Hi. Good morning, everyone, and thanks for joining us for Oppenheimer's 29th Annual Technology Conference. We have with us today AvePoint CEO, TJ, CFO, Jim Caci, and Jamie Arestia from Investor Relations. Gentlemen, thank you so much for joining us today. Before we begin, for the audience, we do have a question bar for you that you can send questions, or separately, you can email me at param.singh@opco.com, and I can ask the question on your behalf. First, thanks again, TJ, Jim, for joining us. I wanted to thank you again for taking the time today.
Again, very high level to begin for some of the audience here, who may not be as familiar with AvePoint, if we could start with where you participate in the market, what are some of the niche areas where you are advantaged, and give a lay of the land for the audience here.
Good morning, Param. Thank you for having us. Yeah. AvePoint, we are an industry leader in cloud data management, governance, and security, where we play as really helping enterprises curate and govern their unstructured data, which includes all emails, chats, files, contracts, and what have you. We have been doing this for 20 + years, and we have done this very successfully globally in the large enterprise public sector, regulated industry, as well as now SMB. Of course, in the AI era, folks very quickly realize that to have good, high-quality AI deployments, you need to also have high-quality data that AI grounds on. We have extended since last few years into the space of AI governance, agentic asset governance, and cost control, and discovery.
This is now the really massive tailwind for us as we lean into this narrative of being the AI trust layer for our customers and partners. We are the layer between the model and the data. As everyone know, the commercial large language models, frontier AI companies, their models are trained on publicly available data. But for companies, for them to leverage AI to be productive for their business, they need to have AI refined on their own corporate decades worth of corporate industry data. That is context aware. That is also dynamic. In this framing, increasingly the ability to be able to manage both the foundational data layer as well as the AI agentic layer to work in cohesion with a right dataset, with a time sensitivity, with a right access control, lifecycle management, permission control, and also, very importantly now, also cost visibility.
It's what allow our partners and customers globally to have comfort and confidence in their AI deployment. The two top of mind issues we help companies, large and small address, one is AI risk and AI cost. That boils down to fundamentally what we do. We are the largest such cloud player, specializing in Microsoft Cloud ecosystem. But we have extended beyond that because customer is multi-cloud. We now also cover Google, cover Salesforce, and Atlassian, and a number of other ecosystems as well.
Fantastic. There's a lot of different topics I want to dive into, but maybe first, the most intriguing one for everybody in the audience and myself has been AI and how different vendors are participating in it. TJ, you touched upon that a little bit, that you're adding the governance and the compliance layer to the agent side, the AI side. Maybe we can delve into a little that, especially with your new product, AgentPulse.
Yes.
How are you specifically addressing that market? Because it's so nascent. How do you get confidence around what the agents are doing? What are they doing from a rules and regulated perspective? Or maybe some of the advantages that AvePoint brings as an organization that would allow people to use your product versus somebody else for monitoring and governing, let's say, agents here.
Right. That's a great question. For a decade plus, our expertise has been helping companies manage the risks and exposures of employees working on corporate data. What data the employees have access to internally, externally, how long those data should reside, should live for, and how do these data get ingested into the corporate system, and how do they get retired? Whether it's archived out, record managed out, tier storage out, or completely purged. In the age of AI agents are no different than employees in the sense that they have access to systems, process data, make decisions. Except now AI is operating at machine speed, so everything is accelerated. We already have this governance framework for quite a number of years in the cloud to manage large data estates. We're talking about petabytes of data, and for our customers around the world.
When Microsoft first introduced Power Platform, Power Apps, we did the same thing. We just elevated our delegated governance framework to include Power Platform and Power Apps. This delegated model means IT doesn't really understand what data and what business applications are doing for the business users. Our framing allows the business user to provide context richness to that. This delegated governance framework, compliance framework, then implement and actuate the corporate policy. In the age of agentic AI, it's the same thing. Basically, business user can provide that context and richness of what this agent have access to. If you vibe code an AI agent, by default, that agent inherits your permission structure. But that may not necessarily be the actual intended outcome.
We have the governance framework to allow users to actually provide that context so that our software can then start to track those agents, just like any other semi-autonomous or fully autonomous computer applications, their access rights, their data rights, and also, of course, their lifecycle, and then monitor their cost. We help enterprises go out and discover all the agents running in cloud, whether it's Microsoft or Google. Very shortly, we're extending onto the devices, so we can actually discover open cloud agents that's running there as well. Today, of course, we support Microsoft Copilot, Google Gemini, and then also Anthropic Claude, and ChatGPT, those type of AI services and agents, and then bring them under control. This is something we've been doing for a long time. It just really, in the agentic era, we then extend to the agents.
