Great. All right. Wrapping it up near the end of the day on day two. Great conference. I'm Derrick Wood, Senior Analyst covering software at TD Cowen. Our guest today is Serge Tanjga from Appian, CFO of Appian. Serge, thanks for being here.
Absolutely. Thank you for having me.
Would love to just start, quick background on Appian and just to get people familiar a little bit about the platform, just maybe provide an example, a use case of the applications and the platform that you guys power.
Maybe let me take a step back for some of the new faces in the room. Appian is a process automation company. We've been automating complex mission-critical processes for over 25 years; we've been at this for a while. We're going to do over $800 million of revenue this year. We primarily play in four highly regulated verticals, those being financial services, insurance, life sciences, and the government. Those roughly make 80% of our ARR. We have a tremendous and very demanding customer base, eight out of the top 10 global banks, seven out of the top 10 insurance and pharma companies, all 15 branches of the U.S. government. Our market is large. IDC estimates business automation platform market to be $37 billion and still continuing to grow healthily. Let me give you a couple of examples to make that real.
One of the large global financial institutions using us to monitor fraud on their money transfers. Millions of transactions. Before Appian, it was done in six disparate processes, or sorry, six disparate systems with a number of worksheets on the side and a lot of human labor to track. It was error-prone and risk of missing fraud. They implemented a single Appian layer on top of those systems, and they reduced the time it takes. They used AI to automate the investigation. It took the time down to 38 seconds to actually investigate a case. Overall process decreased by 98%, three-quarters improvement in reduction in risk. That's one example. Since we're big in the government, I'll give you another one.
A branch of the U.S. military is using us to manage the ammunition lifecycle to make sure that they literally can have ammo in the field when it's needed. That's the definition of mission critical. They had a legacy system in place that was crumbling, frankly, to manage 1 million transactions a month. They implemented Appian not only for its performance, but also for security. It's an IL5 environment, so that's a very high level of security that's above FedRAMP High. They now have near real-time visibility into the state of their ammunition globally, which as you can imagine, is tremendously valuable. If you take those, and actually, we just had an investor day two weeks ago, where we really went out of our way to provide a number of examples to give people a sense of what Appian is used for.
I encourage you all to spend some time looking at it. There's at least 20 examples in the presentation. It really comes down to we replace homegrown systems, other automation tools that have failed, or a tremendous amount of spreadsheets and manual labor to take these highly customized, complex processes and automate them to make them significantly better. Our customers see great returns, which is why they keep growing with us.
Great backdrop. Maybe just shifting into gears on talking about AI-
Let's do it.
the puts and takes. Let's just get right into it. What I love about you guys, this whole idea of determinism versus probabilistic workflows and with the importance of determinism and how you guys fit. You guys have a pretty well-educated view on that part of the stack. I would love to hear about why that's so important when deploying these new LLMs into mission-critical applications, and maybe where Appian sits in that part of the stack. We can kind of contra that. There are a lot of concerns on why. You want to build new apps, vibe code, what do we need Appian for? The risks around AI that everyone seems to be concerned about. We can talk about that as well.
Let's absolutely do it. Let's break it into two pieces. A framework that we think is helpful is to think of AI as doing two things. One, AI as a worker inside a process, and the other one is AI as an author, meaning actually using AI to create an application. Let's break that down into pieces and talk about them each separately. First, AI as a worker. Not so much now, but if you go back six to 12 months ago, there was this belief that you can just let AI loose inside your enterprise, and it will get all your work done. That's not a view that we've ever taken because like any other worker you can introduce in a process, AI has strengths and weaknesses.
Obviously, the strengths are it is very powerful, it is able to reason, it is able to extract complex information and concepts, frankly, only comparable to what a human can do. At the same time, it's relatively low reliability, and also at times it can be very expensive to use. You got to think of its strengths as well as its weaknesses and then deploy it appropriately inside the process. The thing about Appian and the legacy of process automation that we bring, we've been adding workers to the process for the last 25 years. When we started automating processes, it was all humans. You introduce rules engines, then you introduce integrations, then you introduce bots, and now there's various flavors of AI. Ultimately, these are all tools in a toolkit to deploy at the right moments in the process.
How do we deploy AI in a process as a worker? We surround it with guardrails. In other words, we check its work. If we're not confident of the outcomes, we kick it out to the human. We usually use multiple LLMs to make sure that the action that we've asked AI to do is actually done correctly. It can be audited. You can go back and examine why it's done what it's done, and we also use AI to make itself better over time, so that the accuracy that it has, it keeps increasing.
