AI pricing doesn't behave like the software pricing banks are used to. Q2 CFO Jonathan Price joins CTO Adam Blue to talk through why token costs are simultaneously falling and driving total spend up, what Q2 has learned managing its own AI bill, and why the future of AI pricing may look less like "token-plus" and more like value-based pricing built around outcomes both sides can agree on.
"Ecstasy of Order: The Tetris Masters"
AI is getting more prevalent, and it feels like it's getting cheaper and more expensive at the same time. Join Jonathan Price, CFO at Q2 and I, as we talk about how pricing works whether you're consuming AI, you're building AI products, or you're trying to figure out how to integrate AI into your business. There are some real changes in the way that software delivery works when you start to take AI as a component, and we'll work through a lot of the bits and pieces of that and offer up hopefully some useful conclusions of how you can think about.
Adam Blue
Hey everyone, welcome to Cut to Context. I'm Adam Blue and I'm here with Jonathan Price. And he is here today to talk to us about AI pricing. Jonathan is CFO at Q2 and is a recovering investment banker, having worked for some of the biggest banks in the world, helping people understand software valuations and how to grow and drive companies. So thanks for being on today, Jonathan. I'm really excited for this topic.
Jonathan Price
Thanks, Adam. Happy to be here.
Adam Blue
So one thing I've learned in 25 years of working with banks is that when they consume software products, they like the pricing to be predictable, even, and uniform. And they're used to buying seats, they're used to buying modules, they're used to buying even blocks of transactions. AI pricing doesn't work like that. AI pricing is whimsical. So why don't you talk a little bit about what that's like, both at Q2 and then with Q2 customers, and how these new pricing models are the way that CFOs and bankers and other people are really thinking about the cost of doing business.
Jonathan Price
Yeah, I mean, I think you hit on it at the very beginning of that, Adam. It's very clear. I haven't been in it 25 years, but you know, sitting here after eight plus years, when I think about how banks are consuming the technology, the budget certainty is definitely top of priority. Banks, credit unions, financial services in general have consumed it that way. AI comes with a very different underlying cost structure that just forces all of us here at Q2, when we think about internal use cases and then when we think about pricing of our products that we're delivering to our financial institutions, it forces us to think differently just because the underlying cost driver is more variable in nature.
And so it is an interesting time and a journey that we're all going through. And I think you have to compartmentalize how you think of it when it comes to internal consumption, whether it's us here at Q2 or for our customers, how they think about internal consumption at the banks and credit unions versus how you price these products to the customer. And I could talk about both of those in different ways. But the key thing that ultimately companies are going to have to do when it comes to their AI costs at large is understand how much are we spending and what is the value that we're getting from that spend. And I think we've all been through what I'll call an experimentation phase. And I think that's OK. I don't think that should be conflated with failure or we're spending too much without clarity of the end results. Like this I think is part of the journey.
Adam Blue
Yeah, I think that's a great way to think of it. One of the things I find a little paradoxical is it feels like token costs in the main are kind of gently dropping. There's competition, there's the threat of open source and open weights models, there's choices. People are getting better at what we abusively and obnoxiously call tokenomics, which is not about marijuana consumption, but about token usage. And so how should we think about this seeming paradox, which is the total spend seems to be increasing at a time when the unit spend is actually getting less expensive? What do you think that signals?
Jonathan Price
I mean, you're right, both are happening concurrently. I would argue the latter, in terms of the underlying unit costs rapidly coming down, isn't happening that fast yet. I think people largely expect it to occur. I'm not sure we're there yet. It's an interesting problem because I can speak for us here at Q2, we have seen a line item come onto our P&L that didn't exist just two, three years ago and grow exponentially into a very meaningful number now that, by definition, we have a duty to manage. But it's sort of if we believe, and I think we've told this—Matt at our town hall across the whole company—that we are quote unquote all in on AI, we can't just talk the talk. We also have to walk the walk.
So our approach has been oriented around how do we enable our employees with the tools to transform the way they work, with the tools to deliver products to our customers in a new and different way, and with a value prop that's solving problems that we couldn't solve before the advent of AI. And so that comes with a responsibility, yes, to manage the cost and understand what the value is that we get from it. But it also comes with, especially in this first couple of years, I would say a responsibility and maybe better said is a willingness to spend the dollars to enable the workforce to transform how they're working and understand that there may not be perfect clarity in the early days on what the return is.
There's a lot of publicity and articles around AI spend versus the return dynamics going on right now. And the large narrative is that companies aren't seeing it yet. I would argue that, while you can't let it go totally out of whack and create problems for yourself, whether it's a public company or for your own sort of financial health and management. But I would argue that we have to go through this in order to figure out what is the right level of spend and delivery of new ways of working.
