All right. Thank you, and welcome to Sprout Social's discussion of our system of record and action AI strategy at the company. My name is Alex Kurtz, Head of IR and Corp Dev here at Sprout Social. Our discussion will cover foundational architecture of our platform, the data engineering that fuels Sprout Social, an overview of our products and customer use cases, and the power of Trellis AI capabilities. With me are Sprout Social CTO, Alan Boyce, and Distinguished Engineer, Kevin Stanton. Today's call will contain forward-looking statements which are made pursuant to the safe harbor provisions of the Private Securities Litigation Reform Act of 1995. All statements other than statements of historical fact are forward-looking.
These include, among others, statements concerning our expected future product roadmap, including our AI strategy and the development of our Trellis AI capabilities, AI agents, and Agent Studio, the expected benefits of proprietary data assets and data enrichment capabilities, our ability to scale our platform infrastructure efficiently, our plans for product automation and integration capabilities, our competitive position and defensibility of our technology and data, our partnerships with social networks and other technology partners, our expectations regarding the use of AI in our products and internal development processes. These statements can be identified by such words as expect, anticipate, intend, plan, believe, seek, opportunity, or will. These statements reflect our views as of today only and should not be relied upon as representing our views at any subsequent date, and we do not undertake any duty to update these statements.
Forward-looking statements address matters that are subject to risks and uncertainties that could cause actual results to differ materially. For a discussion of these risks and other important factors that could affect our actual results, please refer to our annual report on Form 10-K for the year ended December 31ST, 2025, filed with the SEC on February 27, 2026, as well as our most recently filed 10-K and 10-Qs. With that, appreciate everyone hanging with me on that one. Let's start the call. Let me introduce the team. With me are Alan Boyce and our CTO has been with Sprout for 15 years. Has seen it all.
Kevin Stanton, one of Sprout's distinguished engineers and with Sprout just about the same time, about 14 years, and has been a really critical driver to a lot of our engineering efforts of the company. With that, I'm going to hand it over to Alan Boyce. Take us through the presentation. I will say, along the way, if you wanna put questions into the Q&A function, then we can start reviewing those, and then we get to the Q&A, we'll start asking some of your questions. With that, I'm gonna hand it over to Alan Boyce.
Thanks, Alex. Appreciate it. Hello, everybody. My name is Alan. Been at Sprout 15 years, like Alex said. It's crazy to think of, but Kevin and I both together have 29 years of experience in the social space. Believe it or not, it's pretty crazy. Today we wanted to offer a deeper dive into what it takes to run our platform at scale and why we believe that we're well positioned to take advantage of AI. We both strongly believe Sprout has built a differentiated platform that allows for us to act as both a system of record and action for social, and it's really driven by three core technical architectural areas. The first is the platform at scale or how we manage a billion-plus social artifacts through our platform.
The second is our deeply integrated workflows between our product and partners. Then third is our Trellis, our AI platform, and how we expect it to unlock future value for customers. We're gonna dive into each three of these today. To deep dive on our platform, I'll pass it along to you, Kevin.
Yeah. Thank you, Alan. Hi, everyone. I'm Kevin Stanton. I'd first like to start off by giving an overview of our platform today. We spent the last 15 years building the system of record for how the world's biggest brands manage social. Every public and private message, every workflow and decision flows through our platform, and we handle a ton of jobs for our customers. The AI revolution's changed what's possible, but agents don't run on models. They run on context, and our customers have built theirs within our platform. Even at this high level diagram, there's a lot of complexity to cover, so I'll kind of walk through right to left. On the right-hand side, our ingestion services handle over a billion messages and metrics per day.
It requires an insane amount of algorithmic orchestration just to make sure that we can stay in sync in real time with the APIs that are constantly shifting around. As we move towards the middle, we structure and standardize the data and enrich it for our customers. Our efficient ML models, powered by GPUs, categorize every message in real time and without huge infrastructure costs. At this point, the messages are in the system of social record that powers hundreds of workflows, and we store more than four PB of data. Just a single enterprise customer today has almost 1 billion messages. In spite of this scale and our 30,000 customers, we can deliver search and analytics over the entire corpus of data in seconds.