That's what's resonating with our customers, is that we're not giving some sort of broad promise in terms of what AI can do, but we deploy it very deliberately and surgically when it's the best tool, not uniformly, but when it's the best tool to be part of a larger toolkit to deliver a portfolio, if you will, of tools at your disposal to automate your process the best possible. The reason why it resonates is because we're getting real-world results across a variety of use cases. Now let's switch to your second question, AI as an author. Obviously, vibe coding has been the rage for, what, the last year, and it's very exciting, particularly when you start playing with it.
We were actually just with one of the investors who was telling us that she was trying to build a vibe coded tool for internal use, and you kind of get off to a quick start, and it feels very powerful, but you run into two issues, one relatively quickly, and then one it takes a little bit of time. First is the accuracy. You need to make sure that the application doesn't just kind of get it right but really gets it right. Again, I cannot stress this enough. Maybe for a side tool in some company, particularly a small and medium-sized company, a low reliability application is good enough. It's not true for a bank, it's not true for an insurance company, it's not true for a government agency, and certainly not for the workloads that they're using Appian for.
You run into reliability issues when you're talking about vibe coding. The second thing that you're running into, and if you're talking to customers or partners, this we've been hearing increasingly over the last couple of months in particular, is it's one thing to build an application, but you've got to maintain it. You've got to make sure that you know how it works so you can troubleshoot it and fix it, because otherwise it's just this black box that you've given some amount of work to do, and if it fails to do that work, you don't know how to fix it. We've had one of our partners talk to us about how they've vibe coded a meaningful application just to see if they can do it.
Before they knew it, the team that was needed to actually maintain it ballooned to dozens of people, you ask yourself, "Well, why did I do this in the first place?" AI is an exceptionally powerful tool. It's really impressive out of the gate to solve a number of tasks, but as you think about it, employing it repeatedly at a core of a real enterprise, it needs the support, it needs the harness, it needs the guardrails, and that's what Appian provides.
Great. Where are we in the stage of customers looking at bringing AI into worker or processes? Is there still a lot of proof of concept? Are you seeing customers in deployment and production and having success? What's kind of the typical ROI that the customer will get out of using Appian?
It's a journey, as you would expect, and different people are in the different stages of it. What I would say is, and I'll throw some numbers to kind of help it out, but 70%-80% of our customers are using AI in some form or fashion, although a lot of it is still proof of concept.
Okay.
However, if you want to use AI in production with Appian, you need to upgrade to a tier of service that we call Advanced tier. Standard is what it sounds like, standard. If you want to use AI in production, you've got to upgrade to Advanced.
Okay.
For that out of the gate, you're paying us 25%-35% premium on your license, and that is just for the purposes of building this application and using it. We said in the last earnings call that 40% of our customers are paying us for our AI tiers.
Okay.
Meaning some amount of their ARR comes with the ability to deploy AI features. I would argue that's a very good number.
That's 40% of our customers have, they have voted with their wallet that they want to use Appian for, excuse me, to actually put AI to production. What we see then, what comes as sort of the next step is you build the first use case. To build that first production use case, you're usually going to do it with our professional services, and that's part of the reason why our professional services org has seen strong demand over the last 12 months, because customers want to make sure that, again, implementation of AI isn't nontrivial, and to get to that high 90s accuracy which we are able to achieve across use cases, it doesn't begin there. It requires tuning. It requires working with the out-of-the-box software to get it there.
Then you have the first use case, and what we're now starting to see is some of those use cases becoming quite big. The reason why that's important is, as you think about the path to monetization, our Advanced tier includes what I would describe as a moderate amount of AI usage that's kind of packaged in. Enough for, let's say, one medium-sized application. If you have a very successful use case, one that is hundreds of thousands of documents processed or really any other AI actions that are being taken, that's not going to be enough.
You've got to turn around and buy incremental packages, if you will, of AI consumption from us, and that's kind of like the next level of usage and maturity, and we're starting to see customers get there, either because they have a very large first use case, or they're onto their second or third use case, and now they need to commit incremental capital to our platform. It's a journey. The great thing for me is that every month we see more customers in that final stage, and even then, that doesn't stop them from building incremental use cases because what it really comes down to is value. Customers want to see ROI from their any investment, and certainly from their AI investment. In fact, I would argue that's where, generally speaking, the industry has struggled over the last couple of years. With us, it's straightforward.
We partner with you from the beginning. We help you build the proof of concepts. We help you build the value case. Here's how much money you're going to save. Here's how much more revenue. It's interestingly, it's frequently revenue because you're going to have greater efficiency in your company, and therefore, as a result, you'll be able to write more mortgage applications. You'll be able to go chase more customers to onboard to your platform. When we quantify that, meaning the benefit, the ROI is really just that benefit divided by the cost, and it's very compelling.