And that comes at a cost, and we will figure out what that means and we can talk about some of the things we're doing internally today and that I think banks and credit unions should think about when it comes to spend management that I honestly don't think we could have done six months ago.
Adam Blue
Yeah. Yeah. If you've got a certain amount of people and you want to retain for the most part that level of employment in your business, and you now have to absorb realistically a few million dollars a year in token spend, plus infrastructure, additional people with new skills, raises for people that are good at the work, whatever, right? It feels like there's kind of two choices you have to make. You can either try and fund a lot of that investment—let's just call it an investment—and you can try and fund it by reducing costs in other places, which frankly in a post-COVID area like is pretty thin. You know, people are not running around buying gold-plated coffee machines and having Maroon 5 come and play a show at lunch, the way it used to be, you know, in the go-go days of fintech.
And this is I think always the more interesting option to me, you've got to say this investment has to drive incremental revenue in the form of value for my customers. And I think that that loop, right, that idea of this investment has to create something we couldn't do before that somebody will give us money for because it has tremendous value for them, that is the piece that's so pivotal in all of this. And the velocity with which you have to figure out how to turn these new capabilities and new things you can do with AI into functioning products that create value downstream for your customers. That feels a little terrifying to me.
And I'm probably a little more of a riverboat gambler in the business than you are as the CFO. So talk to me about the sheer terror of what you just said, which was basically we're going to spend this money without clarity in terms of how it impacts the business, which I'm super proud of you, because that's the most un-CFO thing you've ever said in your life.
Jonathan Price
Well, it hasn't been comfortable, especially the, let's say mid '25 to earlier this year. But again, if we're going to talk the talk, we have to walk the walk. And empowering our employees and our customers to have these tools and to then change how we work and deliver products is the goal.
And so I think when I sit here today and think about sort of how we got here and what do we do going forward, the visibility that we have today into the spend at the team level, at the individual level, is so much greater than it was just a few months ago. And one of the things I see now that I think we assumed but didn't really know was this idea of like what models are being used and what is the actual underlying spend against what the current use cases are. There hasn't really been guardrails or really any sort of like rigid framework around what folks around the company or folks that are customers use to do their work.
So real example, when we looked at the first lens of what models are our employees using, because there is a big … big is not the right word. There's a massive difference between what I'll call a high-cost model on a per-unit basis runs us from a total spend perspective versus I'll call it a lower-middle-cost model. And without any guardrails around what model are you using for what use case, highest level, we saw at early points 80-20 of our employees using the highest-cost model to run day-to-day tasks versus a lower-cost model. And it's come down since then, but with no management, no guardrails, no real framework even for how we should manage that, either in terms of what everyone sees or in terms of how we throttle it on a team-by-team or individual basis. And I would argue that gives me confidence that there is a real opportunity with …
Because the key here is how do you manage spend without throttling innovation, without impairing our employees. And if it's our customers, the bank or credit union's employees, to be willing to experiment, be willing to fail in some of these initiatives, but not necessarily go, to your point, through this like fear factor of, oh my God, we're spending way too much money. Now I feel like we have at least some mechanisms, whether it's putting model management and really understanding what use cases require and demand that we have to be willing to spend for a high-cost model versus something that is more day-to-day tasks that just don't require it.
And, you know, maybe it comes a little faster and the thinking mode is more exciting and you want to try it. But at scale, how do we use more often the right model for the right task? That is a really material opportunity to bend the curve on the cost side without throttling innovation, without throttling experimentation.
And there are other examples of that. We're working a lot on training and enablement. How do you optimize for prompting? How do you think about caching and all these things that are going on in the world of AI that while I think our employees and the world is getting better and better at, we don't really restrict or manage historically with any enforcement. And so I think that's, I won't call it low-hanging fruit, but I think it's an opportunity to say as much as the costs have proliferated and it's scary, we have an opportunity now to do both, innovate and push the limits of what we can do with this AI, but also manage costs with more intelligence and rigidity—maybe is not the right word—but more control than we have for the last let's say 12 months.
Adam Blue
Yeah. Yeah, I think there's a story in there about observability and transparency being really key to having any shot at any kind of discipline. Because I think, you know, your average engineer sometimes I think can imagine themselves to be this sort of cackling mad scientist that's going to launch seven concurrent Claude sessions and let them code while they're, you know, scrolling through I/O 9 posts and eating a burrito. And it would be very difficult before AI to go convince seven people to go work all day long on your brand new interesting idea and stop doing all the other work they were doing, right?
But especially with coding, I think, and to some extent analysis and other things, but especially with coding, you could sit down and pull out a Fable model, attach it to an agent, and send it off on a task and let it run for four hours, and you've just spent half a day of quote unquote, you know, digital employee time. I think I just threw up a little in my own mouth with that comment, but it's so easy now to spend those tokens, right? It almost feels like the tools are designed in such a way that it's easy to spend tokens.