Along the top, we integrate with dozens of partner APIs, like Alan mentioned, that unlock the workflows for customers where they need it, like routing cases over to Salesforce. On the bottom, our platform services provide critical capabilities for customers with thousands of employees to execute workflows without stepping on each other's toes and doing it securely. In the past nine months, we've added Model Context Protocol to our internal existing APIs to enable our AI agents to leverage that system of record. We do all this with incredible engineering discipline that I can't stress enough. Like last year, our hosting costs were only 2.7% of total revenue. I want to dig a little deeper into the system of record and why it's so valuable.
As you can see on the left, every action that's taken by a human or an automation or an agent, everything our customers configure, their tags, their rules, how they label data for their business, adds proprietary data to the system of record. Our customers are defining what's important and what's not, like what inbound messages are high priority for customer care. They define how they expect workflows to be done for their business, such as preventing posts that could damage the brand's reputation or create regulatory risk. Every action by a human user or an automation or an agent is important behavioral signal. It is like it's telling us what kind of messages they respond to the most, how they triage, what important metrics they track, or if they prefer to tag and group messages a certain way that their business understands and can report on.
Our system of record encodes how each of our customers do social for their business and their industry. It's a structured, queryable, and governed shared history that allows teams of humans and agents to do workflows at scale without tripping over one another. When an AI agent in Sprout needs to know how to handle something or be prevented from doing something risky, we believe the answer is already in the system. This makes AI agents possible at scale. The natural question that might get asked is, why can't someone just throw an LLM at raw social data and skip all of this? Anyone can write code a demo agent, but making agents reliable requires both a robust system of record and the engineering rigor to prove it. AI agents need evaluations just like software needs tests.
Evals are an entire engineering discipline around measuring and optimizing AI agents on specific tasks in specific domains. If you're shipping agents without these, it's like shipping unreliable code to production. An LLM is like someone who went to school but never worked a day in their life. Evals are how agents learn on the job and get better through experience. Our evals measure industry standard things like hallucination rate, but more importantly, they measure task-specific quality scores and completion rates, adherence to customer rules, and reliable use of our platform's tools. Every agent we build is continually tested against these benchmarks on real data. Evals are only as good as your ground truth, and ours comes from two different places. First is the system of record itself, 15 years of our real customer data that's enriched and labeled through years of real usage.
Second, it's human experts, our ML scientists, product experts, and the customers themselves telling us what good and bad agent behavior actually looks like. As you know, LLMs are non-deterministic. They've never seen our customer's data or specific prompts in its training data. This is why agents seem to work fairly well some of the time out of the box, but that's not good enough for businesses to rely on. This is why OpenAI and Anthropic emphasize the criticality of evals in spite of their own training and in-house evals. When AI agents fail, those failures feed back into our evals. The agents get better, the ground truth gets richer. We have deep insight because of the depth and breadth of our customers in our platform. A brand crisis, a regulatory incident, or a viral moment, these are rare for any one customer, but common across tens of thousands.
The flywheel we've built, we believe, enables us to continually improve the quality of our agents and our platform for our customers. To talk about how these integrations fit into the flywheel, I'll pass it back to you, Alan.
Thanks, Kevin. Yeah. Not only do we have integrations into all of the native networks, but Sprout sits at the center of all of our customers' social workflows. We're connected to their CRMs, their help desks, analytics tools. We're connected to their digital asset managers, e-commerce platforms, and then every major social network on top of that. That embeds us directly in the social transactions themselves. You can think of every integration as another job that Sprout does for our customers, and yet another switching cost for competitors trying to displace us. When you combine those integrations with AI, we believe we occupy a unique position in the governance, observability, and compliance layer for everything social. Social teams have hundreds or thousands of people who rely on us for auditability and oversight.
