"Up until a couple years ago, SaaS pricing used to be kind of boring, especially at the startup stage. Folks would undercharge, figure out who gets the most value for their product, build products for those companies, raise prices, rinse and repeat."
At 80% SaaS margins, that works phenomenally. But when you potentially have millions in hard inference costs (at enterprise scale), your monetization lag can't be months... you have to be ready to adapt pilots, pricing and billing in days or weeks.
Batting lead off at the Monetizing AI summit was Kyle Poyar, one of the most respected voices in pricing and monetization out there. He talked through the state of AI pricing, what's happening now, and where leaders think things are heading (backed by some great proprietary polling data).
Check out the full video (and stay tuned for breakout insights to come).
Why AI cost curves make flat fees dangerous… and how to avoid negative-margin customers
How the biggest players are changing pricing faster than ever — and what to copy vs. skip
The shift from “owning” to “hiring” AI products (hello, people/hiring budgets)
Why hybrid models (platform fee + credits) are winning — and when to layer outcomes on top
How outcome metrics, not just usage, unlock pricing power (if you can measure them)
Real examples: chargeback recovery as outcome pricing; consumer-friendly credit models
Fynn Glover: For AI products, undercharging has very real costs. You can end up with negative margins and unprofitable customers. Some of the biggest players are scrambling on pricing and changing it faster than ever. We used to watch what the largest companies did and follow along. Even they have not fully figured it out.
Buying models are shifting from owning to hiring AI products. That lets vendors tap into hiring or people budgets, not just technology budgets. We cannot afford unlimited plans with one price for all customers given the range of usage patterns and the costs behind them.
As a founder, pricing is one of the biggest mistakes I have made across companies. As an operator, I have worked at companies that were never serious about pricing. It was not a first class citizen in the product, and it was not treated as an operational discipline. Even top business schools do not teach pricing. You can get an MBA at Stanford and never learn it, which is wild because pricing is one of the most important growth levers.
That has been true for a long time, and it is happening alongside a lot of change. Many of you are building AI businesses. I look at what is changing for pricing through two lenses.
First, there is more software and AI than ever. Customers have choice. Years ago you did a demo, sold a few seats, and closed the deal. Today buyers expect to purchase how they want, be billed how they want, and use parts of your product, not the whole thing. You must be flexible with pricing and packaging. It is table stakes.
Second, AI creates a big opportunity but also changes how we price. There is volatility and unpredictability in inference costs. What does that mean for a buyer who faces zero predictability? How do you handle that as a vendor and a buyer?
I ran pricing at a company from $10M to $40M ARR. We left tens of millions on the table because we underused pricing as a growth lever.
Kicking us off, we have Kyle Poyar talking about the state of AI pricing. Kyle is calling in from Boston. Many of you read his work. Kyle, thanks for being here.
Kyle Poyar: Fynn asked me to give an update on the lay of the land for AI monetization. I have worked in pricing and monetization for more than 15 years. Until a couple years ago, SaaS pricing at the startup stage was pretty boring. Teams would undercharge, figure out who gets the most value, build for those customers, raise prices, and repeat. With 80 percent margins, that worked fine.
That model breaks with AI. If you undercharge, costs are real. You can get negative margins and highly unprofitable customers. I have seen a company where the top 5 percent of users drove about 75 percent of usage costs, but pricing was flat fee. Those users were only 5 percent of revenue and all were unprofitable.
AI products are also taking off faster. You cannot afford to get pricing wrong and fix it months later. People bet on LLMs getting 10x cheaper every year. There is some truth, but the best models are still expensive. Buyers want the best tech, not yesterday’s model. At the same time, token volume is exploding with longer context windows, more reasoning, and agents. You cannot punt monetization.
