Everyone, thanks for being here. My name is Billy Fitzsimmons. For those of you I haven't had the chance to meet, I cover application software here at Piper Sandler. Started at the start of 2026. I cover the hyperscalers apps vertical. We're really excited to have Andrew Casey with us, CFO of Amplitude. Thanks for being here.
Thank you, Billy. Appreciate it.
I think a good place to start is people in the room might be familiar with Amplitude, but a lot's changed over the last year. Can you both take us through quickly what Amplitude does, the problem it solves for customers, and then some of the recent changes?
Sure. I've been with Amplitude for two years, and maybe I'll start with the two things that were going on right when I arrived was this effort to, one, sell to more enterprise clients, and two, was a postulate that our founders have about all the applications that have kind of surrounded product analytics were somewhat of a bug. They really thought that they should all be integrated in one platform. Okay, so I guess back maybe to the first question you asked, what does Amplitude really do? You got to think of Amplitude as this observation observability layer with every piece of software that gets created. Our foundations were around product analytics, where our motto was, "We help people build better products." You're getting feedback on how people are interacting with a mobile application, a website, a kiosk.
Anything that had software and a way of digitally engaging with clients is what businesses were deploying, and they were using Amplitude to get that feedback. A lot of it came in the form of, how do I make this website better? How do I make the feature get demanded more? How do I make sure that a marketing promotion is being seen and being acted upon? How do I make sure that cart abandonment isn't a big issue? Those are types of issues that every one of our clients who are using classic product analytics was trying to determine. They may also be deploying applications like Session Replay to get qualitative feedback, or using their experimentation platform to understand if they're making changes to the product, what the impact would be.
They might have used the CDP or an activation software, then take cohorts of customer information and feedback and move those into a Klaviyo or a Braze. They may have used Pendo or WalkMe for a guide and a survey to kind of point people in the right directions on how to interact with their product. All of those, as I said, were created around product analytics. They're taking their feed from the insights that the product analytics engine was providing. We decided that we should drive a strategy around building those products and adding them into a core platform in which customers could easily migrate or move between those different application environments, not have to learn a new UI, not have to have data taxonomy issues, and just drive greater and greater workflow optimizations.
I'd say over the last two years, we've kind of executed on those visions of, one, selling more to enterprise clients, which are increasingly trying to engage digitally with their clients, and consolidating all these other applications into a single platform.
I think one of the interesting things in the last earnings call that we talked about was. Walk us through this. Over the last two years, it sounds like monetization has maybe lagged underlying data ingestion growth.
Oh, for sure.
On the last call, we talked about how those trends are starting to potentially reverse.
Yeah.
Can you walk us through the dynamics around that?
Sure. So when I arrived at Amplitude in Q2 2024, there was a backdrop of decelerating growth. There was a lot of discussions around how many contracts actually have entitlements which were much higher than their actual utilization rates. What Amplitude had been seeing is contraction and logo churn, especially with what I would consider overselling capacity, and some really poor sales practices. In fact, one of the stories I tell is one of the largest contracts we had, I think I told you this before, but just for the audience, one of the largest contracts we had, I realized that it was a 12-month contract, and upon renewal, the customer was entitled to a 33% reduction in the rate of data that they were ingesting in the platform at the same volume.
I look at that contract and say, "Well, you need to take into consideration the likelihood that the customer would renew, and you incorporate that into your ARR." Well, that didn't happen. We were looking at a customer who was renewing. They were using the product, but we were going to get over $1 million in contraction just because they had this renewal opportunity. That's an example of a really poor selling practice that you just wouldn't do.
Today, if that were going to happen, I would tell the rep, "Well, if that's the way the customer wants to do it to make sure they're not going to have an increase in data ingestion fees in the first year where they're ramping up, that may be okay, but I'm going to pay you as if that second year does get renewed." Those are the types of the behavioral changes, the, I'd say, maturation that we drove throughout the go-to-market team. Yes, we replaced a bunch of them. Yes, we made them enterprise sellers, gave them a selling process and criteria and pipeline analytics and just really moved Amplitude more and more into what I would consider a very mature enterprise selling process, along with executing against this consolidation play we saw out there.