Computers, as powerful as they are, can make mistakes faster than any human could have hoped for. That's where our role as a guardian becomes critical. We see it as escalations, auditing what our agents, human and AI alike, are doing, and in verifying they're acting within bounds, confirming that it's safe to proceed. Because we sit at the center of our customer social workflows, we're uniquely positioned to be that layer of trust. As the cost of building with AI continues to fall, other things will start to emerge and stand out. Reliability, stability, security. We're building on long-standing relationships with the networks and foundation our customers have trusted for years. With that platform and those integrations as our base, we wanted to share more about how we see AI transforming work with our customers. For that, over to you, Kevin Stanton.
Thanks, Alan. Here's our AI vision, what we've built, what's in development, and what's coming next. About nine months ago, we set out to build our first agent, and in the process, built an entire agent platform layer on top of everything I showed earlier. Our listening agent delivers insights from massive sets of social data in seconds. When we presented at our breaking ground event in November, we received over 550 demo requests from customers, our biggest response ever. It's in the hands today with over 1,000 beta customers, with full release on track for this month. That platform foundation and agentic UI framework you see in the now column, it powers everything that comes next. In development, we have an insights agent that will span across care, publishing, and analytics.
An Agent Studio which will move us beyond chat to agents that run autonomously on a schedule, on a trigger, in the background. In a revamped automation platform that will let customers combine deterministic rules and AI in a single workflow. Looking ahead, a care agent, AI-created boards, and the ability to deliver insights where our customers work. We believe Trellis can unlock more value faster for our customers than anything we've built before. Our vision is that every workflow in Sprout can be assisted or automated with the right balance of AI and automation and human oversight for each customer. As AI reliability matures, so does the autonomy we can offer, and our platform is built to make that transition seamless. Here's a preview of our agents in action. On the left, our listening agent.
Say you're tracking brand sentiment, and the agent surfaces a chart showing a sharp spike in negativity overnight. You ask what's driving it, and the agent immediately breaks down the root causes. No digging through reports, no switching tools. It dynamically generates visuals and analysis in the moment. On the right, our insights agent. Imagine a customer's content engagement has been declining for weeks, and they need a lift. The agent analyzes what's been performing best, drafts new posts based on those patterns, and drops you directly into our publishing workflow. AI does the analysis and drafts the work, you review and approve before anything goes out. Our listening agent is context aware. It sees what the user sees and can access what the user can access while obeying the same permissions and guardrails. Our insights agent is being designed on the same platform.
This is Trellis Studio, currently in active development, where we move beyond chat to autonomous agents. Imagine waking up Monday morning to an email summarizing the competitive landscape over social from the weekend or flagging a crisis tied to your latest product release. Studio will give customers a library of pre-built tasks executed by our agents, each rigorously evaluated to run reliably on our system of record. Customers will configure them to run on a schedule or fire on a trigger, like an anomaly spike in listening data or inbound messages. Work gets done even when no one's at the screen. This is a platform play. Our vision is for every product team at Sprout to be able to contribute agents to Studio over time. As the library grows, the capabilities compound, and customers will be able to get increasing value without having to change how they work.
Here's what it'll look like to configure an agent in Studio. The customer won't have to write prompts. They'll select from the context they already know which social profiles are important for this task, which tags, which types of scheduled posts. It'll be programmable without prompt engineering. That's the point. Customers will be able to get the power of a configurable agent using the language of their business, not the language of AI. Behind the scenes, every task will have already been rigorously evaluated, so the agent will behave reliably regardless of how the customer configures it. Over time, we'll open up more control, letting customers define agent behavior to fit their specific needs, but we'll do it in stages, always grounded in the same evaluator that ensures quality. This here is a concept for the next evolution of our automation platform.
