--- title: "If you're rethinking your pricing because of AI, read this first" date: 2026-04-18 description: "The most common mistake: teams look at their AI COGS, panic, and mark it up. That's cost-plus pricing wearing an AI hat." author: Kat Laszlo --- # If you're rethinking your pricing because of AI, read this first If you're a product manager or founder staring at your pricing page wondering whether AI just broke it, you're probably right. I talk to a lot of teams right now who know something has to change but don't know where to start, what order to do things in, or which changes are reversible. Here's what I keep coming back to. ## Pricing has two ends Your cost is the floor, important but the least interesting part of the conversation. It tells you how much discount you can afford on a deal and when to pass third-party costs through cleanly. If you stop there, you've left the real decision untouched. The ceiling is the value you create for the customer. That's the interesting part. The gap between the floor and the ceiling is your entire pricing strategy, and most teams spend too much time staring at the floor. The most common mistake I'm seeing right now is that teams look at their AI COGS, panic, and mark it up. That's cost-plus pricing wearing an AI hat. It caps your upside to whatever multiplier you picked, and it ignores how much more the customer would have paid if you'd anchored on the value. ## How far can you reach toward the ceiling? This is the part most pricing advice skips. "Charge on value" is easy to say. In practice, how far up you can price depends on how closely you can attribute the outcome to your product. That's a spectrum. At one end, you own the data and you touched every step. An AI SDR writes every message and every reply lands in your system, so positive reply rate is a clean value metric. You can credibly charge per positive reply because there's no ambiguity about who generated it. I ran pricing for a data company that did this for identity verification, and we charged whether the answer was yes or no, because a confirmed non-match was just as valuable as a match. The null result has a price, because knowing someone *isn't* who they claim to be is the whole point of the check. Move along the spectrum and you need external data to see the outcome. That same AI SDR could price on meetings booked, but now you need a CRM sync, the prospect had to show up, and a human AE was involved, so the attribution gets murkier. You can still measure it, but you're no longer the only input. At the far end, the outcome is far from your input. You're selling enriched data, and your customer uses it to close deals, qualify leads, and target campaigns. You didn't touch the deal, you don't see their pipeline, and you can't measure the outcome. But here's the thing: even when you can't measure the outcome, you can still price in a way that reflects value. The move is to segment by how much value the customer is likely to extract, even if you never see the result. At People Data Labs, we charged different rates per field bundle based on how customers used the data. Employment history was worth more to a recruiting platform than a marketing team. Company technographics were worth more to a sales tool than a compliance product. The price reflected the use case without requiring us to track what the customer did after they got the data. You can reach toward the ceiling even when you can't see exactly where it is. The goal is not always outcome-based pricing. The goal is getting as close to the ceiling as you can credibly reach, given what you can actually measure. A company charging $80k/year on a flat seat-based plan is pricing based on value if they know why that number is right. A company marking up their API costs 3x is not, regardless of the billing model. The billing model is how the invoice looks. The pricing decision is how you got to the number. ## Pick the billing model after you know the value Three things should drive that choice. The first is how predictable your cost per customer is, because if costs swing a lot, you probably need a usage component so you don't eat the variance. The second is how the buyer wants to pay, since finance teams hate unpredictable invoices and enterprise buyers will pay more for a flat annual commit than for a metered bill of the same average amount. The third is how you want to split risk, where outcome-based means you carry the risk, flat subscription means the customer carries it, and hybrids split it. Subscription, prepaid commit with overages, pure usage, outcome fee, hybrid. They're all valid. The right one depends on the buyer and your cost shape. ## Pricing is a loop, not a one-time decision Once you know your value metric, even an imperfect one, pricing becomes a cycle: 1. Measure the output your customer is getting. 2. Charge in a way that reflects that value. 3. Improve the product using the measurement as a signal. 4. Renew on proof, not promises. 5. Raise the ceiling as the product gets better. Every turn makes the next one better, because the measurement tells you what to build, the improvement raises the value, and the higher value justifies the next price. Teams that treat pricing as a quarterly spreadsheet exercise miss this entirely, while the ones who build the loop into their product figure out what to charge next, not just what to charge now. That's the whole game. It's not setting the right price once, it's raising the ceiling over time.