But dark patterns aside, the velocity with which you can now spend money as a line-level employee is fantastic. And if you're producing value, it's amazing. And if you're not producing value, it's just expensive. And so I think we've already grown through that token-maxing era. I think that's been over for a while, but I think even now today, if the observability and the transparency are there, the discipline becomes much easier because now you can have an objective and factual conversation, right?
And so I'm imagining that somebody's the CFO or the digital lead or even the CEO of a $3 billion, $4 billion institution. And they're in the room with their board who is simultaneously telling them, man, you better figure out tokenized deposit and stablecoin and how you're going to compete with Chime and what you're going to do with AI. And at the same time saying, you want to spend how much on what? And so maybe take us through like how to navigate that conversation because I think it's tricky.
Jonathan Price
Yeah, it is. And it sort of goes back to the underlying cost structure of these things and how do you imagine a world and operate in a world without that cost certainty that they've historically always had. And so one of the ways I think about it is as we try to determine how do we price these things and what value if I'm the institution, if I'm in that board meeting with a bank or credit union board, how do I articulate why the spend, this new line item, has value? What I think about is what is the meter that I care about when it comes to the value prop of that product that I may be buying, this AI spend that I may be delivering?
Because, you know, and I do not want to talk Q2's book here at all, but I think it's an interesting example to talk about fraud-related products. Because if I'm a bank or a credit union, there's probably not many more pain points that the institution is talking to their board about and managing on a day-to-day basis than mitigating fraud. When I can think about consuming a new product that may deliver a superior value prop around mitigating fraud, I have measures of that outcome, whether it's dollars of fraud that are leaving the bank, whether it's account takeover activity, there are probably a dashboard and there's probably a number of KPIs internally that the bank or credit union has that monitors the health from a fraud management standpoint.
If I'm the financial institution, what I want from my technology vendors is not really a token-plus like cost model where I'm just paying for the tokens I use plus your margin. If I'm the institution, I don't care about Q2's token usage. I don't care about our margins. I care about what value am I as the institution getting from these potential products.
And so I think where we have to find alignment as a technology vendor and as the financial institution customer is what meter can we all align on that says this is how this product would deliver value for me and this is how I'm willing to consume it. Even in a world where I've never consumed products that way historically, to our earlier discussion, I think there are ways and use cases where we can get alignment and where in that boardroom you can have a conversation with your board about what we think this product can stop objectively and quantitatively, in terms of—stop may not be the right word—but manage in terms of the fraud outcome. And here is what it would cost us.
I think the more we can become aligned on those concepts, which, by the way, isn't a new idea. Like value-based pricing in commerce is not like a new concept. But AI forces us to find alignment around these things because of this underlying cost structure. It is very difficult just for a technology vendor to imagine selling a product to a customer that has the ability for the customer to have unlimited use of it, but a fixed price that they are budgeting for. Because it is very possible that the vendor is going to get deeply upside down or make no margin from it, which by the way isn't what the customer wants of their strategic partner either.
So I think it's just this idea of can we find the right meter that aligns with the distinct value that the customer wants that that financial institution is in that boardroom discussing with what the product can deliver and what the technology vendor can offer. That to me is sort of … it's maybe utopic to think all products and all use cases can find the meter that works for the customer and works for the vendor. And it probably isn't possible in every use case. But I do think the more we can get to that outcome, the easier it's going to be for the banks and credit unions to absorb products in this new model.
Adam Blue
Yeah, I like the way you frame that. To some extent, it's a little bit of a classic economics problem, which is we've had a supply side shock that's made something a lot cheaper. And it doesn't feel equitable, doesn't feel Pareto to say that two companies or three companies in the world should capture all of the benefits of that supply side shock any more than it feels reasonable that the technology partner should capture all the benefits, or that the downstream consumer should capture all the benefits. There has to be an equilibration in the market that makes sense, where everybody participates, everybody works hard, everybody delivers value, and that everybody shares in, you know, what this kind of new surplus is. And I think what underpins that is this sense kind of like, you know, just equity, polite business equity. What's fair for us to keep, what's fair for us to pass on.
I think about another industry that I'm forced to interact with periodically throughout the year, and that's the airline industry. And, you know, we find ourselves in the world today in a situation where you know, decisions have been made and things have transpired that have driven up the cost of a barrel of crude oil, like a lot. And the markets are reacting to a variety of policy changes and announcements around the conflict in the Middle East in a way that drives the price up and down.
And I wouldn't want to run an airline right now because their biggest variable cost is aviation fuel. And aviation fuel gets made from barrels of crude oil, and you're not really sure how many barrels there are going to be and where we're going to run out of reserves. And everybody understands that whole thing. So airlines have always had to figure out how do we manage in this kind of uncertainty around this really key element?