We're building an entirely revamped system where customers can combine deterministic automations and AI agent tasks in a single workflow. Here's an example. Your top competitor just launched a product that's blowing up on social. A competitor spike triggers an agent to analyze what people are saying and why it's resonating, and it pushes that summary straight to your competitive intelligence team Slack channel so they can get ahead of it. This will give customers a programmable workflow language without requiring technical expertise, but they won't have to build from scratch. We envision delivering pre-built workflows out of the box, and eventually, we believe agents themselves can help customers configure workflows. The integrations of our customers that have already been configured can become an output layer for the agents and automations.
They'll use deterministic automation where judgment's required, AI agents where intelligence is needed, and the customer decides where each fits. This is Sprout as the operating system for how brands run social. Agents will be able to run tasks, execute workflows, and deliver results, but they'll also be able to reshape the product itself. This might look like a standard dashboard, but it's not. We call these boards. You'll describe what a user needs, and the agent will compose the experience. The example here was built for a CEO who might want to engage with influencers when they interact with the brand. Instead of giving the CEO access to the full platform, you give them a board, the right data, AI-powered context, and the ability to take action all in one place.
The bigger idea here is that UIs no longer have to be hand-built for every single use case. Users will describe what they need, agents will build it, software that will shape itself to each user. Now back to Alan to talk about how we build at the speed of social.
Finally, in the same way we're building for our customers, AI is transforming how we work internally. Our most productive dev is now called Claude. uncertain allowing us to take the manual toil of maintaining this many integrations and do it in a fraction of the time. Traditional engineering teams are constrained by when devs are working. We've built tooling that allows bugs, errors, and package updates to all be event-driven in the cloud, meaning that when a dev wakes up, there's already work ready for him to review or her 24/7. We're doing it in the same way that we're building for our customers, which is with safety and governance alongside automation and velocity. To summarize, we have a massively complex platform with a track record of scale. We're ingesting more than 1 billion messages a day.
We sit at the center of our customers' workflows, embedding us in the transactions and positioning us as a key trust and compliance layer. We've encoded how social actually works on top of off-the-shelf and proprietary AI models. We believe we do it at some of the best margins in the industry with our hosting costs less than 3% of revenue. If you combine that all with our AI vision, we believe we're well positioned from a platform perspective to deliver great outcomes for our customers. With that, let's open it up to Q&A. Alex?
Yeah. Thanks, Alan and Kevin. Great job. DJ, I see your question. Before we hop to DJ's question, I do wanna
Maybe start with Kevin. If we go back to the platform architecture slide, if you could. I don't know who's running the slides right now. I think it's important, Kevin, you maybe talk through the engineering efforts to bring all these disparate data sources into the platform, what it took to kind of build all that, right? The ingestion mechanisms, and then to take all this at scale and then make it usable for customers. Maybe how that kind of built over time and why it's, you know, we think one of the best in class kind of defensible pieces of our platform.
Yeah. Yeah. Well, I mean, as Alan mentioned, we've both been around a long time, so it started small and scrappy. You know, the platform that we built in the ingestion layer, probably like six or seven years ago, we turned it into something that was a lot more robust and easy for anyone to add, you know, ingestion jobs so that we could keep pace with the change of the social networks. There's a few things that stand out. Like, one is the volume and variety of data. Like, there's no data standard in social. Every network's different, and we're ingesting the full fire hoses of X, Reddit, and Bluesky, and it requires a robust and scalable system just to keep up and enrich and store all that data.
The APIs themselves are constantly changing, whether it's like releasing new features or just changes in the data. Like, the scale of data is so massive that, like, you would think that every message that comes across the wire fits the correct specification, but that's not really true. We have to adapt to any, like, kind of weird curveballs that we get thrown by the networks. The other thing is social data is spiky and evergreen. Like, if you think about your bank statement, once March is closed, your bank statement is your bank statement, but a tweet or another message can go viral months later. You have to be able to intelligently go grab fresh data whenever stuff is happening on social. It's not as simple as just, like, receive it and you're done.
