Mastering AI Pricing — Mayank Pant, Stripe

Description

Monetizing AI is hard. Rising GPU and inference costs are squeezing margins, and traditional SaaS pricing simply does not work for the unpredictable compute demands of new-age AI companies. With models constantly shifting across credits, tokens, and seats, a new challenge emerges: how do we charge for AI without stalling growth? This talk presents a framework for solving the dual problems of aligning charge metrics with true customer value and balancing predictable revenue with rapid adoption. Through real-world examples, we'll explore how to build guardrails that protect your margins and see how Stripe's world-class usage-based billing solution helps AI companies launch quickly and monetize with ultimate agility. Whether you're launching your first AI product or revamping your current model, you'll learn how to make your pricing strategy both profitable and adaptable.

Transcript

Hello, hi. I am Mayank. I am from Stripe and I'm the Billing Solution Architect working in Stripe. And today we want to talk about AI pricing. So we've got a billing engine and a lot of nowadays a lot of AI companies come to us talk about how they want to do billing, how we can help them with their AI pricing. So based on the research that our Stripe team has done over the last two years and based on our experience, we want to share some of the learnings with you guys. So as we all know, AI economy is growing at a record pace. They are growing 3x faster than traditional SaaS. And based on Stripe data, this is what we see, right? The top 100 AI companies took 20 months to get to 20 million ARR versus the top 100 SaaS companies, which took 65 months. So we are growing at a 3x faster rate than what we've seen before. And though this is great, this is exciting, but this is bringing new challenges. The speed of movement is so great that the companies are now going global faster. They are scaling up faster. And pricing AI is becoming incredibly challenging because of the speed that we see. So we know that earlier it was traditional SaaS pricing. We had 80-85% gross margins. They were not changing, right? But with AI, it's completely a different thing, right? The margins are low and the margins get impacted based on how many users are using your system. So neither pure subscription nor pure user-based model offer us a complete answer, right? And the simple reason is we are margin-risk from power users. 5 to 10% of your users can use 80% of your compute. External costs are unpredictable. That is your infrastructure pricing, right, which can really hurt your margins. So, what we mean by that is for a product company talking in terms of tokens, API calls might seem very normal. But people on the other side do not understand that technical pricing. For me, for example, if I go into Gamma, I don't really mind how many API calls am I using. I want to see how many slides or how many decks was I able to get out of the price that I'm paying, right? And pricing is not able to keep up with the product velocity. What might be a premium feature today, in six months, may be a standard feature across. So are you able to keep up with that pricing? So what we see is 33% of our AI-powered business cite unpredictable compute costs as their concern. 41% say that defining the value that is being delivered is a concern, right? And 84% of them talk about that we are rolling out product faster, but our pricing is not keeping up with it. So that is why what we say is iteration is a competitive advantage. The first price that you put in is a hypothesis. It is not a commitment because you are bringing in new features every week, every month. And with those new features and maybe the old premium features are becoming standard. So the pricing has to evolve with the feature that you put in, right? So frequent pricing change is a signal of growth, right? You build the infrastructure. You put your pricing in the infrastructure that allows you to iterate rapidly, right? So this chart is one of my favorites. It shows that hyper-growth companies, which are giving 100% plus year-on-year growth, they are changing their pricing three plus times in the last two years. They are not standing with their pricing. 49% is high growth. They have changed three plus. And the low-growth companies are only 22% who have changed it. And that means that those low-growth companies have a static product, right? And that is not where you want to be in today's age. So now where are these companies moving, right? If you look at it, hybrid pricing, which was only 6% in 2024, is now 41%. Outcome-based pricing is 5%, right? And where has it picked up from? Seed-based pricing, SaaS pricing, subscription pricing is all declining because the models are changing. So as I said, hybrid pricing, 7x increase. And now 56% of AI company leaders are using hybrid pricing. We talked to all the companies such as Intercom, Lovable, 11 Labs, OpenAI, Anthropic. They are all building on Stripe, billing on Stripe. And all of them are using hybrid pricing. And I've also talked with companies who were SaaS companies so far, which used to help customers build workflows. And they were in SaaS pricing. But now as soon as they bring LLM AI into their product, then they are also planning to move on to hybrid pricing. Because SaaS pricing then starts eroding their margins, right? So if this is where we are going, then how do we start thinking about pricing? How do we come to a right price? How do we define our