform or sensor participation in determining whether an incident is or is not a gunshot.
If you talk to the Incident Review Center reviewers, they will share with you that there are other intangible factors, the type of experience that they have gathered over the years. For example, even gunshots sound differently in rural areas versus how they sound in dense metropolitan areas or close to a body of water. They will also probably share with you that patterns of activities are very unique to different regions. Those patterns are different in Oakland than they are in Pittsburgh or Las Vegas or New York City. That is the type of knowledge that we have gathered over the years, and we have contributed that to the vast knowledge base for our team, and that is exactly and precisely what sets us apart from competition. That is our competitive advantage.
I cannot think of any other company in our space that has the extensive team that we've got with 25 highly qualified reviewers with the average tenure of 39 months. That tenure has been lowered just a little bit by recently adding a few folks to provide service for our D.C. office. They perform with an extreme accuracy of better than 99%. I don't think that there is any other company who can claim to have the type of volume we've got to continuously build upon that knowledge base. Just in this past year, each of our reviewers reviewed more than 260,000 incidents. In 2020, we were all faced with a challenge that really tested our resilience, and that was COVID. Shortly after the pandemic broke out, we conducted a risk analysis and concluded that it was not possible for us.
We would not subject our team and ultimately by extension, the service that we provide to our customers to such risk. We had to go to a remote mode. It took us only a few days to provision the logistics, put our team through an orientation period, and go 100% remote for the entire company, but also for the IRC. That happened on March 12, 2020. The reason we were able to do that was, of course, the capabilities within the organization, but some of the investments, technology, and infrastructure investments that we had done prior to the pandemic. As an example, we have an internal tool we call the Dealer Agent, and its job is to distribute incidents to reviewers in a smart way. It looks at multiple factors.
For example, the overall size of the queue, how many incidents are being reviewed by each of the reviewers, some of the other factors. Also it dynamically prioritizes the incidents so that those incidents that are highly probable to be gunshots or have a higher probability are always prioritized ahead of the ones that are less likely to be gunshots. Peer review is another example. In general, communication is extremely important for the IRC. Being located in the same room, they used to communicate on a continuous basis. Everything from when they publish an incident, they provide awareness to their peers, or when they wanted to ask for a peer review or simply just communicating a five-minute break. We could not let that fail.
We took advantage of a channel-based messaging tool that we had already implemented, and we created dedicated channels to the IRC so that they could take advantage of that and communicate as if they were in the same location, even though they were in 25 different locations. Peer review, Paul briefly talked about it, but that is a capability that our technology team created for the team based on our unique requirements, and that has been extremely instrumental. Last but not least, as leaders, it was extremely important for us to be able to observe IRC's performance, and manage that performance in real time and as it was happening. Having access to a number of dashboards, some examples you are seeing on the screen, made that very possible and very easy for us.
For example, at any given time, we can see the flow of incoming incidents into the system, and we can see how those incidents are being distributed, the queue size for each of the reviewers, the velocity, the speed that the incidents are being reviewed, and if we have any bottlenecks or if we are experiencing any delays. If we have an unusual surge in incoming incident volume, we can quickly see which zones are most contributing to that, and we can perform a diagnosis. If the reason that we're having a surge is related to unusual firework activities, for example, or wind or things similar to that, very quickly we could apply our suppression tool to normalize the situation and continue to perform our review process extremely effectively and efficiently. Here are the results.
This graph shows the year-over-year change in the volume of incidents being published, gunshots being published to our customers. As you see, since the beginning of the year to the end of the year, it continued to increase. Even though year-over-year, we had a 60% increase in the number of gunshots published to the customers, we maintained our accuracy, and we even had a slight improvement over 2019. We continued to perform remotely, and we very successfully managed our two most important peak times, July 4th and New Year's Eve, completely remotely and with extreme success. In fact, those were the most successful July 4th and New Year's Eves in the history of our company. We recognize and we are confident that this trend will continue. The incident volume will increase, plus as a company, we will grow and we will expand our footprint.
We are very well positioned to scale in a cost-effective manner and continue to maintain our performance. With that, let us transition to our customer support. Our customer support organization is structured around Tier one or basic support and Tier two or advanced support. Our Tier one support is performed by the IRC. Because IRC is live and it's operating around the clock, that has given us the advantage or the opportunity to provide 24/7 live support by a live human being to our customers, which is extremely unique. I can say that probably we are one of the only vendors who provide that type of service in public safety. About 83% of all support inquiries are resolved at Tier one, and we have a very impressive track record of a 99% first-day resolution for issues that are resolved by our Tier one support or IRC.
