How to Price Your AI Product: A Complete Framework for Choosing the Right Model
7 MIN READ
|
Updated on August 25, 2026
Picking an AI pricing model often starts the same way: look at what a competitor charges, land on something close to it, and move on. Then, the model scales, the invoices start coming in, and the margin doesn’t look anything like what got modeled at launch. Nobody can quite say when it went wrong, because the pricing decision and the profitability problem happened in two different meetings, months apart.
Nearly half of AI company leaders, 45.9%, manage pricing and profitability as separate disciplines. They set a price, and they track their profit margins, but they don’t connect the two.
Chargebee surveyed 1,336 AI company VPs and C-suite leaders across the U.S., Australia, Canada, the U.K., and Germany in 2026, and the 24% who connect pricing to margin run gross margins roughly 10 points higher than everyone else.
Here’s how to join that 24%. Skip the “usage-based vs. subscription” debate; it’s the wrong first question. Seven factors, specific to your product, determine whether a model protects your margin as you scale.
Copying Competitors Is Costing You Margin
More than one in four AI companies choose a pricing model by copying a competitor, the single most common method. It’s also the second-worst for gross margin: companies that do so see 37.7% margins on average, against 45.3% for companies that run structured pricing experiments. That’s a 7.6-point gap, and the experimenters are 1.9 times more likely to hit margins above 61%.
A leader at a $20M–50M ARR company running a hybrid model described the cost of hesitating: “We should have raised prices sooner. We were afraid of churn but never tested it. When we finally ran a price increase test, the retention impact was almost nothing.”
The fix is direct: test a price with a real segment, track retention and margin for a full cycle, then decide, rather than anchoring to whatever a competitor already charges. That discipline is what the seven factors below help you apply.
The 7 Factors That Determine Your Pricing Model
Chargebee’s research points to seven important factors that carry the most weight when AI companies choose between usage-based, outcome-based, hybrid, and other models. They don’t work in isolation, and your product may pull in different directions on different factors.
- Output measurability. The first question is simple to ask and hard to answer: can you define what “work done” by your tool looks like? If you can, outcome pricing is on the table, since you can bill on what the AI achieved. If you can’t yet measure it cleanly, don’t force it; stay on usage or effort-based pricing until the definition holds up. Fin (formerly Intercom) had to answer this when it launched its AI support agent. Seat-based pricing didn’t fit a system that resolves conversations on its own, and customers balked at paying per interaction while still covering the cost of their human team. The team settled on one outcome everyone could trust: a conversation resolved by Fin with no human involved. That only worked because the company had already been tracking resolution rates long before Fin existed, so the metric and the billing system to meter it were both sitting there waiting.
- Usage spread. Pull up your heaviest users next to your typical ones: your P90 customer (using more than 90% of your customer base) against your P50 customer (the median user). If the P90 consumes five times or more what the P50 does, flat pricing is bleeding margin on your best accounts, and it’s time to evolve. If the spread is tighter than that, usage-based pricing alone may not unlock much value; check whether usage and value move together before you commit to it.
Sudowrite ran into this in stages. It started at a flat $5 a month, rising to $10, then $20, with no usage limit at any tier. By the time it reached $20, some of those customers were costing the company $400 a month, because Sudowrite lets writers choose their own models and chain several expensive calls together. The spread forced a pivot to a words-based metric, then another to credits. The company is now on its fourth pricing model, treating it as a live experiment rather than a finished answer.
- Cost proportionality. Does your cost of goods scale in a straight line with usage, or does it cliff at certain thresholds? Linear costs pass through cleanly under usage-based pricing. When costs cliff instead, credit bundles and cost caps protect margin better than open consumption billing does. Either way, cost sets your margin floor, not your price.
Salesforge, an AI sales platform, builds this logic in from the start: every product’s pricing model mirrors how its underlying cost behaves. Flat provider cost gets a flat fee. Cost that scales with the customer runs on volume or graduated tiers. A provider that charges per event gets billed per event to the customer. Before any price is set, the team models the worst case, not the average, on the idea that a price only holds if it survives the customer who uses the product hardest. Competitor pricing gets checked afterward, as a sanity check rather than the method. When the math stops working, Salesforge’s first move is upstream, renegotiating with the supplier or switching providers, before the customer’s price ever changes.
- Customer predictability. Do your customers budget spend on a set cycle, or does usage swing with the seasons? Customers on a fixed budget cycle need a ceiling: committed contracts with overage caps. If they don’t budget on a cycle at all, pure consumption pricing can work without a commitment- or cap-based structure.
Gorgias sells heavily into e-commerce, where Black Friday and Cyber Monday create enormous, brief spikes in support volume. A merchant on a pure usage model could sail through the year, then get hit with a painful overage bill during the exact week revenue pressure is highest. Gorgias moved more customers onto annual plans instead, giving merchants room to absorb the seasonal swings without getting punished for holiday demand.
- Value correlation. More usage doesn’t always mean more value; sometimes it plateaus, and sometimes it varies wildly by user type. Where usage and value keep climbing together, expand on usage; your heaviest users are your best customers. Where value plateaus, a cap-based or milestone model protects margin better than open-ended usage billing does.
T2D2, which uses computer vision to detect structural deterioration in buildings and infrastructure for Fortune 500 clients, built its pricing around exactly that correlation. The team combined seat-based pricing with usage caps that create a natural upgrade path: as a customer’s consumption grows, they move automatically into a higher-value tier, no manual renegotiation required. Pairing that with a self-serve track for smaller teams alongside its enterprise sales motion helped triple revenue in 24 months.
- Competitive anchor. Know what the default pricing model is in your category before you decide whether to follow it or fight it. Aligning with the default cuts friction. Diverging from it can differentiate you, but it costs more in customer education.
Super, the AI search tool built by Slite’s founders, chose to fight it. Products like Fin had popularized outcome-based pricing, and coding assistants had leaned into usage-based billing. Super went with seat pricing and a fair-usage limit instead, on the theory that internal teams under-explore a product when every click visibly burns a metered credit. They know it leaves some usage-based revenue on the table, but they’re betting adoption is worth more.
- Contract horizon. Whether you sell monthly and self-serve or on annual commits shapes how usage pricing has to work. Monthly plus usage is the natural fit for product-led discovery. Annual plus usage needs a runway commitment and clear overage terms built in from the start.
CodeRabbit needed both to work under one roof. With over 50,000 installs through GitHub Marketplace and thousands of paying customers, the AI code review platform relied on a free, self-serve tier for developers discovering the product alongside a genuine enterprise motion for larger teams buying on annual terms. Building flexible subscription management for both meant it could expand from pure product-led growth into enterprise sales without slowing down the self-serve signups that brought developers in the first place.
The seven factors above aren’t a one-time decision. 37% of AI companies plan to change how they price in the next 12 months. Credit bundles, which decouple price from raw consumption, grew 126% year-over-year. Outcome-based billing has moved from a handful of early movers to roughly 13% of the market. Revisit the framework against your own product every time your cost structure or your competitors’ pricing shifts, not just once at launch.
Where to Start
Choosing the right model doesn’t help much if you can’t act on it: 87% of AI companies lack spend alerts, 79% can’t produce enterprise invoicing, and 78% can’t reprice without an engineering sprint. Not sure where your own model leaves margin on the table? Chargebee’s AI Pricing Diagnostic is a directional tool that scores your product against these same signals (output measurability, usage spread, and cost exposure) in about three minutes, then builds a 90-day roadmap to close the gap. Plus, you can watch How to Nail the Price of Your AI Product to learn everything you need to know about pricing your AI product from Chargebee’s experts.