Again, no different than previously how we govern employee access, except now everything's just operating at machine speed.
That's great. And that's one point, TJ, I kind of wanted to delve into. I think people kind of forget that there is some inherent advantage of taking the skill set from dealing with human-based governance and compliance to machine governance and compliance. And of course, everything is at machine speed, which means it's exponentially higher.
That's right.
Maybe for my edification, too, if you can kind of talk about what are some of the inherent advantages in practical terms that you are leveraging from your existing expertise, one. Then two, when you're dealing with things at a machine speed, what are some of the newer items you had to introduce to be able to handle that, both from a monitoring perspective, but also from a response perspective? Because you also need to kind of manage much larger databases, have the correct, accurate information, and kind of stop things at machine speed as well. So maybe some clarity that would certainly be better for my edification here.
Yeah. That's a great question, Param. Firstly, the benefit that we have the right to own the space is that, again, we've been in this space for a long time, and our software are certified by some of the most rigorous security agencies, FedRAMP certification, U.S., ISMAP in Japan, which is like 3,000 security check certifications. So we are running government data centers, already in commercial data centers. So that fully vetted compliance and regulation and scaling and that capability around cloud security, cloud app, has already been vetted out for the last 15 + years for some of the most rigorous commercial and public sector enterprises, including the biggest banks on Wall Street. So that's a massive basically credit and trust that's already there in a highly scalable software.
Now, the challenges in the machine speed that you rightly point out is that the reaction in monitoring it have to also step up in machine speed. This is why actually last week at Black Hat, we released what we call Kinetic Classification, which is basically dynamic reclassification of content based on policy changes, role changes, data changes. So then we can actually, in real time, modify the context richness of objects, so that as AI act on those different type of things, we can actually evolve and change. My background's in machine learning and data mining, actually.
Right.
In our space, there is a concept called drift. I think most of your audience understand that now. AI, especially GenAI, is a non-deterministic technology, general purpose technology. But what it can do over time is it can have drift, just like in social media, this echo chamber that people talk about. The fact is 60% of unstructured data today are now generated by AI. If AI is acting on AI-generated data, if you do not actually go proactively curate it and manage it and govern it and retire it, that drift problem happens very quick. All of a sudden, you are detaching from AI. Basically, another way to think about it is it will hallucinate more. It detaches from reality very quickly. It is that continuous proactive coverage.
Lastly, I would say because we have a singular platform that does not only begin and end with governance and compliance, but we also have resiliency and data protection and data integration. What happens then is if you have bad actors, whether it is human actors or agentic actors that go and modify or destroy part of your infrastructure, we can actually recover also at machine speed. This is also something we released last week. Again, on top of everything we already have, what we call rapid intelligence recovery to prioritize the assets that you want to recover very, very quickly. This actually came into light very relevant. When we talk about Q1 earnings, we talk about the Middle East conflict. When that first occurred, Amazon data centers were hit in Middle East, regional data centers.
We were able to help our customers recover very quickly in other environments. Customers realize that you cannot just only rely on one hyperscaler infrastructure. The demand for resiliency, for recovery, this technology has been deployed for 20,000+ customers in term of ransomware detection and recovery, and now are using it for AI damage detection and recovery.
Right.
That is the resiliency side that we do. This is end-to-end platform that provides this risk management and cost management for our customers.
No, that is phenomenal, TJ. It is really good to see some of the progress being made here as an industry. Maybe we could quantify some of these things. It seems like a newer vertical for you, even though the technology is more expansive from what you are already doing. How would you quantify, say, from a cross-sell perspective or maybe an incremental ARR and TAM to the business here?
Yeah. We just finished our Q2, where we announced for the 13th straight quarter of double-digit net new ARR growth. We had the largest net new ARR quarter as well. Clearly we are showing acceleration in our business. 87% of our business is fully recurring. That is what the ARR captures. Then we have 12%-13% that is in services. Increasingly, that is becoming AI foundry type of services. What we found is that customers also recognize it is no surprise that Anthropic, OpenAI now starting consulting arms. Microsoft Consulting now pivoting towards frontier company. You actually need intelligent engineers who understand data, who understand integration and intricacies of systems, connectivity of systems, to land AI. We also have that piece going for us as well. We see strong momentum. We share specifically to AgentPulse as part of our Control Suite.