We continue to invest in tools and capabilities for IRC so that they can be even more efficient. The picture you see on the screen, that is a chat tool that we have very recently deployed. It replaced our older chat tool. In addition to having a more modern look and feel, it is tightly integrated with our applications. It can be launched directly from our applications, and it looks like an extension to our own system. In addition to that, it has other capabilities. For example, we can automate some of the responses for those situations that really do not require any human interaction, such as resetting passwords. It also based on the past history of some of the questions being asked by the same customer, same person.
It can even queue up for us some of the articles or other resources so that we can provide better service and more rapid service to our customers. Our Tier two handles about 17% of support inquiries, which are more sophisticated or more complex and require more time to get resolved. We have about 93% of resolution under seven days. One of the things is that sets our Tier two apart from the rest is that we have a designated model of TSE to customer. Each of our customers know their TSEs as a person with a backup. The advantage that has provided to us is that since they are always working with the same person, we have been building a very trusting, intimate relationship with our customers.
Also, it has provided the opportunity for the technical services engineers to build a comprehensive understanding, comprehensive view into their customers, know their geography, the universe of their issues, things that they are more sensitive about, and that overall has provided a better level of customer service to our customers. We also have implemented another change into our support structure. As we continue to perform and deliver reactive support, which is fundamental to any support organization, we are doing proactive support because our philosophy is that before an issue escalates and is sent to us by a customer, we should have a great view and a great understanding of our own performance, and we take actions before it reaches a critical point. Of course, that impacts customer satisfaction, it improves customer satisfaction, and it reduces cost of support for us.
The reason we can do that, Joe covered this, we have access to a number of dashboards. We can see unusual activities. If there is a drop in activities for certain zones, we can take a deeper look and take action. If there is any degradation of service in terms of infrastructure or a significant change to sensor availability, we recognize those, and as we work very closely with our IT team and operations team, we notify the customer so they are not surprised at all by any potential degradations. Our technical support engineers perform a continuous assessment for incident detection, classification, and location. If they see any trends or any troubling situations, they bring that up to our attention so that we can work together, and we can solve that problem.
On a monthly basis, we share a scorecard with our customers. That is in complete transparency, informing them of how many incidents we have detected in their coverage area, how many gunshots have been published, how many false negative, false positives, and missed incidents, because we believe that type of transparency actually helps us in building stronger relationships with our customers. Result of all of that has been that when we ask the customers a specific question as part of our NPS survey about their experience with customer service, about 92% of them indicated that they had extreme satisfaction, or they were very satisfied with our support. We recognize that providing good customer service is not a one-time achievement. We will continue to invest in some of the internal tools and improving our processes to maintain and improve our support to the customers.
On forensic front, we are going strong. As you guys may recall, back in 2019, we introduced Investigative Lead Summary or ILS, which is an automated report that can be generated on-demand and on a real-time basis from our applications. Those satisfy about 90% of investigative needs for the police department. In 2020, we made another improvement to ILS, that was the ability to generate ILS report from our notification API. That was extremely useful, very instrumental for those agencies that have implemented a tight integration between our services and their public safety applications. For example, for New York City. Combination of those has increased the utilization of ILS.
In this past year, we had more than 70,000 ILS reports being generated. It was so useful that it reduced the demand for detailed forensic reports, which, of course, we continue to produce for those cases that went to court or for all officer-involved shootings. DFRs are reports, generally detailed reports, that are created by human, by our forensic experts, and we continue to perform that function. Because of COVID, our presence in court and expert witness testimony, it did slow down. It was challenged. Even with that, we participated in 18 expert testimonies last year, and some of them were delivered over Zoom. We have received multiple messages and emails, unsolicited, talking to us about the effectiveness of our expert witness testimonies and our DFRs as corroborating evidence to the cases, and we are extremely proud of that.
The reason we have been able to maintain our position is that we are completely neutral, and our forensic services are completely based on science. I hope that this summary was useful to you. With that, I would like to invite you to watch a brief demo of our Connect product prepared by Wendy Ethridge, the Director of our Connect solutions.
