SaaS to AI Pricing: How to Move Your Base Off Seats

To move from SaaS to AI pricing, change the unit from a seat to the work the product finishes, keep a platform fee for the record and workflow, and move the installed base in waves.

New customers go first, existing accounts get a shadow bill for 60 to 90 days, every account sees its projected bill before cutover, the heaviest and happiest users move first, and the hardest-hit accounts get a cap or a credit. In Pricing AI Products I set the danger line at the 10% to 15% churn spike that an unprotected price change routinely triggers, and the whole sequence exists to keep a migration under it.

Seat prices become pricing debt

A seat price fit software that helped a person work. It misfits AI that does the work, because the customer buys the AI to need fewer people. Pricing debt is the accumulating gap between what you charge and what you should charge, and it compounds every month you hold the old number. A vendor that should charge $150,000 a year for the value it delivers but holds a legacy $80,000 price carries $70,000 of pricing debt per account. At 50 accounts that comes to $3.5 million of annual margin forgone, and the income statement never shows it because the revenue never arrived.

AI adds a cost problem to the debt. Serving one more seat cost almost nothing. Every AI resolution, document or generated file burns compute. ICONIQ’s 2026 State of AI report put the average gross margin on AI products at 45% in 2025. A team that carries its old seat price into an AI product often finds the margin gone, because the price assumed a near-zero cost to serve.

Where the human sits sets the unit

Human-in-the-loop products keep a person checking the AI, as in sensitive content moderation, and their economics still resemble software. Human-outside-the-loop products deliver the outcome alone, as in autonomous support, and need usage or outcome pricing because there is no seat left to sell. As a product moves from the first mode to the second, its price has to move with it. Table 1 sets the two models side by side.

Table 1 · SaaS pricing and AI pricing compared

The hybrid that holds

The surviving structure has two parts. A platform fee covers the record and the workflow and gives an enterprise buyer a budget line it can forecast. A prepaid commitment, drawn down by units of finished work with a cap and an overage rate, pays for what the AI completes. Keep the record on the platform fee, because a customer charged each time it opens its own records will ration access to them. Set the cap with care, because a customer who opens a bill twice the size of the last one stops trusting the meter.

Every unit has been tested in public. Microsoft priced Microsoft 365 Copilot at $30 per user per month in July 2023, a seat price that stays flat while a heavy user’s compute grows. Salesforce launched Agentforce at $2 per conversation in September 2024 and added Flex Credits at $0.10 per action in May 2025. ICONIQ found usage pricing at 42% of AI companies, up from 35%, and outcome pricing at 23%, up from 18%, with the average company running 1.7 models at once. Price outcomes only where the customer can verify them in its own numbers, the way Fin counts a resolution only when the customer confirms the problem is solved or asks for nothing more, because an outcome the buyer’s finance team cannot find will not survive the first renewal. The level of the price, set against the labor budget the AI replaces, is covered in AI agent pricing against the labor budget.

Reset windows

Customers accept a price change only at certain moments. I call them reset windows. A major release that visibly adds value opens one, and so does a renewal, a public shift in your cost base or a move that reprices the whole category. Between windows, an increase reads as opportunistic and meets resistance, while inside one the same increase reads as the plain result of something the customer can already see in the product, the contract or the market. Tie each planned price step to the next release or renewal and take the smaller, well-timed increase, because a company that reprices a little at every window pays its pricing debt in installments its customers barely notice.

A barbell for the first rollout

Learn two numbers before you touch the base. One is what the AI is worth, and the other is what it costs to serve. Load both ends of a barbell and leave the middle empty. On the heavy end, give the AI to the broad base embedded by default, with a generous allowance and a cost-safe meter the CFO signs without a fight. On the light end, give a handful of willing accounts close to unlimited use at close to no charge, and watch what they do when nothing constrains them. The meter on the heavy end suppresses the usage you need to measure, so the light end is your only clean read on the ceiling. The mistake is to meter everyone cautiously in the middle, learn nothing about the ceiling and protect a margin you could have grown.

The repricing sequence

An installed base is harder to reprice than a new product. Existing customers are anchored to a number. Bain’s October 2025 review of more than 30 SaaS vendors found about 65% had added an AI meter on top of seats and the other 35% had raised seat prices to bundle AI. Every vendor in that sample still bills by the seat and has this migration ahead of it, and Table 2 sets out the sequence I use.

