For twenty years, buying business software meant counting people. You paid a set price per user per month, multiplied by headcount, and the budget was easy to write down. AI agents break that arithmetic. When software does work on its own, the vendor needs a new way to charge, and a growing number now charge for what the software does rather than who logs in.
That can be fair, and it can also make your software bill hard to forecast. This guide covers how AI software pricing models differ, what vendors and analysts have published, and what to put in a contract before you sign or renew.
Why Per-Seat Pricing Is Giving Way
An agent that answers support tickets or updates records does not need a seat. If it works well, you may need fewer human users, which is exactly what a per-seat price punishes the vendor for. Pricing analyst Kyle Poyar's survey of more than 240 software companies, run in April and May 2025, found seat-based pricing down from 21% to 15% in a year, while hybrid pricing rose from 27% to 41%. Poyar notes that the typical respondent is a US company with $1 million to $20 million in annual recurring revenue, so it describes software makers of that size rather than every vendor.
Gartner forecasts, in a July 2025 research note that Deloitte quotes and a June 2026 CIO.com article repeats, that by 2030 at least 40% of enterprise SaaS spend will move to pricing based on usage, agents or outcomes. That is a forecast, and we read it second-hand, not in Gartner's note. The same article quotes a Twilio strategy director saying hybrid wins as the dominant model, with pure outcome pricing "aspirational for most categories". Bessemer Venture Partners reaches a similar view in its February 2026 AI pricing playbook: hybrid models, a base subscription plus usage or outcome tiers, "win when you're uncertain".
So expect a mix at your next renewal, not a clean switch. Our look at how AI agents will change business software explains why the shift is happening.
What Vendors Actually Charge Now
These are vendor-published figures. The Microsoft and Intercom pages were read on 25 October 2026; Salesforce's figures come from its May 2025 announcement and may have changed. Prices change often, so confirm them before you budget.
Salesforce Agentforce. At its May 2025 launch of flexible pricing, Salesforce said Flex Credits are sold in packs of 100,000 for $500, and each agent action uses 20 credits ($0.10). It also listed per-user licenses, and a $2-per-conversation option that existing customers already used (vendor data).
Microsoft Copilot Studio. Microsoft lists a prepaid pack of 25,000 Copilot Credits at $200 per month, or pay-as-you-go billed at the end of the period through an Azure subscription (vendor data). Agents published to Microsoft 365 Copilot are included for users who hold that license.
Intercom Fin. Intercom's pricing page says Fin is priced at $0.99 per outcome, which includes confirmed resolutions and conversations where the customer does not ask for more help. Intercom's page also says a minimum monthly commitment applies when Fin is used with an existing helpdesk, and gives 50 outcomes as an example (vendor data).
Zendesk. In August 2024 Zendesk announced that customers pay only for issues resolved autonomously by AI, with a starter level of usage included (vendor data). The announcement gave no per-resolution price.
ServiceNow. In March 2024 Constellation Research reported that ServiceNow's CFO described future pricing that may include seat-based fees alongside consumption (Constellation's paraphrase of the earnings call).
Our reading of the list: most of them pair a base fee with a meter. The seat has not disappeared. It has been joined by a unit you cannot count in advance, such as an action, a credit or a resolution.
Comparing AI Software Pricing Models
Model | How you pay | Main risk to you | When it fits |
|---|---|---|---|
Per seat | Fixed price per named user | Paying for users who rarely use the AI; cost does not follow value | AI is an assistant to staff and usage is steady |
Usage or credits | Prepaid or metered units (actions, tokens, messages) | Bills vary with behavior you do not control; vendors may change what a credit buys | Technical teams that can monitor usage and set limits |
Per outcome | A fee per resolved case or completed task | Disputes over what counts as an outcome; costs rise as volume rises | High-volume, well-defined work such as support tickets |
Hybrid | Base subscription plus a metered or outcome layer | Two bills to forecast; the base can look cheap while the meter grows | Most mid-size buyers, if the meter has a cap |
Capped or enterprise agreement | Fixed fee for defined AI access | Exit terms and renewal pricing after the cap period | You want a firm budget and can commit for the term |
Bessemer frames the trade-off from the vendor's side: moving from consumption to outcome pricing means "accepting more cost risk in exchange for tighter alignment with customer value". Our view: as a buyer, expect the vendor to try to move that risk to the part of the contract you read least.
The Budget Risks Finance Teams Should Plan For
The first risk is volatility. In the CIO.com article, the Twilio director calls budget volatility the most immediate problem, and says the shift "transfers forecasting risk from vendor to buyer". A seat price capped your bill at headcount. A meter does not.
The second is a moving rate. Gartner's Hannah Decker told The Register in January 2026 that vendors often keep the right to alter the multiplier, meaning the rate at which credits are used up, so costs can rise while your usage stays flat. Salesforce disputed Gartner's wider view that it was moving away from capped agreements, so read the same note as a warning about contract terms, not a verdict on one vendor.
The third is the renewal. The same Gartner note flagged Salesforce increases of 6 to 15 percent as likely to carry into renewal costs, and warned that generative and agentic AI products could push totals up sharply. Decker advised that if you move to a defined quantity, renewal price increases need limits.
The fourth is the definition of "resolved". Intercom's page counts a conversation where the customer simply does not ask again. That is a reasonable rule, and it still means your invoice depends on how a vendor measures success, so test it against your own tickets in a pilot.
A Buyer Checklist Before You Sign
Usage cap. Ask for a monthly spending ceiling with alerts at 50% and 80%, so a busy month cannot become a surprise bill.
Overage terms. Get the price per unit beyond your allowance in writing, and whether the service stops or keeps billing when you hit the limit.
Rate lock. Ask that the credit or unit rate cannot change during the term without your consent.
Renewal limit. Cap the price increase at renewal, in percent, for every part of the bill.
Definition of a billable event. Get exact wording for what counts as a resolution, action or message, plus a way to dispute counts.
Usage reporting. Require a dashboard or export that shows consumption by team and by month.
Pilot pricing. Run a limited pilot at a fixed fee, and agree that pilot results set the volume commitment, not a sales forecast. Our guide to scoping an AI pilot shows how.
Exit and data export. Confirm you can export your data, prompts, configurations and logs in a usable format, with a wind-down period, and a clear rule on what happens to prepaid credits you have not used.
Ownership and privacy. Check what the vendor may do with your data. See keeping business data private with AI.
This checklist is our suggested practice, not a figure from a source, and the 50% and 80% alert levels are examples. Have your lawyer review the final wording, since terms differ by vendor and contract.
Where a Custom-Built Tool Fits
Buying is not your only option. In our view, a tool built for your own process has a different cost shape: a larger upfront project, then hosting, model usage and maintenance that you control and can measure. There is no per-seat fee or vendor-set meter, but you take on the build and upkeep, so it rarely wins for common needs like a standard helpdesk or CRM. It tends to make sense when one workflow is central to your business and the per-unit charges of a bought product would grow with your volume. Our comparison of custom software and off-the-shelf walks through that decision, and our guide to calculating the ROI of automation helps you test the numbers.
Where to Start
List the AI features you pay for today, how each one is charged, and what your last three months of usage would have cost under a meter. That one page makes your next renewal conversation much easier. If you want a second opinion on a proposal or help weighing buy against build, our software consulting team can review it with you, and our custom software development team can price a tailored alternative. Book a call to talk it through.