Hello, my name is Wendy Ethridge, and I'm the Director of Analytic Solutions for ShotSpotter. I've been with ShotSpotter since January of 2019. Prior to coming to work for ShotSpotter, I served as a criminal intelligence analyst for the Colorado Springs Police Department, where I specialized in gangs and sex crimes for almost 10 years. I then transitioned up to the Denver Police Department, where I served as a Deputy Director and managed the department's crime analysis unit. Today, I'm going to show you ShotSpotter Connect. What problems are we trying to solve with ShotSpotter Connect? Policing, like other professions, continues to evolve, and so do society's expectations that police departments will find ways to more effectively deploy their resources and do it in a way that is transparent and equitable.
That requires sophisticated analytic tools, the careful selection of data, and the right set of practices that most police departments simply don't have access to it. ShotSpotter Connect is our risk forecasting and patrol management solution. It provides chiefs and command staff with automated, data-informed patrol operations that are designed to reduce crime and maximize the efficiency of the department while fostering community trust. Connect is built on the foundation of protecting civil liberties and provides directed patrols across the entire city to not over or under-police one area. Connect also provides transparency into officer whereabouts and encourages interactions that are focused on community engagement and not enforcement. Let's imagine you're a supervisor who needs a patrol plan for your officers who are coming on shift today.
When you log in to Connect, as you can see here, at your fingertips is a data-driven patrol plan with precise risk forecasts by beat and shift that are crime and time specific. These forecasts are designed to maximize crime deterrence and have been generated by highly sophisticated algorithms that analyze and model real-time data sources. As a supervisor, you review the directed patrols and make any adjustments before presenting the plan to your officers during roll call, just moments before you send them out on patrol. With ShotSpotter Connect, you are confident that you have equipped your officers with the most effective way to utilize their unencumbered time they have during their shift. Now, let's imagine that you're a patrol officer who just received your assignments for the evening.
Upon opening ShotSpotter Connect, you can easily visualize your focus area for your upcoming shift, like we see here, and glean any analyst or command-enhanced details. Each of these directed patrols is a 250 m by 250 m data-driven risk forecast designed to take you into the areas at the highest risk of crimes so you can prevent those crimes from occurring. You can see the patrols for this shift. The presented forecast may, at times, validate your officer knowledge and experience. Where in the past you've had to rely on educated guesses, you're now informed on where and when to focus your patrols with a much higher level of accuracy, increasing your confidence that the work that you're doing is making a difference. As you enter a directed patrol, you are presented with a set of tactics to select from.
These tactics are designed to prevent crime while fostering positive community engagement. I'm going to select a foot patrol. Upon selecting the tactic, your timer starts and begins to count backwards from 15 to 0. This is the recommended dosage, and it's based on well-known research called the Koper Curve, where we learned that if officers spend 10 to 15 minutes in an area this size, again, the patrols are 250 m by 250 m, 2x to 3 x throughout their shift, it's the equivalent of placing a marked cruiser there on a saturation patrol for the entire shift. We get in there, spend 10 to 15 minutes, create a presence that lingers and keeps the community safe long after we leave.
Once I document any activities that I've completed while I'm on my patrol, and you can see I'm going to document two community interactions, I can simply close out my patrol and end my session. A dosage meter indicates to the supervisor and the other officers on patrol that I have now spent time in this patrol during my shift. Now, let's turn our attention to the reporting module so we can see how the officer's activity is documented and presented to supervisors and command staff. There are currently four reports in ShotSpotter Connect. This first report is a shift report that shows supervisors all activities for either the current or previous shifts. You can see there are a number of metrics on this report that supervisors can use in real time to coach and direct officer activity.
The second report in ShotSpotter Connect is an officer report, and this is where we can look at officer activity and team activity over time. All of the reports have interactive charts, and all of the data can be exported for further analysis with other tools. The last two reports in ShotSpotter Connect focus on tactics and crime types. You can see here that this is the crime type report, and this allows officer, and analyst, and supervisors to see what crime types officers are focusing on over time. You can see again, all of these charts and data are interactive and exportable. Finally, the last report is going to be the same data, but reversed. Now we can look at what crime types we're focused on and what tactics that we're actually applying there.