It's both revenue generating and cost efficiency. On the cost efficiency side, as you automate more, do you take people out of those operations and there's headcount reduction? Maybe you redeploy them in other areas.
What we're seeing customers do is redeployment. What we're seeing customers do is using AI efficiencies to be more ambitious somewhere else in their business.
That's what we hear from our partners as well. What AI is unlocking is a greater ambition to automate more as opposed to just save money on what you've already automated, and it comes in two flavors. One is processes that you couldn't justify the investment before AI, but with the introduction of AI, either because it's cheaper to get it established or greater savings once it's actually running, more projects are kind of crossing the hurdle bar. The other one, of course, although it's early days, is the holy grail of legacy modernization. AI is a way to modernize your true deep legacy that is frankly holding back every meaningful enterprise. That's the other place where we're seeing. It's early days of becoming more ambitious, and that's another way to kind of grow the pie, if you will.
Do you guys embed AI building, like coding tools inside the platform for the application building part, or is it more about the actual process and production?
It's both. When it comes to in-process, we kind of offer AI functionality in variety in different ways. You can insert AI to complete very specific actions inside of our platform, very surgical and very efficient. We also have a DocCenter, which is our AI-enabled document processing, which is a very popular use case. Of course, we have agentic. Again, it's important to deploy agentic only when it's truly a case that requires it, otherwise you're just wasting money.
We also have a tool called Composer that is an AI-enabled coding. Just recently at our user conference a month ago, we've announced our own MCP server, which will over time allow for you to use your own favorite coding tool to actually code inside of the Appian platform, which sort of be the best of both worlds, which is you get to pick your own UI, but at the same time, you have an application that benefits from the platform.
All the benefits that come from it.
In terms of back on the monetization side, you get a certain level of credits at the upgraded tier.
Then if you really go into large-scale processes, then you've got to buy extra tokens, credits, and that, I guess, have you seen what's the uplift beyond there? Maybe it's too early to tell, but anything you can share online. Is that recurring revenue? Is that consumption revenue? How does that show up?
on the revenue side?
I'll fill in a couple of extra steps. The first step is that first upgrade, where you take a portion of your ARR estate and you enable AI on it. Like I said, 40% of our customers have roughly done that with some portion of their ARR. That doesn't mean that 40% of our ARR is actually on that AI-enabled tier. In fact, at our investor day, we showed that we just got to $100 million ARR on our Advanced tier, which is our predominant way to sell AI, and that's on a base of just over $600 million of ARR that we've done last year. Obviously, that's like, it's an opportunity to continue selling AI functionality to those customers who have already chosen to buy it, in addition to obviously selling to the remaining 60% that haven't gotten there yet.
well, the third step is the consumption element, which is you will buy more AI usage. You can buy it going forward, which I think is going to be the preferred way for customers to buy it and the preferred way for us to sell it, because then it's committed revenue.
Yeah.
You will pay for it in arrears, which will cost you a higher price. We'll see. I'm sure some customers will choose one versus the other. It's early days, we're going to see how that plays out. Not in the call it next year or two, if you take a further step back, as we develop further AI capability, we'll play the tier game again. We have a Premium tier, which is another 25%-35% upside, where at some point in the future, once we kind of feel like we're past the early adoption phase, we're going to put incremental AI functionality there, we're going to do the monetization game all over again.
Is this going to be a gradual adoption curve? Is there something that could catalyze and create a little bit more inflection? I know it's still pretty early, but are you looking for anything that could Or do you have to go through sales cycles, and these are complex systems, and just go chip away, and that's the name of the game?
It's a great question, and so I'm obviously always hesitant towards the year, but I'll try to give you a little bit of color there. The good and the bad of something like us is that it's a long sales process, and because we're selling at the heart of some of the most demanding enterprises in the world, you would expect them to make us jump through all sorts of hoops, which is why it takes nine to 12 months on average, and sometimes, frankly, much longer to sell these deals, particularly the larger ones. The upside of that is that it's exceptionally sticky. Our gross retention rate is truly best in class, and that just means that once you win the business, you're deeply embedded inside the enterprise, connected to the rest of their data system. You're staying there, right?