So without getting too cute with it, I wonder if metaphorically, you know, the AI token has kind of become the aviation fuel of the information technology world in a sense. And as a technology provider and as a consumer of technology on the financial institution side, we're going to have to think pretty hard and be creative and disciplined around how we build those models to price those things, to distribute the surplus, and then sometimes to absorb the deficits that arise. Because token costs could go up and down, you know, depending on how many data centers get built or don't, how much the electrical grid can expand or not, what people's reaction to that is. Those could have long and far-reaching implications for the overall cost of employing AI as a technology. And so based on, you know, your experiences and working in software, what do you think are the things that a technology partner can do and should be doing to be the best possible partner to their customers, given that sort of variably random operating environment?
Jonathan Price
Yeah. Yeah. The way I would think about that one is almost take everything we've talked about around economics and cost off the table for one second, and as the technology partner, what are the use cases? What strategically do our customers want to do with AI? Because one of the things I think we did early on when we were experimenting—I know, Adam, you were involved in this in 2024—was just saying AI is here, but what do we do with it and what do our customers really want with it?
And I think one of our early examples was we were going to go develop a copilot. And that copilot was going to help our customers manage the lending process more efficiently, help them think through what are these more manual tasks that, in a world of AI, we can do more efficiently. We learned a lot from what I would call a failed pilot, which basically said that our customers certainly want to solve that problem, but they don't see enough value in the concept of that product to—and they don't know how to—size what to pay for it. That is the product that we should anchor on.
And that changed how we frame what we should be thinking about in terms of how to deliver value for our customers. From just being this like, let's deliver AI and let them figure out what's good for them, what's not, versus, no, what is practically useful for them today, which may be different a year from now, which may be different further down the road. And what we keep hearing from customers today is there are areas where they're using AI internally and where they're willing to use AI when facing off with their customers, be it a retail, small business, or commercial customer.
And again, we don't need to get into Q2's product strategy here or where we're going, but some obvious areas where financial institutions have started to use AI in the field, both internally and with their customers, are around areas like how do I make the bank more efficient? That is an area where AI makes sense. It is a use case that we can wrap our heads around. We can manage the risk around. And so as the technology vendor, it then becomes our responsibility not just to deliver AI for AI's sake or products that they can figure out what to do with AI, but how do we go deliver a bespoke solution to help bankers be more efficient that they want to consume? Because they should not be buying products from their technology vendor just because we come up with them. They should be buying products from their vendor because they see the value in it. And so the more we can align on their strategies and bring our roadmap in line with that, I think that's our duty as a strategic technology partner.
You can say the same with another area where they're looking for help with AI-driven value proposition upleveling the outcome on their behalf is the fraud management topic we talked about earlier. So, how do we deliver a different AI value prop, a product that's enhanced because of the power of AI that helps them do that better?
And those are the ways we're sort of framing up how we be a better partner to our customers in this moment, as opposed to the learning from 2024 was you can't just go throw AI at them because AI is here and we have a product idea. It was a good beta test pilot that sort of showed us that A, that product may not be the right place to start. But B, you need more definition, you need more clarity, and you need more strategic alignment, and that can guide where we go over time with our product roadmap.
Adam Blue
Yeah. Yeah. I think that product orientation is really clean, which is can you build a product that fulfills a job to do in the financial institution that's important? And can you do it with a level of efficiency that leaves surplus on the table for both sides to capture after you've delivered it? And if you can't, then you know, blow it out the airlock and keep moving. That's I think that's the lesson.
All right. Well, this has been fantastic, Jonathan. Really appreciate you being on today. We'll leave everybody with kind of our interesting element to open your mind. So there's a great movie called "Ecstasy of Order: The Tetris Masters." You've played Tetris, Jonathan, I think.
Jonathan Price
Too much in my life.
Adam Blue
Everyone's played Tetris. So yeah. These are people that don't do anything else but play Tetris. These are the finest Tetris players in the world. And I think it's a fantastic movie because it's interesting to see what it takes to be that good at Tetris. But it's also interesting to see what it does to a person, to their mindset, to their personality, when they try and impose order on an inherently unordered thing.
Because you know the magic of Tetris is … The tetramanos, as they're called, from the original game, can be arranged perfectly, but only for so long. And Tetris is such a beautiful visual metaphor for, let's see, everything in your human life. Like if you exert enough effort and enough focus and concentration, you can clear the pit of squares, but only for a moment, because there's always new pieces. You never know what the piece is going to be.
And the way these people approach the playing of this game and the inherent futility of it, it's really fascinating and sort of noble. And I think there's something there to take away from that around this conversation as well. So again, thanks Jonathan for being on today. I think this was really great.
Appreciate everybody watching. Please check out Cut to Context, subscribe to the YouTube channel, and catch us wherever your finest podcasts are bought and sold.
Jonathan Price
Thank you, Adam.