price? And how then do we iterate it? So we've got a five-step framework for this. And step one of that framework is you define your value, right? So what type of value can my product provide? But not what your product is doing. But what the customer perceives your product to do. Again, I took an example of Gamma. For me as a customer, I need that product to give me my presentation, my decks. I don't care underneath how many API calls is it making, where is it going. For the customer, it is the quality of the deck and the relevancy of that deck. So 53% of hyper-growth companies have offered clear value-based pricing that the customer understands versus just 26% of low-growth peers. And the way we look at it is there are four broad frameworks. The first kind of these companies are the companies that are providing automation. As a company, I'm helping them to save time. And as a customer, what they are looking at is, okay, with the time saved, I'm saving on the cost. Second is augmentation. The number of people remain the same. But the quality that they come out with is much better. They can produce images better. Probably for a campaign provider, they can build better campaigns more quickly. So those kind of augmentations, the company looks at is, okay, I have the same number of people, but they are more efficient. They can deliver more value. Third is enhanced service. Maybe it gives you access to a proprietary software or some new data set that you wouldn't have access otherwise. For example, if you look at Stripe and our payment infrastructure, we can do fraud recognition much better because of the volume that flows through us. And finally, the fourth kind is which gives you improved results. For example, a company like Intercom who says, I will price you based on the number of tickets that I solve without human need. So that is impacting the direct bottom line. So those augmentations, the company looks at it, okay, I have the same number of people, but they are more efficient. They can deliver more value. Third is the enhanced service. Maybe it gives you access to proprietary software or some new data set that you wouldn't have access to otherwise. For example, if you look at Stripe and our payment infrastructure, we can do fraud recognition much better because of the volume that flows through us. And finally, the fourth one is which gives you improved results. For example, a company like Intercom who says, I will price you based on the number of tickets that I solve without human need. So that is impacting the direct bottom line. So once we realize how we are helping the customers, how our customers perceive it, that is when we know the value and we know the price that we can command in the market. Once we have understood this is the value that we are providing, then you define your charge metric. What is the billable unit that is best representing my value, right? And then you build features into a currency that customers understand. For example, is it consumption based? For an infrastructure company, consumption based might be the right one. How many API calls did I help you to make, right? And this aligns to the cost of the company that is providing their service. Second could be workflow based. How many images was I able to generate? How many documents was I able to summarize? And this aligns to the product. And the third is outcome based. As I said, how many business results did I generate? If I am helping the company to hire people, how many candidates that I have put forward, how many candidates were hired out of those candidates resulting in saving time? Or how many qualified leads did I generate? So this aligns to your customer ROI. So now that you have understood the value that you are providing and you have decided your charge metric against it, we will see how they change it. Right? If you move from consumption based to outcome based, it is definitely harder to implement. Consumption based is much easier to implement. But it is harder to allow to value. If some company tells me and says, I allowed you 1000 API calls. I do not know what those 1000 API calls mean. I want to see how many decks I generated as in my previous example. But if you go from outcome based to consumption based, it is easier to sell. I can definitely go to a customer and say, I will increase your outcome. I'll increase your candidate hired. But to attribute value is difficult. So that is where the balance has to be made. And data is required to satisfy and to make your case. One pro tip is to translate value with credit, bundle the features into credit. Say that I am giving you 100 credits for the month. And then beneath, under the hood of the credits, you can have your own models, how you get to that. But that helps the customer to understand, okay, 100 credits and it will translate to this particular ROI. Once you've done that, then that is where you pick your pricing model. Now, is it a subscription fee? The usefulness of subscription fee or SaaS fee is predictable revenue. It gives you a committed customer relationship. And then the second could be the usage fee, which scales with customer value, which protects margins. But the downside, as I said, is for the SaaS fee, your power users might burn your margins. And for the usage fee, the customer might be hesitant in experimenting with your product. Going full deep because they do not know what kind of an invoice will they expend on this. So, pure subscription, pure usage base was a thing of the past. Now, as we saw in the