I think all of that culminated into the Q2 earnings announcement where we said, look, remember when we were talking for so many quarters about the churn that was happening, what the reasons were behind it. We kept using, "Hey, this is inhibiting our growth." If you would have looked back at that time in Q2 2024 and said, what was the real percentage utilization customers actually using their data versus what the entitlement was, I would tell you that was around 60%. It just pretended that there's going to be more additional churn happening.
If you take that same view this last quarter, I'd say that the data utilization versus entitlement was at 85%, and that's a very healthy place to be because it incorporates not only customers who had just signed and they're ramping up, but also customers who are now getting to the point where they've either been upsized or expanded into that entitlement. Now we're looking at a lot more renewals where customers are coming in and they're having to increase the amount of data ingestion that they're contracting for because they've outstripped that entitlement.
I have a lot of questions on everyone's favorite topic, AI. Can you contextualize for us Amplitude's role in this AI era? There are several products you have to address.
Yep
AI. Let's start with Wave Agent Analytics and your custom agents and the opportunity that gives you going forward.
Can I do it in reverse order?
Of course.
Because I think it's more of the evolution of the product itself. I would tell you that, for a long time, Amplitude looked at AI as not very interesting. Then there was a period under which increasingly it became clear that AI was going to be a key enabler for us, and that more and more of software's going to be created using agentic capabilities. The exponential increase in software was going to create a much broader surface for us to deploy Amplitude for instrumentation observability. When I explain Amplitude to friends and people about what we do, I say, "You got to imagine that software has slowly but surely had many layers over a period of time." Amplitude has the opportunity to be that key observability instrumentation layer for every piece of software that gets created.
Because there's more software being created, that expands the relative TAM for us. Now, we started implementing AI with a global agent and with these custom-built agents that you're talking about. We also enabled our MCP server for a couple of reasons. First, we wanted to make Amplitude easier to use. For a long time, and this predates you, but a long time, people would say, "Amplitude's hard to use and it's hard to implement." We'd say, "Well, we agree, and we're working on it." We did a number of things to make it easier and easier, like a single line of code. We consolidated the SDKs, but it was still an interface that was hard to use unless you were a data scientist or an analyst who was very accustomed to working within an application environment that brought all these disparate data sets together.
By implementing a global agent, now, increasingly, our customers are interacting directly with that prompt and just saying in a natural language way, "I'd like to see what the actual conversion rates are with the new marketing program I've rolled out. Can you draw me a couple of charts?" Then once that's done, interface again in that way to augment or, "Can you give me feedback on the cohort of customers that would be most likely to demand this product based upon historical frameworks?" That interaction is one that was really focused on how do we get more and more of not just the real data scientists and people who are very interested in the Amplitude instrumentation, but rather the broader marketing analysts, business analysts, pricing analysts that could use the data and do it in a much easier way.
That broadening the adoption, broadening the use cases was why we started integrating the MCP server that brings in more data sources and the global agent. We figured if we charged directly for that would create a pricing barrier where adoption would be mitigated. We didn't want to do that. It was so important for us in this consolidation strategy to make it broader, more use cases. Okay? Now, as we implemented these modules, experimentation, Guides and Surveys, activation, you started including specialized agents that actually worked within those applications. Session Replay is probably the one that had the most immediate impact because Session Replay gives you qualitative visual feedback on how anybody is interacting with an application on a website. Okay? Engineers would actually used to have to go look at all these different recordings. It could be thousands of them.