Lastly, I would like to show you how supervisors and analysts can create an ad hoc patrol to draw officers into an area for proactive interactions with the agency. It's as simple as selecting the edit tool and drawing a directed patrol on the map, just like I'm getting ready to do. I click on the map, I get a drop-down arrow. I'm going to select a command-directed patrol and hit Save. We can see that this patrol that I just drew on the map has diagonal lines through it, and that is how officers will know that this patrol is command or analyst-directed and not a ShotSpotter risk forecast. That is ShotSpotter Connect in a nutshell. I'll be glad to answer any questions during our Q&A session later this afternoon. Thank you so much for your time. Have a wonderful day.
Thank you, everybody. Today, you've seen a lot of information about our technology and why we're so competitively strong. We continue to invest and work on our technology, our products, and all the uniqueness that strengthens our position. Hopefully, you enjoyed that Connect demo that we just showed. It's a very powerful tool. I would like to go over a couple of the financial summaries, though. If I focus a little bit on the left, many of you who've been following us for a while know some of these. In the top, we focus on annual subscription-based SaaS revenue. It's greater than 95%, basically, at this point. It will change a little bit now that we have LEEDS acquisition, and they add a little bit of professional services to that. We operate with very high margin and very low variable costs.
We also have significant leverage in every operating expense category that allows us to be profitable in just about everything we do. If you go to the second column, we have very efficient sales and marketing. This allows us to have very low customer acquisition costs. I do have another slide where we'll talk a little bit about what those ended for us in 2020. We also have very low customer attrition and churn, which allows us to keep our high revenue retention that we have. In fact, you may recall that last year at the very beginning of COVID, we thought we may have attrition that could have been as high as 5%. Ultimately, as you heard Gary mention earlier today, we ended up less than 1%. That's pretty supportive of our customers for us. Lastly, we also have very strong project and unit economics.
We break even in less than a year for every contract that we conduct. On the right, you can see there's something very unique about our company versus most other companies. We're GAAP profitable on only a $10 million quarter. That allows us to maintain our expenses and our operating controls. It also allows us to help build those customer relationships. Our goal is to keep those customer relationships very long, hopefully 10 to 15 years. That would be our goal. In the next slide, for those of you who have seen day one, you may have seen that Ralph presented a slide of our history. I thought it would be interesting to show it again for those of you who didn't see us on day one.
Also for those of you who have followed us since the IPO, you may recall that we said our goal was for us to grow a 30% CAGR revenue growth for the first five years. We're proud to say that we met that exactly. It went 30% even through 2020. It will also end up about 30% if we hit the mid-range of our 2021 expectations. Also, you can see that our gross margin has also continued to improve throughout the years. You do see a little bit of flattening in 2020 and 2021. Those are related to our customer success organization expansions in 2021, where we start doing some of our 3G sensor replacements. That's addressing gross margin just a little bit.
What that does not do is that does not change where we expect our gross margin percentages to get to, which we do believe that we can get to 70%. One thing you don't see on this slide is that our adjusted EBITDA continues to grow past the profitability, and adjusted EBITDA has continued to grow significantly in the last two years as well. That's the history. Let's talk a little bit about where we're going and what we are committing to in terms of expected revenue and profit growth. It is important to understand this is not guidance, this is our goals, but it also tells you that we believe over the next five years, we can do significant things that are positive, which allows us to double our revenue growth from a mid-range of $21 million-$59 million, doubling that to over $118 million.
That would be approximately 15% CAGR growth on average. Even more importantly than that, though, if you think about what we've been able to do in terms of profitability and especially adjusted EBITDA in the last couple of years, we expect that we can increase our adjusted EBITDA on a dollar basis by 4 x over the next five years. That's significant. I also think most of you understand SaaS companies, other ones, and what the rule of 40 is, which is really a metric that talks about revenue growth and profit growth. These numbers put us well above the rule of 40 as we continue to grow. In terms of the customer economics, we are very efficient in terms of the sales and marketing that we spend. In 2020, we only spent $0.51 to generate that new annualized revenue dollar of $1.
We also achieved a revenue retention rate over 100%. We're at 107%, a little lower than the years slightly below that, but still quite impressive. We expect to continue that as we continue to grow. Last time we did an analyst day, we put this slide out there. We thought it might be very interesting to show it again. If you think about our guidance, our guidance for 2021 is between $58 million-$60 million. A lot of us have a focus of sort of the middle of that, in $59 million. If you take a look at the four areas that are basically building up our actual revenue, and this is GAAP revenue, we start the year with ARR. Basically, it's a little over $46 million. We have LEEDS producing some revenue for us this year. We've already talked about $10 million.