Big players are changing pricing faster than normal. Salesforce introduced per-conversation pricing for Agentforce, then added a flexible credit model a few months later. OpenAI historically leaned on seats for ChatGPT and enterprise features, then moved to pooled credits for some products. Another headline case this year was a vendor shifting from “unlimited” to a credit model and facing significant user pushback. The point is even the biggest companies are iterating and they do not have a final answer.
Buying models are moving from owning to hiring. Think transportation. You can buy a car (on-prem license), lease (subscription), rent (usage), call an Uber (on-demand usage), or hire a chauffeur (managed outcomes). As options expand, more people can access the service. Upfront costs drop, total spend at success can rise, and the market gets bigger.
AI products follow a similar path. Buyers want flexibility. They may start with one model, then switch for predictability or cost control. Do not think one model only. Offer options and choose what best aligns with your business.
As you move from license to subscription to usage to outcomes, the vendor takes on more risk and charges closer to realized value. You can capture a larger share of the economic value you create, but only when you can prove and measure it.
Flat-fee subscriptions
Seat-based pricing
Usage-based pricing
Hybrid models — usually platform fee plus usage or credits
Outcome-based pricing
We surveyed companies on their primary model 12 months ago, today, and in three years. A year ago, flat fees and seats dominated. Usage was about 20 percent, outcomes about 3 percent, hybrids about 27 percent. Today and looking ahead, hybrids are surging. The most common hybrid is a platform fee plus credits. Seat count is not a great proxy for value in AI. Value can increase even as fewer people log in. We also cannot afford unlimited plans given usage variance and costs. Expect big declines in pure flat and seats, big increases in hybrid credit models.
The north star is outcomes. Few predict flat or seat models as their future primary approach. Many expect outcome-based or hybrid with an outcome component, such as a platform fee plus an outcome bonus.
Get clear on the outcomes you provide and how to attribute them to your product. That helps if you want to charge on outcomes, but it also helps hybrid and credit models. You can explain the ROI behind a credit so buyers accept variable spend.
A classic framing comes from Twilio. If a buyer struggles to connect SMS volume to value, anchor to the workflow. Appointment reminders reduce no-shows. What is your no-show rate and cost today? What reduction is realistic if we message the day before and one hour before? A message costs less than a penny. The buyer can accept a variable model if the ROI is obvious and quantified.
Q: Examples of outcome-based pricing in AI? Kyle: Chargeflow is a good example. They use AI to manage chargebacks and only charge when they recover money. They may take 25 percent of incremental recovered funds. Industry win rates are often 10 to 20 percent. Chargeflow customers see 60 percent or higher. As they win more, they build a moat. Even a cheaper competitor with a lower win rate yields less net value. Outcome pricing can create pricing power if you lead on outcomes. Intercom has outcome elements in some products. In CX, companies like Sierra, Decagon, and Crescendo are exploring outcome models. I am seeing it appear in more categories.
Q: How does this translate to consumer AI? Kyle: Consumer use cases are simpler, with less usage variance. Flat rates are more common, but there are creative models. Some consumer apps help travelers get airline compensation for delays and take a cut of the recovered money. Credit models are also familiar to consumers. Think ClassPass credits, meal kits, or gaming. Credits are not new. They make sense when usage varies.
Q: What if outcomes take months or years to realize? Kyle: Aim for micro-outcomes you control and can measure sooner. You do not need to charge only on revenue or total cost savings. In support, resolved conversations can be a proxy. In services, completed tasks or verified work products. Choose metrics that correlate with end value but are observable quickly. That keeps the model tied to value without waiting for long-cycle impact.
Q: Penalties when outcomes are not met? Kyle: The common pattern is a platform fee plus an outcome bonus. If outcomes are not delivered, the bonus is not paid. The key is alignment on definitions and measurement up front. Example: what is a resolved support conversation. A customer closing a chat does not always mean resolution. They could return later with the same issue or churn. Get crisp on definitions, evidence, and attribution so both sides feel the charges are fair. Most friction comes when vendors take credit for value the customer believes they created.
Kyle: Thanks for having me.