So invariably, what they are doing is they are sampling. Well, agents are really good at accumulating lots of data and summarizing it for you. That is exactly what the Session Replay agent was doing, is it is giving engineers the ability to say, "I want to look at 10,000 session replays. Distill for me the most important aspects, and what does that tell you from? What is the insight you get from it?" Customers were, again, looking at these specialized agents as, "Wait a minute, this is enabling me to do my work more efficiently." By the way, having the agents deployed in a consolidated platform, not individual products, but in a consolidated way, as you start to stitch together workflows. Let us say now from a product analytics perspective, you get some insight on your website.
Hey, there is a lot of people who are interacting on this banner that you say is a promotion, but it is actually not driving conversions. Like, okay, that is the insight. Run the experiment. What is going on? Let me see the qualitative feedback. The agents are actually prompting these interactions. If you do not have paywalls and they are all integrated well, suddenly the customer starts getting an experience which is much more holistic about how you actually drive that engagement to ultimate conversion. Both those instances, the ease of use with the global agent and the specialized agents, provided this framework under where the power of all these things working together in a platform was really exemplified. Even if the customer was not entitled, because we dropped some of the paywalls and they could start experimentation or testing the activation product.
Invariably, they would run out of entitlement, and they would either have to contract with us or stop using. That then provided us greater and greater pipeline for customers who might be interested in expanding with us, even if it was not in the moment. Okay, so that is kind of the backdrop on how we started making really good progress using AI to drive our consolidation strategy.
Then I would say when you talk about products that are for fee, like AI Feedback, Agent Analytics, and Wave, which I will explain each one. AI Feedback, got to think about it as terms of, it is a way of getting customer sentiment and feedback beyond surveys. Think what Qualtrics does today. They do a bunch of surveys, they get community information. But what is the response rate? It is pretty low usually with surveys. It is not really product data, it is sentiment data.
You want to marry product data, sentiment data that you are getting potentially in surveys, that is Amplitude Guides and Surveys. But then how about just this holistically around Reddit feedback, LinkedIn feedback, what they are saying about the product, your customer support, all that is connected through AI Feedback, and so you get a more holistic view of what the customer is actually perceiving your product to be good or bad at. That influences what you would build next. Agent Analytics is, we kind of stumbled upon as we were building our own agents and realizing that there really was no application environment to measure how well your agents are doing what you intended them to do. Think about the implications here. Every enterprise is building or deploying some type of agent to enhance customer service, improve workflows, to drive increasing monetization efforts.
All of these things then are surfaces in which you may want to understand exactly what your agent's doing. In some cases, make sure they are actually doing exactly what you want to do, as opposed to compromising your environments or providing personal information externally. In fact, there is increasingly more and more healthcare companies that are looking exactly to us to help them not only with the modernization of their web interfaces with their clients, like Sutter Health, but they are also starting to look at how the agents and the chat and the customer service environments can be better instrumented and managed to provide the outcomes that you are actually looking for. If you say, "I want an appointment set up," great, agent may jump in. If that agent does not actually get you to the point of booking the appointment, then it failed.
If you do not understand the reasons why it would fail or be successful, because you are not actually understanding everything that is happening outside the agent, then the investment is wasted. Agent Analytics is this opportunity for us to really help instrument and manage workflows that are being increasingly directed by agents. Sometimes it is not agent to human, sometimes it is agent to agent. We just went GA with Agent Analytics. We are charging based on the number of sessions, so think of that in terms of you engage with a customer service agent, that would be a session. The real critical aspect of getting this right as we evolve and add more customers is, what is the right yield? What are you seeing in the session volume versus the yield that the customer is getting, whatever their outcome is.
I do not think that we are absolutely locked in on the price point and the strategy at the moment. I think that we are increasingly going to find that there are variations in sessions by use cases and by verticals, and then that may cause us to have a different framework as we start deploying the Agent Analytics in different ways. Wave is a really revolutionary product, and it comes from Spenser and Curtis, our founders, talking about that software really should be able to improve itself. One of our taglines is, "We help people build better products." The next step from that is why can't the product build better for themselves? Why can't it constantly update and understand what is going on in security patches or integration issues or personalization?