We GA'd in Q1, and by Q2, the number of Control Suite customers already doubled. The demand is so high that we actually made it into an independent SKU in July, just last month. A light version that allow customers to quickly discover all the shadow AI that is happening in their enterprise, that requires their attention to go remediate and bring under governance and control and monitor cost for. We also share that when, on average, when customer deploy AgentPulse product from us, our customers are actually already managing actively, on average, 5,000 AI agents. What is remarkable, it is the speed of growth of that number of agents. We see that it is doubling. Literally, we release in Q1, right? We see that number is doubling on a per customer basis every quarter, every three month. We already have customers that is monitoring and managing 100,000 agents.
Again, mind you, any AI agent can be a light agent, can be a workflow. It does not have to be a full-on virtual employee type agent. Those are still very rare and few, because those are very costly for companies to run as well. They could cost hundreds of thousands in token consumption on a yearly basis. But any agent that employs vibe code for workflow, that is just running out there. It is an agent. It is a computer process.
Got it. No, that's pretty helpful, TJ. We've taken this one lens of how you are addressing AI from a product standpoint, but how about using AI for your existing products? What are you using internally to maybe upscale your capabilities or maybe improve time to market on introducing new versions of the same product? Anything you can share from internal use of AI here?
Yeah, that's a great question. So, obviously we are dev shop. We're a very large engineer organization, so our developers are using GitHub Copilot. Behind the scenes, GitHub Copilot allow you to change different type of frontier models for usage of coding assistance. Of course, from a QA perspective, that's testing harness, automated testing harness are now leveraging AI. Internally for customer support, for sales support, for finance, we also have our own internal AI initiatives and projects. Even back in 2022, we released these very intelligent AI services that allow employees to go find resources very quickly and be able to engage our partners and customers much quickly and more intelligently. So there's that internal consumption of AI. But very importantly, it's the integration of that into our product. So, for example, in our product, we have this assistant called AVA, AvePoint Virtual Assistant.
They actually help our customers proactively identify risks and exposures they have and proactively highlight different aspect of our AvePoint Confidence Platform that can be brought to bear to help customers solve problems, to help them save license costs, and storage costs and all that. So there's already this internal, essentially, product level AI capability that actively recommend function and features. Lastly, because we collect a lot of signals for our customers across Microsoft Graph, so we have this product called tyGraph that collects all the interactions, sentiments, these essentially metadata across all the conversations, emails, and file access patterns, et cetera. That's a very nice intelligent database. We also now offer agentic experience with that layer because our customer come to us, some of the big four consulting company audit firms, they say, "Hey, not only IT and CISO want to use your product for intel.
Now business users want to query that for information, whether it's HR or finance folks, but they don't want to use your UI." Can they just use AI to directly interface? We also have made that possible. So it's the agentic experience. So we go from UX to AX experience as well. That allow us to expand significantly the type of persona we can help within the enterprise.
Oh, that is great, TJ. One of the products that I was looking at that I was personally excited about with some of these AI developments was Opus, was lifecycle management. Maybe that ties into what we are seeing today with identity and identity resilience and what is coming up in the space.
Yeah.
Maybe we can talk a little bit more about some of this identity and overall lifecycle management. Particularly because, with everything that is going on in the world and international threats and so on and so forth, it is the topic du jour, and that is where everyone is spending a lot of incremental dollars on. Maybe, TJ-
Yeah
you can talk about Opus a little bit more and what you are seeing there.
Yeah. That is a great question. Our background as a company, we came from the enterprise content management background. Think about the old days, you have Documentum, OpenText, FileNet, those type of products. Record management, document management, compliance, data classification, it is core part of what we do. What Opus does is actually, it started off as a record management product. It has an AI classification capability to help discover taxonomy. We also use this, by the way, in our monetization suite when we help customers classify, tag, and then migrate. Do data analytics and migrate data between systems. Migration is basically data movement. Data would never stop moving between systems on-prem to cloud to cloud. Think about you are moving data from maybe Microsoft stack to Amazon stack to Google stack, and from Salesforce to Dynamics 365 CRM to HubSpot. Different systems.
When you do migration, do movement, we actually help support a lot of that. This is why part of our recurring revenue is the migration side. We use the Opus capability to do taxonomy discovery, help them classify and tag, and move assets over. We always give the framing, when you move the house, you don't move everything as is and put it exactly where they were because the new house have completely new structure. You get rid of stuff, you doing your sale, you dump stuff, and then you remove and you reorganize. That's what a typical enterprise system move looks like. Opus allow customers to do what we call one-shots or even a guided capability to classify. Kinetic Classification is another layer on top to do the machine speed classification in real time.