In terms of Respond, which is typically our domestic Flex that we continue to sell, we're talking right now that's about $4 million. There might be some less if we continue to have some attrition. Hopefully, we can do as well as we did in 2020, keep the attrition very, very low. The other one is just everything else. Basically that's Connect, that's Security, that's international. You can see that's a bit of a low number right now. Things could continue maybe they're a little faster than we expect. If you add all of this up, it adds up at $60 million. It is something that we feel pretty strongly that we'll be able to achieve.
It's always nice to get to the last slide, but I think this is a really important slide because you know many companies right now are searching ways to be significant in ESG, environmental, social governance, and their significance. The best thing about ShotSpotter is we're already there. If you think about what we do, let's just start with environmental. From environmental, some of the things that we do, primarily through our labs division in South Africa, we cover an area of Kruger Park, which is basically putting our sensors out there, helping prevent poaching from white rhinos. In Southeast Asia, we are doing blast fishing, which also helps keep people from doing blast fishing, which is destroying the coral reefs. From environmental, it's something we feel very strongly about.
You will see more things in the future about what we're doing with labs and things like that. In terms of socially, I would say in terms of a company, what we do is incredibly important. Not only are we helping people save lives by addressing the gun violence that's going into their communities, but we're also helping police agencies transform the relationships that they have and build communities. Very strong in terms of social. Lastly, if you think about in terms of the DEI or governance, we're a company that we feel quite strongly about this. We're very strong in terms of diversity. It's important for us. It's something we focus about a lot, and the same thing with our corporate governance. That's really the summary for us. We are doing pretty good in time.
We're not quite at the two hours yet, we are going to go into some question and answer period, and we have also already received quite a few questions and answers. What we're going to do is I'm going to try to hit as many of those as we can, and then we'll be able to answer those with some of our folks. I'm going to go ahead and start. Let me see the first one. First one was related to our Houston deployment, and basically, Chief Acevedo was going to be very important and influential for us in Houston. He is now no longer the Chief of Police in Houston, he's moved to Florida. I'll turn it over. Gary, why don't you give a couple thoughts on that one. Gary, are you there? Make sure you're not on mute, Gary.
My apologies. Can you hear me now?
Yes. Thank you.
My apologies. The bad news is Chief Acevedo is leaving Houston. We were very sorry to hear that announcement. The good news is he's going to Miami, another ShotSpotter customer, very important customer for ShotSpotter. We believe he's going to play an instrumental role in helping us continue pushing that tipping point across the state of Florida. The other good news is when I first sat down with Chief Acevedo several years ago to start talking about bringing ShotSpotter to Houston, he turned to Assistant Chief Troy Finner. He assigned Troy to work with us. Troy has been a great supporter. He's been involved every step of the way. He believes in what we're doing. He was recently named as the new chief in Houston. We've got strong support at the top of that organization, even after Acevedo has departed.
Great. Thank you, Gary. Appreciate that. We're going to try to move through these as fast as we can. There was a question about what percentage of the 90 mi in the sales objectives for Respond were already signed or booked but not deployed during or turned on yet. Honestly, we can't really answer that completely. What we can do is say what we talked about on our last earnings call. At that point, we talked about the miles that were actually live versus the ones that were under contract. We said there were 779 mi live and 813 under contract. You can think about the ones that are not live but under contract. That's about 34 of those miles. Next question was about the actual pricings in terms of several of our products. Talks about Security and the average sales value, and same question in Connect.
Gary, I'll go ahead and turn that back over to you. Hopefully, you can provide some information there.
You bet. The question seemed to be about whether or not the average of $85,000 for Security and $100,000 for Connect are reasonable averages. The answer is yes, on both accounts. What I would say, however, is we have a tiered pricing model for Connect. The annual subscription for Connect can be well below $50,000. It can be well over several hundred thousand. The price for those Connect deployments, the ongoing annual recurring revenue, can vary considerably based upon the scope and scale of the agency. On the Security side, pricing model is really based upon the size of the footprint, and those footprints can vary significantly, and so the price can vary considerably as well.