previous slides as well, everybody is moving into the hybrid model. So, what is the hybrid model? A hybrid model will have a base fee and it will have a scaling fee. The base fee basically helps you to establish a relationship with your customer. And the scaling fee or the usage fee on top of it will allow them to experiment as much as they want. And then to pay for the value that they derive out of your platform. So, you are not alienating any of the users, any category of the users. Once you've established your pricing, then you build in the guardrails. Because a wrong bill can erode a lot of customer trust. You might be doing great for three, four, five months. But if on the sixth month your bill goes wrong, bill goes very high, then those customers go. And then you work so hard to retain those customers, but then they leave you. So, what safety feature do you consider? You are giving them flexible pricing. Now you're building guardrails around it, right? The design principle is simple here. Build fair pricing, but do not surprise. So, what we advise people to do is put in usage caps. I paid $20 and then I've been giving 100 credits. I say that after 100 credits, build in a usage cap. You either pay more to go ahead or we will stop and you wait for the next month. So, it is the customer that has remained in control of their usage. You build an automated notification. You tell them when they've used 50%, 70%, 90% of their allocated limits. Because this is building trust with the customer. We are not trying to cheat them. We are just trying to inform them that this is what you've used. These are the things that you can go ahead with now. You can top up. You can do a manual top up. You can do an auto top up. Or you can pause and then start the next month when fresh credit gets allocated to you. And then you set rate limiting so that no wrong code is burning through the limits, right? So, this will protect you and your customers both. Once you've done these steps 1 to 4, then step 5 is you iterate, right? You keep iterating. 84% agree that fast pricing adaptation is a key competitive advantage. As I said, your first model is a hypothesis and you keep building on that pricing as your product keeps evolving. You prioritize speed. You do not wait for the right price point and wait for a year before you put it. You put a price point there which you think is the right price point and then you iterate. You talk to your customers. Those guys who churn, you talk to them and ask them why they are churning. Is it a product market fit issue? Then you work on your product. But is it high price? You keep iterating. 84% agree that fast pricing adaptation is a key competitive advantage. As I said, your first model is a hypothesis and you keep building on that pricing as your product keeps evolving. You prioritize speed. You do not wait for the right price point and wait for a year before you put it. You put a price point there which you think is the right price point and then you iterate. You talk to your customers. Those guys who churn, you talk to them and ask them why they are churning. Is it a product market fit issue? Then you work on your product. But is it high price? Then you work on your pricing, right? If they upgrade, you ask the same questions. You run A/B tests on pricing to find the optimum point, but you prioritize speed and you keep iterating. And then you continuously realign to your value. Now, realigning to the value means as you are rolling out new features, you've already given them hybrid pricing. You've already put in number of credits. Under the hood, you can keep changing what those credits mean. Do those 100 credits that you've given them mean 5 API calls of a certain category, 10 image generations of a certain category. Whatever. So under the hood, you can keep changing them. But you constantly realign the value. So this is how all the companies have been thinking about pricing. This is how we've been trying to help them. But what is most important in all this is what kind of infrastructure do you have for your billing, for your pricing that is helping you to do this? If every change costs you 3 months, 4 months and a lot of engineering effort, then it is not worth it. So that is why the infrastructure you choose determines how fast you can iterate. And from our side, the most flexible and complete billing solution in market is Stripe, and it is not us that is saying it. 78% of AI companies are building on Stripe, which is a testimony in itself. Most of the AI companies that you see here are building on Stripe. Anthropic, OpenAI, Lovable, Loom Labs, Intercom, they are all launching their billing and pricing. You might have associated Stripe with payments only. But in the last 2-3 years, we've invested a lot into AI billing. So we've got Stripe Billing, which allows you to go with subscription pricing, user pricing, hybrid pricing. And as these companies are scaling so quickly, within 10 to 15 months now, starting with retail, PLG motion, going to enterprises, we've got Metronome that allows you to build all the difficult and complicated contracts with the enterprises, [SPEAKER_01] having minimum commitments, pre-commitments, overage prices. [SPEAKER_01] So we've got that, and we've got the whole platform that allows you to do payments, tax, invoicing, revenue recognition on this AI pricing. [SPEAKER_01] So this is where we are right now, helping all our AI companies in the market. So I'm happy for any questions that you have. Yes, please. I think