What Opus also does, a very popular and strong feature, is storage optimization. It allow customers to save. Basically, if you move all your data in cloud, everything in cloud has a cost. There's compute cost, there's network cost, there's storage cost. Compute, of course, everybody know about token maxing. That was last year's thing.
Right.
That's just GPU compute. There's also a CPU compute. Now, do you want to leave all your stuff there to incur cost? You may not. Part of what Opus does is storage optimization. It will basically, based on its study and analysis of your data set, it will say, "Hey, this set of data hasn't been touched for a year." It's under this classification or this version. Because in the office world, in productivity world, whether it's Google or Microsoft, you can keep many version of the same document to track changes from an audit imperative perspective. But do you really need 100 version of the same document? Can you actually version out the previous 90 versions, the previous year versions and do duplication? All those storage optimization is also part of Opus capability that actually help our customers save cost and stay compliant.
Got it. That's great, TJ. Before I move on, I do have a question from the audience. I do want to remind everyone, you can definitely type in your questions or you can email me at param.singh@opco.com. For the audience question, TJ, Jim, how do you price across your different product portfolio at this time?
Predominantly, it is a subscription pricing. It is priced predominantly based on the number of employees, just like Microsoft pricing Office 365 and how Google prices Google Workspace. So it is employee count pricing, seat-based pricing. In some cases, we do consumption pricing. Even migration, for example, we have partners. We have a very robust partner ecosystem. Nearly 60% of revenue is done through partners. We have partners that are system integrators that would continually migrate one client after another, managed service providers. So they are basically subscribing to the software. But we also have partners that would do large projects in one and done. And those large projects, when the data size is of certain amount, we start flipping over to a capacity-based, consumption-based licensing. Same thing apply to backup. So resiliency side.
If you have very few employees but tons and tons of data, and we have many design shops are like that, then we also have consumption-based pricing because if we do not do that, then we will start to essentially operate at a loss. But predominantly, we are seat-based licensing mode. We are increasingly covering what we call IaaS and PaaS, so Infrastructure as a Service, Platform as a Service. These are basically compute cloud workloads. So you think about AWS,
Yeah
think about Microsoft Azure, think about Google GCP. So because a lot of AI actually run and applications run in compute cloud, not in productivity cloud, which is where emails and documents are hosted. So over there, the predominant market in framing of licensing is consumption-based. So we also have consumption there. We really follow the market. The hyperscalers are effectively the market makers. They will say, "Okay, for productivity, we do seat-based. For the compute side, for the IaaS and PaaS, we do consumption-based.
I think, Jim, you want to do a chime in?
Yeah. No, I think TJ did a very good job of summarizing that. We have always talked about really treating the value of our software as truly an ROI proposition for our customers. We still strongly believe that. So whether that ultimately winds up being years from now, whether it is consumption or ultimately, if some sort of seat-based licensing continues in one way, shape, or form, the end result really always comes down to ROI. Are you getting value from the dollars you are spending? We believe that our software provides that value, so ultimately, however it is being charged, it is about providing value to our customers.
Got it. I had a couple on pricing as well. So one, I am curious about how things change with agent bots. That has to be priced very differently. Obviously, you cannot do it on a per agent basis either, because it is going to be, I do not know how you would do it, right? Because there are millions and billions of agents getting turned on and they are ephemeral, and some are longer than others.
Yes
So forth. That is one part of it. Then separately, I had a question on what you are seeing as an industry, right? Seat-based pricing, in general, is under flux. The idea is under the AI landscape, headcount will get reduced over time. So is there a way to mitigate that? Because you might see more usage per user rather than an expansion in the number of seats. So how are you guys thinking about that second dynamic as well?
That is a great question. So first, I will address the second part. So I know last year everyone was worried about in the SaaS apocalypse phase everyone is worried about the massive seat reductions and also obviously AI replacing human phase of the hype curve. We think that has largely gone away, because also at the same time, the way hyperscalers think about it, Microsoft and Google specifically, they think about in a world where you have full-fledged virtual employees, right?
These are very sophisticated AI agents. They will have full email access, CRM privileges, and data access privileges. So they are actually licensed as a person. They are licensed as a non-human identity, but same as seat-based licensing. So there is that, right? In the very sophisticated side of it, AI, essentially a virtual employee, as far as a hyperscaler is concerned, it is a licensed entity, just like a carbon-based employee, right?