Great. Thank you very much, Gary. The next question. There's a question regarding what Connect does and makes it different to other similar technologies. Wendy, I'll actually ask you, can you give a summary of Connect? Wendy, you might be on mute.
Let's key it over to Paul then.
Okay. Paul, do you understand the question? There's a question regarding Connect. What makes it different to other similar technologies? Can you just give a summary of Connect, Paul?
Sure. The question, I think, is specifically relating to PredPol. There are two areas that are in the depths of the product. There's the how we create the modeling, the crime risk assessment model, and also how we deal with allocation, which officers go where. On the data modeling, the key difference is the, as I opened the presentation, we model using data that underpins multiple crime theories. The machine learning is done purely on crime data. That's a significant difference. The other is, as I said, is on the allocation engine that Simen went through. PredPol doesn't have such a thing. Even with the product as it stands today, I think we're in a pretty strong position versus PredPol.
Alan, Wendy is connected in an airport. She was having some microphone issues, but she is online. If you want to turn to her, she may be able to just briefly add to Paul's response.
Sure. Wendy, over to you. Okay, maybe not. Let's go to the next question.
I'm not sure if anybody can hear me or not. Unlike any other tool, just like Paul just said. Can you hear me, Alan?
Yes.
Okay, good. Connect incorporates those multiple crime analysis theories we are modeling. They include the evidence-based risk terrain modeling, near repeat patterns, collective efficacy, and temporal cycles and events. Like Paul said, this is not only hard to attain without sophisticated modeling, but also presents the most reliable risk assessment that focuses on crime deterrence while mitigating any bias.
Great. Thank you very much. Really appreciate that. Okay. Let's see. We have a lot of questions. I'm trying to make sure we can get through most of them, at least. There was a question related to the CapEx, related to the sensor upgrade program that is required by the 3G service sunsetting by AT&T and Verizon. I'll go ahead and answer that. We expect that overall it's about a $5 million cost that will be adding to depreciation over five years. It'll start small, then basically increase over the next couple of years. The good news is that as we're replacing these sensors, that number by itself ultimately may be replaced by other sensors where they have just gone past their depreciation. The actual cost isn't going to be impacting us near as much.
In terms of timing, we're expecting to do AT&T, most of those by February 2022, Verizon by the end of 2022. Next question. Let's see. Okay. The question is related to in terms of M&A. What are we doing in terms of M&A? What are we thinking in terms of that? Ralph, maybe you want to put some comments there.
Yeah, sure. Thanks, Alan. My apologies for being frozen here. With respect to additional M&A, I don't think we would never say never, but I think our feeling is that our plate is pretty full. We want to focus on successfully integrating the LEEDS acquisition and then kind of executing on this Precision Policing solutions platform story that we've chatted with you all about. There's a lot of work to do there. We just want to digest what we've acquired already. We think we have a fairly broad product suite. It doesn't appear that we need much else. We just have to focus on execution.
Great. Thank you, Ralph. We had a question about in terms of what does the IRC require in terms of adding new customers, or when are we required to add actual new IRC people and what value? Nasim, you can just give a couple thoughts on that if you'd like.
Of course. There is no hard number that exactly based on how many miles or how many incidents we're going to add more people. Reason for that is the behaviors, as I said in the beginning of my presentation, is going to fluctuate over the course of different shifts, different days, and so forth. We perform a continuous analysis of the overall workload. In this past year, each of the reviewers reviewed more than 23,000 incidents, but that is not the limit. Based on that, I would say that it has to be higher than that number. At the same time, it's about the overall performance of the team and overall workload and volume over the course of different days and different shifts.
One of the models that we have followed is that when we think that we need specific staffing over just a few hours or a couple of days, we are able to take advantage of part-time reviewers to fill in those gaps and remain efficient.
Great. Thank you very much.
That was a great answer, Nasim. Maybe I can just add onto it because I think people are trying to understand the business model implications, and I think, in a very kind of brief answer, it's highly inelastic, I would say. We can grow miles, customers, alerts that we have to process, alerts that we are publishing. To the extent that that grows, we don't necessarily have to grow headcount and expense in the Incident Review Center. It's certainly nonlinear, highly inelastic.