The last thing is just, you know, like, essentially our platform is keeping all the social data in sync with what's happening on the social network, so our customers can do their jobs, and making sure that that's arriving in real time requires being able to keep pace and make sure that we're not, like, tripping over changes in the network or any, like, anything that's happening at any given day.
That's helpful. Just a quick follow-up, Alan. Can you just talk about the partnerships that we have with the big networks and how that's evolved over time and how they've actually have become more selective in who they provide access to because of, you know, a lot of kind of bad use by other partners over the last few years? Just kind of like, these are legally bound contracts. It's built on trust over many years of use, kind of building on the stuff that Kevin was talking about.
Yeah, totally. It's been a long road with the networks, and we've grown with them. It's one of my favorite parts of Sprout is that social is kind of never solved. We have taken the long game with the networks. Early on, they, a lot of our competitors would kind of break agreements in order to have the best features in market. We never played that and it's paid off for us in a big way in that we have kind of preferred access and in ways that we both, with the networks and ourselves, both want a healthier social environment for customers and businesses to run on. For us, it's just about partnering and solving their problems, their customers' problems and our customers' problems, and they all overlap.
Right. I'm gonna start with a question from DJ Hynes at Canaccord Genuity, and he's asking about our AI monetization philosophy and how you see the business model evolving over time. I think it's important just to preface this question with a couple points before answering Alan. We are very early in our market adoption within our customers, right? We typically have several users to a small team working in marketing teams that have deployed our technology. This is not the situation where we have thousands of seats spread across large organizations and there's all this kind of inconsistent use right, Alan, that you would see maybe with some other SaaS vendors, right? Who've been around a long time and they're in incumbent markets.
I think it's also important to understand that, yes, monetization will likely change over time, and Alan will talk to that, but we're also not one of these vendors that has been, you know, pushing our user count like deep and wide into large groups of people. Like, these are very early marketing teams that manage social, and we feel like there's more opportunity with those teams. Alan, with that, maybe talk a little bit about the monetization philosophy that we think about with Trellis.
Yeah, totally. We think on it in a couple different frames. It's a really great question. I don't think anybody has kind of solved this, but AI is not one thing. We've got 40+ models running in production today that are doing everything from sentiment analysis to like deep image scanning. Each one of these comes with completely different kind of like cost and benefit analysis. We are also looking at like a chat experience that can break out of the app and bring more CMOs into the like a standalone Trellis agent that can sit alongside the CMO's desktop. I think we're gonna really have to end up with a pricing model that matches both the user and the efficacy of the thing.
Chat, while we have it today, we'll have like probably a per seat license for that to just cover basically the hosting cost side of it all the way to background agents that are doing more and more work for our customers, in which we'll be leaning into usage-based pricing or, kinda like outcome-based pricing and things like that. I wouldn't think of AI as just one thing in our platform. I think it solves many different things.
This is a question. Thanks for that, Alan. A question from Arjun Bhatia from William Blair. Is the Trellis Agentic Suite incorporated into the core, or will customers have to pay an additional fee to access it? With the upcoming capabilities. I know you kind of touched on it, but, like, how do we think about how we'll kind of be embedded in our core licensing? I know we're still kind of reviewing that right now.
I just love speaking with investors because we say it's embedded in the core. I go right to architecture and technology. I'm like, it's deeply embedded in the core, but we're actually talking about pricing and packaging here. No, it'll definitely be an additional cost. These are doing big lifts for our customers. We reserve the right to have some layer of AI that's just basically replacing how user interfaces get work done. For any of these background agents that are doing work when your teams are offline, we definitely want to monetize that because we're bringing huge amounts of value to a customer as they can't get now.
Yeah. Got it. Another question here. Just talking about how do larger customers leverage social data? We talk a lot about that as the sort of core of our strategy, right? Social intelligence. What are they doing day-to-day with it? Right? On this slide, we have a lot of different workflows that are happening, right, Kevin and Alan, and they're publishing, they're doing listening and analytics. Like, what is their main use case around social data, and how does it kind of work through our platform? Like, what are the top kind of value that they get out of that?