Right.
We right now do seat-based licensing because also the customers are really frustrated over the consumption-based licensing, especially when it comes to AI. Last year it was token maxing. This year, of course, the price has gone up. Nobody is doing that anymore, right? It is kind of also silly to approximate productivity and return on investment by counting on how many tokens you are actually consuming. It is really the business outcome that really matters.
This year, it is really much more intentionality around measuring business outcome of with deployment AI and also obviously managing AI risk, right? Because every other day we hear about frontier models letting their dinosaur out of the cages and go do some damage on the internet. That is also another side problem. I think right now for us, we are monitoring very closely. We stay with seat-based because customers also want that for predictability.
Nobody wants to all of a sudden budget a certain amount, and also with AI consumption, you vastly exceed that. What do you do then? Do you just shut off all AI access and then your employees are used to using AI and they cannot use it anymore? That is also not a good thing to have. Predominantly, we still stay as seat-based. We are monitoring this very closely. We continue to, because we have economy of scale, and we also behind the scenes are balancing different type of model usage cases. We are controlling our gross margins on the product side. We do monitor that. But for now, we like to predict the consistency and predictability for our customers.
I know we are getting close to time, but since we started late, maybe I can sneak two different questions in. Firstly, you have a lot of different products that are coming into the market. You are also expanding beyond your core Microsoft capabilities. You mentioned it earlier in the conversation. You mentioned Salesforce, Google, Atlassian, if I remember correctly.
Mm-hmm. That is right.
If you put it all together, how should we be thinking about your TAM and ARR expansion? Again, you mentioned marquee quarter of net new ARR in your 2Q results last week.
Yeah.
How should, as an audience, be thinking about net new ARR going into next year and that expansion? Within that, how do you kind of balance your investments in R&D to support these new initiatives and expansion capabilities versus operating margin expansion?
Yeah, that's a great call. We actually are looking at. We think the TAM is massive. Just to give you a sense, even Microsoft, just Microsoft 365 alone, it's roughly around 450 million user seats. The entire Microsoft Copilot activation is just 30 million seats.
Right.
It's been three years. They went from 20 million to 30 million a quarter, but it's still been three years. There's really a massive uplift. We think the market is huge in term of not only the number of user seats, but also the different type of cloud coverage, the different ecosystems, and different use cases. This is also why, in the most recent earnings, we announced that we are stepping up the investment to increase investment on go-to market, especially the channel side, which help us reach very good sales efficiency. When we went public, it was 44%, sales and marketing is 44% cost revenue. Now it's 31%. We think we can do better.
We need to bring on the partner and channel development and also brand awareness campaigns, marketing, because these things take time to ramp, to build up for acceleration into next year and beyond. There is also technology investment as well, because again, the token consumption cost is going up. At the same time, we are getting more efficiency. I think the TAM is getting much bigger. We want to actually present to the market at our next investor day the new reframing of the TAM, because we think that TAM has now completely. There is a very dynamic change to our TAM. We are very excited about the expanding TAM.
No, I think that is great, TJ. This is a question, my last one, and this is what I ask everybody. One is what you are most excited about and one, what keeps you up at night? I think that most of the call kind of already highlighted all the excitement that you have.
Yeah.
Maybe we could focus on what is the number one thing or a couple of things you are most concerned about or keep you up at night here?
I think what keeps all CEOs up at night is security. Honestly speaking, because now, cyber, it is operating at machine speed, right?
Right.
We have established a very good reputation in term of governing and protecting our customers' data, some of the most sensitive agencies in the world. But it is an everyday battle. Now it is, you say, agentic type of fighting machine with a machine type of case now. I think everyone is very worried about this. The frontier model companies marketing tactics does not really help either. There is no silver bullet, honestly, right? There is no one provider that can help you solve all your security concerns. It is truly a multilayered, multi-facet approach to security. You have to protect your endpoints, you have to protect your emails, and you have to educate your employees to be vigilant. Also the nature of GenAI with these prompt injection attacks, it is very hard to catch as well.
You can even do a prompt injection attack on commercially sanctioned tools if the attacker is creative enough. These are new security surfaces that is keeping most CISOs up at night, I think.
No, that is great. Look, for me personally, TJ, Jim, I am most excited about the next iteration of AvePoint's journey. Thank you for making the time today, TJ, Jim, Jamie, and thank you for the audience for attending our conference today.
Thank you for having us, Param. Thank you.