AI 如何改变 SaaS 定价：从按席位收费到按成果收费

For a long time, software pricing followed a simple rule: one person, one licence, one predictable monthly charge. That rule worked when a tool mainly waited for a human to use it. Artificial intelligence has changed the balance. An assistant can now answer customers through the night, draft material, qualify enquiries and carry out routine tasks with little direct supervision. The bill is therefore less about who can log in, and more about how much work the system performs and what that work achieves.

Why counting logins no longer tells the whole story

A single employee can now set off far more activity than before. One support specialist, helped by AI, can handle a larger queue, and one marketer can produce many more variations. Not all AI activity costs the same to run either: a brief reply and a long, multi-step workflow place very different demands on computing resources. There is also a possibility that seat pricing never had to consider: an agent can complete work with no human seat involved at all. Seats are not vanishing overnight, and they often remain the entry fee for access, but they are now frequently only one line in a larger agreement.

Four models, in plain terms

Seat-based pricing charges a fixed amount for each named user. It is easy to forecast and approve, but it pays for access rather than use, so unused accounts still cost money and busy AI workloads are not reflected in the charge.

Usage-based pricing charges for measured activity, such as messages handled, records processed, minutes used or tokens consumed. It can suit uneven workloads because quiet periods cost less, but it can be volatile. A surge in demand, a poorly designed automation or a workflow that repeats itself can push the bill sharply upwards.

Credit-based pricing turns different actions into a shared currency. A business receives or buys a pool of credits, and each feature draws from that pool at its own rate. This can make a broad suite easier to manage, but only if the rate card is transparent. Without clear information about what each action costs, whether credits expire and what happens when the pool runs out, a generous-looking allowance may not stretch as far as expected.

Outcome-based pricing charges only when an agreed result is reached, such as a customer query being resolved, a lead being qualified or a defined task being completed. This comes closest to linking payment with value, but everything depends on the definition of success: what counts, who decides, and what happens if a result is later reopened or found to be wrong.

In practice, most agreements blend these approaches, with a base charge for access plus usage, credits or outcomes for AI work. Comparing only the headline seat charge, or only the headline outcome charge, therefore gives a misleading picture.

Why sellers are moving in this direction

The first reason is cost. Once built, a traditional feature costs little extra each time it is used. AI features are different, because each request consumes computing resources, and heavier work consumes more. A flat charge for unlimited AI use can therefore become difficult to sustain as adoption grows. The second reason is value: when an agent performs work that people used to do, charging purely by headcount can seem disconnected from the benefit received. The third reason is competition, as newer AI-led products often start with usage or outcome pricing, pushing established suppliers to add similar options.

The full ExactReads guide to AI SaaS pricing models discusses public approaches from support, model and customer-platform suppliers, including outcome pricing where only defined results are billed, token costs sitting beneath finished products, and credits used as a common currency across many features. The lesson is consistent: the model matters less than the rules around it.

What buyers should work out before signing

Start with your own volumes, not the supplier's demonstration.How many conversations, leads, documents or agent tasks do you expect each month?Split that total by result, because successful completions, escalations, failures and reopened cases should not be lumped together: only some may be billable, while the rest still need human time.Model the peak rather than only the average, since a launch week or service incident can look very different from a normal month.Ask which actions consume credits or tokens, whether retries and failed attempts are charged, and whether unused allowances roll over.

Before committing, ask for a sample invoice based on your volumes, with each line explained, and set alerts and spending limits before go-live.

Watch the risks too. A case counted as resolved simply because a customer stops replying may be a person who gave up, not one who was helped, so audit a sample early and compare billed results with reopen rates. Automation can increase spending quietly, and several teams drawing from one credit pool can exhaust it faster than expected. At renewal, compare the whole contract value from year to year, rather than only the familiar seat line, and keep your own measures of volume, quality and cost.

AI is moving SaaS pricing from paying for access towards paying for work and results. That shift can be fairer for both sides, but only when work is defined clearly, measured honestly and tested against real volumes. For the fuller explanation, including a model-by-model comparison and a practical buying checklist, ExactReads publishes the detailed guide this summary is based on. Bring your own numbers to the negotiation: in AI pricing, the cheapest contract is rarely the one with the smallest headline figure. It is the one whose rules you fully understand.