Table 2 · A sequence for repricing the base without a revolt

Shadow billing pays back the most. Customers forgive a price change they saw coming and churn over one that arrived without warning. Cursor learned this in public. It changed the meter on its Pro plan in June 2025, heavy users met charges they had not expected, and the company apologized in July and offered refunds. Prove the new model on new logos first, ideally in a secondary segment or a new region where a stumble costs least. Put it in front of your own sales team before a single buyer sees it.

A worked migration

Take an illustrative document-AI company with 1,000 customers on a flat $1,000 a month, or $1 million of monthly recurring revenue that has stopped growing. The shadow bills split the base three ways. The top 200 accounts would pay an average of $1,450 under a usage meter. The middle 500 stay close to $1,000 and the bottom 300, the lightest users, fall to an average of $750. Sequenced, the top 200 move first and add $90,000 a month, the middle 500 move next with little change, and the bottom 300 move last and give back $75,000. Monthly revenue ends at $1,015,000, a 1.5% gain on the same base, with every customer now on a price that grows with use.

Now flip the whole base in one month and skip the shadow bill. The same end state arrives with the objections and the revenue dip at once, and a 12% churn spike, inside the range an unprotected change triggers, removes about $122,000 of the $1,015,000. The migration that should have added $15,000 a month loses about $107,000, so one missed protection costs seven times what the whole repricing was worth. I use this arithmetic with founders who want to skip the shadow bill to save a quarter.

Watch the churn threshold weekly

The threshold splits two outcomes. Below it, losses concentrate in the lightest accounts, which often cost more to serve than they pay, and expansion from the repriced heavy users covers them. Above it, churn spreads into accounts you meant to keep and references turn negative. Track churn every week of the migration. The threshold is crossed fast once it is crossed at all. Report net revenue retention for the new-model cohort on its own line, apart from the grandfathered base, so the old contracts cannot hide whether the new price retains, because that one line tells you the migration is working while there is still time to slow down, add protection or pause. When the light user says the bill went up for no reason, answer with the usage and the outcome. If the new price does not track value for that segment, a cap or a lower floor is the honest fix.

The back office moves too

Finance feels the change first. Usage and outcome revenue recurs in amounts that vary, so the smooth ARR line of seat pricing becomes a lumpier stream that finance has to forecast. Billing, revenue recognition and collections all change at once, and a sound price on a broken invoice destroys the trust the price depends on. New Relic shows where a finished migration lands. By the quarter ended June 2023, consumption revenue made up more than 88% of its revenue and gross margin had risen to 77.6% from 70.5% a year earlier, TechCrunch reported. Budget for a liquidity buffer in the quarters when revenue arrives unevenly. Treat repricing as a capability you will use again, because outcome pricing will give way to whatever agents make normal next. The wider pivot this pricing change sits inside is in the SaaS to AI pivot.

Questions

How do you move from SaaS pricing to AI pricing?

Change the unit to the work the product finishes, keep a platform fee for the record, and move existing customers in waves. Nitin Kumar’s Pricing AI Products sets a five-move sequence: grandfather, shadow bill, show the bill first, phase the cutover and protect the hardest-hit accounts.

How long should you grandfather customers on seat pricing?

Twelve to eighteen months, with the end date stated at the start. An open-ended promise splits the base into two pricing systems you maintain forever and removes the pressure that moves legacy accounts to the new model.

What is pricing debt?

Pricing debt is the accumulating gap between what a company charges and what it should charge. A vendor holding $80,000 for work worth $150,000 carries $70,000 per account a year, or $3.5 million for 50 accounts.

How much churn should you expect when you change pricing?

An unprotected price change routinely triggers a 10% to 15% churn spike, which Nitin Kumar treats as the revolt threshold. Shadow billing, projected bills, phased waves and caps keep a migration below it.

What is a shadow bill?

A shadow bill meters each customer on the new model for 60 to 90 days without charging it. It gives the company real usage data and lets every customer see its projected bill before the change takes effect.

Should AI pricing be usage-based or outcome-based?

Use a hybrid for an installed base: a platform fee for the record and workflow plus a prepaid commitment drawn down by units, with a cap. Move to outcome pricing only where the customer can verify the outcome in its own numbers.

Sources

Nitin Kumar, Pricing AI Products, Chapters 28 and 29, 2026. ICONIQ, 2026 State of AI Report. Bain & Company, Per-seat software pricing isn’t dead, October 2025. TechCrunch, Cursor apologizes for unclear pricing changes, July 2025. TechCrunch, New Relic quarterly results, July 2023.