That's the good and the bad. However, we are seeing the one area where we are investing incrementally and focusing in order to make sure we kind of catch as much of the early AI benefit as we can is around our document processing solution DocCenter. The reason for that is a couple of things. Number one is it's a horizontal use case. Every enterprise is dealing with documents, and by the way, those are usually not pretty little PDFs where everything is printed. We're talking prescriptions that are handwritten. We're talking things with coffee stains. We're talking things that were copied 20 years ago, and they're sideways. It's universal, and it's a hard problem to solve. Our solution is excellent for two reasons. Number one, accuracy is very high.
We are able to get people to 95%+ accuracy versus legacy solutions that are 60%-70%. Second of all, it's not a separate solution. You don't process documents for the purposes of cleaning up the data. You do it to use it somewhere downstream. What Gartner ranks us is number one as integrated intelligent document processing because we've put it again in the process. What we're seeing is, and we talked about this at the Investor Day, successes across whether it's the insurance vertical, the government vertical, the life sciences vertical. DocCenter has a large part to do with it, and we're getting better at productizing it. We're getting better at quickly selling it at implementation such that you quickly see that high accuracy.
I don't know if that qualifies for inflection. We'll have to see where it goes. What we like about it is that it will allow us to quickly demonstrate at scale meaningful AI value adoption to our customers that will then only turn around and look for their second and third use case over the next couple of years.
Okay. Shifting gears a little bit, just talking about the go-to-market.
Maybe the reason why, you guys, you mentioned that 80% comes from four verticals. The focus is highly regulated industries. I guess maybe start with why is that the sweet spot for Appian? Then on the go-to-market, you guys made some changes over the last 12, 18 months.
Just give us an update on what changes were made, what the forward focus is, and how execution has gone in recent quarters.
Let me start with the current sort of mix of the business. Where we've had success is where the flexibility, the versatility of the platform, and frankly, where some of the security and auditability features resonated the most because the platform is exceptionally powerful. I have to tell you, whenever we hire a new sales leader, I try to meet them after a few weeks, and that's the first thing they say, "This is a great product. We can go and sell it." There's nothing to say that we cannot be successful outside of those verticals. In fact, we have great customers in the energy space. We have great customers in manufacturing, retail.
I'm not suggesting that we're constricted to those four, but we've had success there precisely because that's where the strength of the platform most resonates and, which is your second part of your question, where we've been most successful when it comes to go-to-market execution. We have made changes in the go-to-market. If you've followed us for a while, you know this. We've had times when we've allowed our focus to dilute from the areas of highest value inside an enterprise, existing or new one, and selling by value versus focusing on volume. We made a change and focused up-market is how we refer to it, focused on large strategic deals. Our Chief Revenue Officer joined us, I want to say 18 months ago, roughly, and he actually presented at the Investor Day and walked through all the changes that he's made last year.
Let me summarize that a little bit. First of all, we did significant changes to the leadership team. We brought in people who have experience selling value and strategic deals sort of across the board. That was Number 1. Number 2 is we went and rebuilt the foundations of our sales team, whether that's forecasting, whether that is qualifying deals, discipline of pricing and packaging and discounting. We started building a much more repeatable and rigorous sales process when it comes to deal qualification, when it comes to evolving executives in the selling process. We've done all those changes last year, which were significant, while having the best year of new business growth at Appian in four years. Back then, it was a much smaller company.
While making significant changes, we were able to also execute through it, which again, I give a lot of credit to our sales team. Now as you look forward, we still see a tremendous amount of opportunity, and it comes in two flavors. The first one is continued improvements in productivity that will come from continuing to fine-tune the process, focusing on the top of the funnel, focusing on our partner ecosystem, a number of other levers we can pull. Then secondly, because we've improved our productivity, I said internally, and I say it externally as well, we've earned the right to grow our sales arm. We kind of put a pause on that for two years while we made the changes that we had to make. Again, our paybacks have improved sufficiently that I said, "Look, we have the opportunity.
We know we're not market constrained. We're not product quality constrained. We used to be sales execution constrained, not as much anymore. It's a unique opportunity in order to establish that foothold as the provider of AI at scale. We're growing our sales arm this year. As long as we can do those two things at the same time, meaning continue to expand our footprint while maintaining and improving our profitability, that's a very powerful way to kind of build compounding growth, which is what we aim to do. Almost halfway through this year, I'm very confident that we're off to a good start.
Yeah, speaking of using AI internally for operational efficiencies. You can grow head count, also grow margins. How do you think about unlocking more margin efficiency with use of AI or anything else?