Great. Thank you, Ralph. Gary, this next question's going to come to you. It's basically asking, I'm going to combine a couple different questions. In terms of the pipeline in 2020 versus the pipeline in 2021, what do you see the difference between the two, and do you see that there was a consistent demand patterns basically being created through 2020?
When you compare the sales pipeline today to what we had this time last year, while there's not significant shifts in the size of the pipe, this overall Respond pipeline, we've got a much healthier pipeline in terms of where we are in the sales cycle, the level of activity on some key opportunities, and some of that is demonstrated by the number of deals that are in negotiations today, significantly in excess of where we were this time last year.
Great. Thank you very much. There was a question about whether we're using third parties to help find grants, and whether that's something that we feel some of the new initiatives might be positive for us. Trying to think. Maybe Ralph, maybe that's for you.
Yeah. Again, Gary, jump in here as appropriate. I think we've always had, I would say, a highly distributed set of resources to help customers identify funding grant opportunities. I think what's different here is that we have a kind of dedicated one throat to choke resource to go get at some of these grant opportunities, and that's all they do. We're going to be very intentional about using that resource because there's so many opportunities that abound now with the recent federal ingestion of capital at municipalities, and they're actually talking about returning to earmarks, which is a very exciting prospect for us as a company. You can look at it as adding net additional capacity to helping our customers identify financial resources. I don't know, Gary, if you would add anything to that or differ.
The only thing I would add is we've had assistance from a consultant out of Washington, D.C. for a number of years who's been instrumental in helping guide agencies regarding various grant funding programs. With the staff I referenced earlier now having the ability to provide assistance in actually writing the grant application and submitting the grant, we're providing a lot of added value to these agencies and helping them find funding beyond what's available for them locally, and the feedback has been very, very positive.
Excellent. Thank you very much. We can probably fit at least one or two more questions in here. There was a question in general about our expected Investigate case management software. Who are we going to be trying to focus selling that to at first? Is it current customers, or is it new customers, Any, basically, timelines for that? I guess, Gary, if you have any additional thoughts on that, maybe you can provide something.
With respect to Investigate and building a pipeline, we're certainly going to start with some known agencies where we believe, based upon our knowledge, contacts in the agency, there is a clear need for that type of case management capability. I don't see us limiting our focus to our existing customers. As we take steps here to start building a pipeline here in 2021, we will definitely be looking within the ShotSpotter customer base and outside of the ShotSpotter customer base for prospects.
Excellent. Thank you. Last question we're going to do. Nasim, it's over to you. The question is, "Is the IRC back to operating in the office? And if not, what do we expect or when do we expect that occur, and how is that changing efficiency or performance, related to what we had in prior years?
Sure. As of this moment, we are still in complete remote work mode. We're not back to the office. By no means we intend for that to be permanent. We will go back to the office. We do recognize that it may be a little bit different from the previous model. In terms of some of the changes, yes, from the perspective of performance, we have maintained our performance. We are also very sensitive about corporate culture and other things. During this time, we have been hosting several team-building activities and so forth to maintain the morale, in addition to training and skills for the team, but the morale and the way they feel about their job.
As the executive team, we're working on the timing to make sure that it is safe for the team to go back to the office, but we will not do that even one day before it is the right time.
Great. Thank you very much. We do have a couple more minutes. There was a question, Paul, I'll turn it over to you. Would there be any value in having the 5G connectivity to our sensors or not?
Not really. The bandwidth requirements between an acoustic sensor and the cloud are actually pretty limited. 3G, we've been able to operate a very successful service based on 3G. As Joe was talking about earlier, we're migrating to 4G. Frankly, the move there is being pushed by the carriers rather than ourselves. 4G is fine. 5G is really unnecessary.
Okay, great. I think that's it, really, at this point. All the questions that we were able to answer, we've answered. Thank you very much. We really appreciate you taking the time to spend a couple hours with us before the start of the weekend coming up. Ralph, any additional thoughts?
Yeah, no, thanks, Alan. We certainly appreciate everyone participating in today's session, and we hope it helped you better understand our business, and frankly, why we're so confident in our future to help transform how policing is getting done in this country and throughout the world. We're thrilled, excited to share our story with you, and I think as Alan would probably suggest, if you all have any follow-on questions, feel free to reach out directly to Alan or myself. We'd be happy to answer your questions. Thank you very much for your interest, and have a great day. Continue to be safe.