AI has really enabled that vision to come to fruition in a product, and you got to think about Wave as a product manager, a data scientist, and a developer all rolled up into one. The customers that are beta testing us today, they are using Wave in lots of different use cases. But one is using us in loan origination. Wave is already surfacing ways in which the web artifacts that customers engage with should look at not just the rates of the loans, but what the payment is for the customers should be paying.
I know that's intuitive, but sometimes you surface these major little things that cause loan conversion to be increased because the customer's now associated with, "Oh, this is the payment I'm going to make associated with this." Not only did it surface what the issue was, it built the code, deployed the code, tested the code, updated consistently. As there's more feedback from interactions, Wave is actually driving continuously improving products with those interactions for that use case. Where would this go? I think every business that wants to digitally engage with a client is using it in an agentic way, is going to have a workflow they want to observe and they want to drive autonomous engagement around and they want to drive improvements.
One of the interesting things we're seeing now from all these router optimization companies, like OpenRouter and Fireworks AI and others, they're looking for ways in which they can get the best, lowest cost model to meet the performance that companies are expecting. But what if you could actually do better? What if you could actually take token spend to outcome based upon the use case, optimizing for the router in between for the cost? Suddenly you've got a really, really strong feedback loop associated with the ROI you're making investments in AI. So that's some of the initial use cases of Wave. I would say, just in product development is another major one. You have Wave deployed in your code base and constantly making upgrades and testing based on changes that your development team's contributing to your repositories, to security fixes that are coming on, to bug fixes.
There's really a broadening TAM associated where Wave can be deployed. I think our initial focus will be in e-commerce and retail, because everybody's got a digital interaction where they're trying to figure out how they optimize conversions and limit or reduce cart abandonment. But Wave and Agent Analytics don't need core Amplitude. They can actually be deployed in data environments of our competitors. They can be deployed in different data sets that we haven't been exposed to in the past. So it's a very interesting opportunity and growth driver.
Exciting new things. I want to make sure I ask you about Statsig as well. This was technology Amplitude got from OpenAI. I think when it was announced, frankly, my read talking to investors, there was a lot of confusion around it at the time. I think people understand it a little better now. Can you just size the opportunity for us, what the technology is, what Statsig does relative to core Amplitude, and then the cross-sell opportunity here?
Sure. First, I think you have to understand a couple of things are going on in the market, too. First, every major enterprise is figuring out how they are going to manage their data strategy. A lot of them are consolidating with data warehouses, like a Snowflake or a Databricks. That is happening. The other thing that is happening, we talked about earlier, is all these new AI development tools, they have done nothing but create exponential increases in how fast that software can be developed. What has not been really addressed is how good that software is. Is it actually doing the things you intended to do? There is a bottleneck in what we call the QA and test and learn process. You deployed something, what was the impact? That is increasingly becoming a tool called experimentation in the product development process.
Statsig, on its own, had done a very good job about appealing to software developers and AI developers to go test and experiment their new code as they were putting it in production. A customer of ours, Granola, they do a great job of using core Amplitude, but they are constantly running experiments on their code and deploying it based upon what they are learning from the process. They develop, they experiment, and it is happening just in a flywheel consistently. The software development process for them, the whole product development lifecycle, is increasingly turning very rapidly, and that is allowing them to drive new innovations and fix bug problems. Statsig did a really good job of appealing to that use case, data warehouse native, and development teams, engineering teams.
Amplitude had an experimentation product, but it was mainly focused in cloud-based environments and for other use cases. As we have seen more and more this whole product development life cycle start to open up opportunities for us, we saw a key opportunity to, one, get the best technology in the marketplace associated with data warehouse experimentation, and two, appeal to that increasing group of developers, both internal and external, that are going to want to use and have a tool set for running the experiments at scale. Since we have brought Statsig on, I would tell you that I have had numerous conversations with both existing Statsig customers and with potential, and they are just reinforcing the thought process there. This is going to be an enormous opportunity for Amplitude to take that technology forward.