Yeah, I can totally take that one. On the data side, like, these big customers are trying to understand their competitors, their customers, their landscapes. They're using social for marketing, customer care, product development, lead generation, research, risk management. Fundamentally, they just need to be where their customers are. You will see us continually adding new networks as they expand and as they kinda differentiate so that customers can keep track. But you can think of social as this massive amount of real-time information about what anyone's feeling anywhere in the world, and we keep trying to find ways to use AI and insights and other visualizations to help people reason about that.
Okay. Have another question here. On the topic of preferred access with the networks, what type of social data access does this give you versus what a generic AI agent would be able to ingest? I think this is a really important question because ultimately, a lot of the LLMs also have social networks, right, Alan and Kevin? Social data is gonna be probably one of the most important training data like data pieces, data sets, right, that the world's gonna use going forward, right? Because that's all human interaction on social. We play this really important neutral party, trusted party. Maybe Alan and Kevin, like, an AI agent maybe trying to scrape a social network, what does that mean legally? Kind of what you've seen out in the market.
Maybe you could define that a little bit for us.
Yeah. I'd say it really starts with like preferred is a couple different layers of things, and it might be increased rate limits, it might be access to APIs before they're built. Several of the networks have come to us to ask our advice for how to build real-time streaming networks because we're one of the biggest consumers of all these networks on the planet. With most of the networks, we're actually working with them ahead of time to build these APIs to leverage for our customers. One of the key distinctions that allows us to remain neutral and kind of at the center of aggregator of all this social data is we're not doing any of the scraping on our behalf, and so we're not training LLMs on their data.
We're not kind of doing any of these like privacy type things that other vendors have tried to do in the past and gotten shut down for. We're only trying to answer our customers' questions about their customers. Most of the networks see us as this extension of their own ad platform customers. With that, we get usually like much more elevated access to PII because we just do the right thing with that data and don't share it when we don't need it, don't train when we shouldn't be training. That allows us to kind of keep the trust of the networks.
Yeah, that's an important point. Well, listen, maybe just one last question here, that I got. When an enterprise customer commits to a new, you know, Sprout multi-year contract, you know, especially the larger ones, and this is a wide-ranging question, right? 'Cause it could be across different products that we have. What are the top three kind of features or capabilities that make the biggest difference, right? When they're evaluating us, and versus what they can't do with the standalone kind of native products, in the networks?
That's a really good question. It is broad, but I will take my swing at it, Alex. I think I hear consistently across the largest customers, I hear three consistent things. The first is having all the data in one place. That's being able to connect the competitive landscape with the current social zeitgeist across your publishing, like, history, along with what your competitors are doing, is just all a desire. That leads into the second one, which is every CMO we talk to wants to connect their internal teams. Every company has creative folks, and they want them to be closer to the performance marketing team, and then they have a social team that's sitting somewhere else that doesn't know what either of those other teams are doing.
With our platform, we're connecting the creatives to the performance marketers to the earned, owned, and paid side of everything that lets them increase that flywheel of how they're doing marketing and care in a quicker, kind of more collaborative fashion. Then the third thing is just listening and Trellis together. I mentioned that social is this giant mountain of real-time customer information. Our listening product with Trellis lets people get access they don't need to be social experts, they don't have to be analysts. They can really, like, ask the normal everyday questions that they would have and get surprised by the answers that they're getting back.
Okay. Thanks, Alan. Thanks, Kevin. Well, wanna stay on time here. I think this was a great use of 30 minutes. Appreciate all the work that you guys put into this. Thanks for all the great questions. I believe we're gonna host the recording of this on the IR site so people can come back and reference it. If you have any questions, feel free to reach out to us and have a great rest of the week. Alan and Kevin, thanks for putting the time into this. Thank you very much.
Thank you.
Yeah. Thanks.