How to Price AI Products

Price an AI product against the value the customer can measure, and treat your token bill as the floor under the price. Choose a unit that grows with the customer's result, launch at 10% to 15% of the value you create and climb toward 30% to 50% as the results come in, the range I set out in Pricing AI Products. Then fence the meter so the buyer can predict the bill.

Price deserves that much care because it moves profit more than any other lever. A 1% rise in price lifts operating profit by about 11% on average, against about 7% for a 1% cut in variable cost and under 4% for 1% more volume, in the studies of the S&P 1500 the book cites. Founders still set the number once under launch pressure and leave it for years, and an AI product punishes that habit harder than software did, because its costs move with every request and its value swings widely from one customer to the next.

The pricing thermometer

Three quantities decide every AI price, and Figure 1, the pricing thermometer from Chapter 2 of the book, stacks them. Cost is what you spend to deliver the product: compute, data and the engineers who build it. Price is what you charge. Value is the benefit the customer receives in saved cost, added revenue or new capability. The gap between price and cost is your margin, the reason to sell and scale. The gap between value and price is the buyer's incentive, the reason to sign.

A generative tool that lifts the conversions of a $100,000 campaign by 15% creates $15,000 of value. Priced at $3,000, it leaves the customer $12,000 of incentive and still earns the vendor a healthy margin over its compute. Cost-plus pricing starts at the bottom of the thermometer and adds a markup, and for an AI product it fails twice. It ignores the large value a fraud model or a support agent delivers, and it ties the price to a token cost that keeps falling, so every cut in model prices becomes a discount you owe the customer instead of margin you keep.

The AI pricing models

Five models cover almost every AI product on the market, and each rests on a value metric, the unit you count to decide what the customer owes. Table 1 lists them with a live price for each.

Table 1 · AI pricing models with a live example of each

Model definitions from Chapter 3 of Pricing AI Products. Vendor prices as published in October 2026.

The average AI company runs 1.7 of these models at once, ICONIQ found in 2026, usually a subscription or platform fee the buyer can budget plus usage, tasks or outcomes on top. The choice among them turns on how fast the customer sees the value, which AI pricing models compared works through model by model, and on how much usage varies between your smallest and largest account.

The capture ratio

The number to watch is the share of value you keep. I call it the capture ratio: the value you capture divided by the value you create. Classic SaaS settled near 10%, a $100,000 contract on a product worth $1 million a year to the customer. Outcome-priced AI can push the ratio toward a third or a half, because the price moves with the result.

The ratio turns every pricing choice into one question: does this raise my capture ratio in a way the customer still sees as fair? Per-seat pricing drives the ratio down over time for an AI product, because the value climbs while the seat count stays flat or falls. An AI that lets one analyst do the work of five has delivered a fortune, and a per-seat contract collects the same fee as before, or less once the team shrinks.

A worked launch price

Take an illustrative AI product that saves an enterprise customer $500,000 a year. Winning the account costs $50,000 in sales effort, and serving it costs $30,000 a year in compute, support and success, so the first year costs $80,000. Launch at 10% of value, $50,000, and every new customer loses $30,000 in its first year, so each logo the sales team wins deepens the hole. Price at 30% of value, $150,000, and the same account earns $70,000 over its cost while the customer keeps $350,000 of the value, a trade any finance team will sign.

The product is identical in both cases, and the capture ratio is the only thing that moved, from 10% to 30%. The book calls the $80,000 mark the viability line, the price below which the unit economics cannot fund the cost of winning and keeping a customer. Plot it before you set a launch price, because early traction below the line looks like growth until the losses compound and the next round has to pay for them. Pricing AI Products vs Monetizing Innovation charts this account at both prices and sets the method beside the nine rules of the 2016 book.

The value metric decides the growth

Two products on the same model can end up in different businesses because of the unit they count. Chapter 8 of the book scores every candidate unit on four gates, from 0 to 2 each: can you run it, does the customer pull on it, does it feel fair, and is each unit worth about the same. A unit that scores under 5 of 8 gets fixed or dropped before anyone argues about the number.

For an AI sales agent, a price per contact enriched scores 4, because enrichment is an input the buyer works to use less of. A price per meeting booked, at roughly $40, scores 7, because the meter climbs with the customer's success. Price as close to the payoff as your attribution can defend, then move the meter toward the outcome as your evidence improves. Seats suit AI only where value still grows with the number of people using the product, and the SaaS to AI pricing migration covers moving an installed base off them.

A meter the buyer can approve

The force that caps a metered AI product is rarely the rate. Buyers fear a bill they cannot predict, so they ration the product to a cautious few or refuse to sign, and the book calls this consumption anxiety. Four tools remove it without lowering the price.

• A subscription floor that makes most of the bill known in advance.
• A cap the customer sets, so the worst case is fixed before the month starts.
• Prepaid credits that roll over, so a slow month carries forward instead of wasting money.
• Alerts at half and four-fifths of budget, with a live usage dashboard behind them.

Credits also fix your cash flow. A media company that buys 100 video-generation credits for $5,000 knows its cost in advance, and you collect the cash before you spend a dollar on compute. The same cap sets your own margin floor: at a 40% floor, a $0.99 resolution price allows about $0.59 of compute per resolution, which is the arithmetic behind pricing an AI agent against the labor it replaces.

Reprice on a cadence

An AI price fits for a quarter, because model costs fall and capability rises every few months. Keep early commitments monthly or quarterly, and lengthen them once the value has steadied. Treat each real jump in capability as a reason to revisit the price, the same reset a renewal or a major release gives you. Plan the climb from 10% to 15% of value toward 30% to 50% before launch, because the longer you wait, the harder the jump becomes to justify.

The fastest money most AI startups can find sits in their own base. A company with 1,000 customers paying $2,000 a month that raises the price 5% with little churn adds $100,000 of monthly recurring revenue, $1.2 million a year, without another dollar of acquisition spend.

Questions

How should I price my AI product?

Price it against the value the customer measures, with your cost as the floor. In Pricing AI Products, Nitin Kumar recommends launching at 10% to 15% of the value you create, climbing toward 30% to 50% as results come in, and choosing a unit that grows with the customer's result.

What is the capture ratio in pricing?

The capture ratio is the value a company captures divided by the value it creates for the customer, a measure Nitin Kumar uses in Pricing AI Products. Classic SaaS settled near 10%, and outcome-priced AI can reach a third or a half of the value created.

What percentage of value should an AI product charge?

Pricing AI Products by Nitin Kumar puts the launch price at 10% to 15% of the value created and the mature price at 30% to 50%. A product that saves a customer $500,000 a year would launch near $50,000 to $75,000 and grow toward $150,000.

Should AI products use cost-plus pricing?

No. Nitin Kumar's Pricing AI Products treats cost as the floor under the price, because cost-plus ties the number to a token bill that keeps falling. Each cut in model prices then becomes a discount for the customer instead of margin for the vendor.

What is consumption anxiety?

Consumption anxiety is a buyer's fear of an AI bill they cannot predict, a term from Nitin Kumar's Pricing AI Products. It caps adoption more than the rate does, and a subscription floor, a customer-set cap, rolling credits and budget alerts remove it.

Sources

Nitin Kumar, Pricing AI Products, Chapters 1, 2, 3, 6, 8 and 10, 2026. Salesforce, Agentforce general availability, October 2024. Chargeflow pricing on the Shopify App Store. Zeliq, Clay pricing 2026. ICONIQ, 2026 State of AI Report.