Land with new clients, land in G2Ks, sometimes not associated with even the core use case that they built, but rather appealing to if companies want to become AI native, one way they can do that is by changing their development cycle and incorporating experimentation at scale to do it.
I want to tie all these things to financials.
We've talked about a lot of new products, accelerating product velocity. You have a lot more to sell. How do we just think about the growth rate of the business going forward for Amplitude? I also want to get your perspective on margins, margin structure
Sure
and the ability to provide leverage over time.
So let me start there, because
Yep
Ever since I have become the CFO, I have been using this phrase called growth with leverage. If you look back at any of our transcripts, I talk about it. We are going to go drive growth, we are going to make investments around it, but that does not mean it is at all costs. We have to be profitable. That was the first year people were very cynical about it, whether we could do it or not. In 2025, we broke even, and the first couple of quarters in my guidance for 2026 would suggest we are continuing to do that. That has just been my mantra, and it will continue to be. We are going to continue to make investments, but we are going to grow with leverage.
Spenser has talked a lot ever since, frankly, our investor day in March 2025, that he believed that 20% was growth, like the minimum for this business. We are very early, it is still developing. There are lots of opportunities, but we really should be growing much faster than we were. When I joined, it was growing 6%. We have gone from 6% to 22%. Majority of that organic, some of it inorganic with the Statsig acquisition. The reality is, I am not really satisfied at where we are. I think we have got a tremendous amount of opportunity. We are still pretty much a direct sales model with relatively few reps. Those reps aren't as productive as they could be, and we really don't have a major distribution channel. I think some of these new products could afford us broadening of distribution channels.
I think that there are opportunities for us to continue to accelerate growth, and certainly there are areas for optimizations in our cost structure as we go forward and getting greater and greater leverage. I will talk about gross margins because you asked about it, but when I set the budgets for the company, we set a budget where we expect to have revenue growth on, and the fact that our RPO has been growing for six consistent quarters over 30% gives me greater and greater visibility on what revenue is going to be. Then we set the target for expenses at a lower percentage of revenue than was the prior year. That means that they are going to have less to make investments, and they have to figure out and prioritize. That is everything from R&D to our sales and marketing and G&A.
Sales and marketing, at first join, I think it was like 47% of revenue. Way high, very inefficient. It is down to 39%, but it really should be more like 32% at our scale. We still have some path to go there. On R&D, I will try to keep my best around 18%-20%. We were actually under-invested when I first joined, it was 16% of revenue. You might see that get higher in any one given quarter if there are acquisitions to do or investments to make, but I think that is a good range for us. G&A was 17% when I joined. It is down to 12%, and I think there is still room for us to drive greater and greater efficiencies. We are going to get leverage on the opex side. I know we are almost out of time.
Yeah.
Gross margins were impacted this last quarter for a couple of things. First, we took on Statsig. Statsig's hosting environment, we had to slam into place in order to ensure that customers did not have service disruption, and it really wasn't optimized. We didn't have a long-term structured contract with Google. We weren't even sure we're going to keep Google. There's a lot of work to do there, but I have a lot of confidence we're going to go work in Statsig environments to be more consistent with Amplitude. The other thing is, which weren't necessarily bad things, but we had costs associated with inference and data ingestion that were outpacing our ability to actually drive conversions. The inference costs were somewhat our own strategy, broaden adoption, broaden usage, but those are starting to show up in new contracts and customers expanding.
On the data ingestion side, we've had a pretty good history of once customers are above their entitlements and moving that towards a monetization. I would look at that as like a precursor to increasing ARR and revenue. We're 99% of our business is ARR, is subscription related. When you see those types of things happening, it's really about us seeing success associated with adoption.
Perfect. We'll wrap there. Thanks for joining, Andrew.
Sure.
We appreciate it.
My pleasure.