it's super interesting. With the advice around changing your pricing model fairly often, one of the risks is that it can cause churn with customers, and then they might turn because of the instability and it's less predictable to them what their contract is going to be. Have you got any advice around how to handle that? Yeah, so what people are doing and what we also enable them to do is they sell credits, right? But what do those credits mean? Like you brought in, let's say, today in January, you had one feature. That was your premium feature. It was not replicated anywhere in the market. And you had assigned five credits for that feature under the hood, right? In six months, that feature becomes a standard feature. The pricing had dropped. And in the meanwhile, you brought in new features because you are also competing in the market, right? [SPEAKER_02] So for the customer, they just see 100 credits. [SPEAKER_02] Under the hood, you are changing these calls or permutations. [SPEAKER_02] Like what feature means how many limits. So it will remain transparent for the customer. It remains fair for the customer. But based on your product features, you can keep changing pricing. [SPEAKER_01] Plus we have features where it allows you to grandfather pricing, like you bring in a new version. [SPEAKER_01] If I have been using it, I still keep getting it on the same price. But the new users have to pay more. Yes, please. At what ACV do you start to get better rates and more of an enterprise management with Stripe? Like how much payment volume do I have to do before I can start to get better discounts? So again, more on the sales side. But it depends on the volume, like your payment volume and your billing volume. We have a platform. We allow you to use as much of the platform as you want or as little of it as you want, right? And then depending on the volume that is coming, we bring it all together. Let's say you are using payment and billing together, but not using tax. That is fine. You bring your payment and billing and then our sales people start getting into it. The sticker price, of course, is there for everybody to see. I'm not really sure on what is the threshold that starts bringing the price down because that's more on the sales side. But do come over to our booth. We've got some sales people there. [SPEAKER_01] They'll be able to give you a better answer to this. I've explained the mechanism, but the threshold they'll be able to answer. [SPEAKER_01] Yes, please. [SPEAKER_01] The stats on the presentation were really interesting. [SPEAKER_01] Thank you. [SPEAKER_01] One thing I'm wondering, how does, for example, the pricing model of the big AI labs relate to this with the rate of pricing? [SPEAKER_01] I feel like the AI labs often have low-floor plans which have a very constant pricing. [SPEAKER_01] And then the rollout of the features always happens first on the expensive plans and then they take on the pricing. I don't really see like they don't really use very good pricing anyway, right? No, they use it under the hood. So for them, for you, even if you go to Eleven Labs, you will see four kinds of plans there, right? Let me call it good, better, best and then enterprise, right? You just go for the best, right? [SPEAKER_03] They will keep adding features or removing features from one plan to the other. The pricing will try to remain constant with the pricing for you. [SPEAKER_03] But inside it, the features will keep moving and that is why they want to give it credits. [SPEAKER_01] I don't really see they don't really use a very good pricing anyway, right? No, no, they use it under the hood. So for them, for you, even if you go to 11 labs, you will see four plans there, right? Let me call it good, better, best and then enterprise, right? [SPEAKER_01] You will just go for the best, right? [SPEAKER_03] They will keep adding features or removing features from one plan to the other. The pricing will try to remain constant with the pricing for you. [SPEAKER_03] But inside it, the features will keep moving and that is why they want to give it, give you credits. [SPEAKER_03] Or we advise that you give credit to the customer so that all these features then start interacting with pricing. [SPEAKER_03] Like the customer facing prices, they stay constant like link to which plan? [SPEAKER_03] Yes, let me say not the pricing, the customer facing plan remains constant, right? [SPEAKER_03] Price might change again. [SPEAKER_03] But the features will keep changing because it has been abstracted by credits on top. Oh, okay. Right? That's yeah, you had a question, sorry. Yeah, so my question is, does your platform provide a way to record every transaction? Yes. Every transaction. Every transaction. That costs something in the user, we can track it. [SPEAKER_03] Yeah, so what we do is if the prompt, you make a prompt, the customer is ingesting some calls, you will send those calls to us. You will tell us we have all kinds of pricing in there, tiered pricing, dynamic pricing, dimension based pricing. You tell us what kind of pricing it goes to and then it comes in. [SPEAKER_01] We are able to rate it and price it for you. [SPEAKER_01] And then you can get the whole report on exactly why the invoice is what it is so we can give you all the detailed pricing. [SPEAKER_04] Thank you. [SPEAKER_01] All right. [SPEAKER_01] If you have any other questions, then please feel free to come to our booth, floor three. [SPEAKER_04] Thank you so much. [SPEAKER_01] Thank you. Thank you.