You would hope that we would eat our own cooking, and we do. We're seeing benefits of AI across the board. I would say most acutely we're seeing, or most positively, I should say, we're seeing it in our R&D efforts. Credit to our R&D leaders. They're really rethinking software development lifecycle with the use of AI. We shared some of the early metrics on our investor day, but highly enthusiastic about, frankly, the return that we can get from our existing R&D investment. It's about accelerating innovation without actually accelerating investment, and we believe we can do that as we think about it on a multi-year time horizon. That's an exciting opportunity, both for the purposes of driving top line while at the same time expanding margins.
Second of all the stuff that we help customers do, we also do ourselves, where we're using our own system, anywhere from customer support to triage tickets and deal with them faster to some of the back office processes, whether it's procurement, whether it's hiring, in order to make sure that we're efficient and really to grow our business, which we think we can do substantially without growing our head count footprint or at least slowing down the growth in our head count footprint. We are doing it this year. We're doing solidly on the revenue side. We're very proud with our results in the first quarter, for example. We're forecasting another year of margin expansion, so we're going to do over 100 basis points of margin expansion, and that's after a couple of years where in 2023 and 2024 combined, we did almost 20 percentage points.
If you think about our sales and marketing leverage, our R&D leverage, as well as our G&A leverage, there's opportunity across the board. We can keep doing both.
Has the competitive landscape changed for you guys at all? Who do you guys see in the market?
The competitive landscape has not changed, and that's not just in my one year at Appian, but even as I look at our competitive reports and quarterly win rates going back a few years. Basically, we see Pega, which competes for the higher complexity workloads and we've competed with them as the principal competitor over time. The other cohort is your ServiceNow, Salesforce, Microsoft. As I look at our win rate, they've been stable and strong. The one area that I'm particularly excited about, although I do acknowledge that it's early days, when AI is specifically a requirement, which by the way, is far from always the case, speaking to your point that it's still early, but when AI is a requirement, our win rates are significantly higher.
That is, I think, because we credibly speak both about the benefits that you can get with AI, but how to do it reliably and safely and with ability to govern and audit the AI actions. That's frankly where I think we differentiate ourselves from some of our competitors.
Yeah. Makes sense. There's been a topic of token maxing recently, and I'd be curious if It seems like there's a very inefficient way of using LLMs, and it's breaking the budget for a lot of people, and there's the big surprise costs. I'm kind of waiting for, is this the time when enterprise software vendors with guardrails and actually the context layer can come in and say, "Look, we could do this a lot more efficiently than you doing it on your own with the frontier lab model." Has that come into conversations at all yet in terms of like, "Oh, yeah, this is what we can do to drive more efficiency of use of this LLM?
It's coming in two flavors in the conversations, and I will say that's picked up on the customer side over the last two or three months, let's say. The first is just general bristling around spending more money with the labs directly to drive whatever internal productivity improvements. Those numbers are growing rapidly. I, the customer, am not sure I'm getting a return on that investment. We're hearing some of that. That's kind of orthogonal to what we're hearing, but we're hearing more of it.
What we're seeing in customer conversations is customer will come to us with a poorly defined AI problem, and what I mean by that is they'll come to us and say, "Hey, I have an agentic use case." We sit down, our professional services team sits down with them, and we tell them, "You could use our agentic offering or anybody else's agentic offering, but you could use our agentic offering, and we would argue that that would not be a good decision because this problem isn't broad enough to require agentic. This problem is better solved with more surgical AI usage, which is going to be more accurate and more efficient when it comes to underlying infrastructure.
You will have better ROI, not only because you'll spend less money, because your accuracy will be higher. Like a silly example that I say is like, you shouldn't ask AI to do basic math because there's much cheaper way to do basic math, and at the same time, one in a million times AI will be wrong because it's a probabilistic technology. We're seeing customers wising up to both the benefits and the limitations of the tool, and then we, as a vendor, have a lot of credibility to turn around and say, "We'll let you use it in a way that is going to actually deliver best long-term value." Now, what's interesting is if you roll the clock forward, think about where we're going to be as an industry in 12-24 months.
Despite the costs going up right now and customers complaining about it, there's also a general belief that AI is subsidized right now, or rather that the current economics are not congruent with the amount of investment that is going by the companies that are making that investment. At some point, and some forward-looking customers are asking this question as well, it's like, "Well, what happens if indeed this is true and the subsidy ends?" The answer for us is, well, we can still provide an ROI for you because we are independent or agnostic of your model layer. You can work with us regardless of what the use is in the back end.
Again, we're optimizing your use case from the outset to use your AI as just another worker in the process where it's needed and where it's the best use of your time and money as opposed to just anywhere.
Great. Well, great discussion, Serge. Thank you. Thanks, everybody.
Appreciate it. Thank you, everybody.