For as long as most businesses have bought software, the bill has scaled with headcount. A CRM cost a fixed amount per user per month, a helpdesk platform charged per agent seat, a scheduling tool priced by how many staff logged in. That model made sense because the thing being paid for was, fundamentally, human access to a system. It is starting to come apart, and the reason is specific: an AI agent embedded in that same software does not hold a seat, does not log in once a day, and can complete a task in seconds that used to take a person ten minutes. Charging for it the way you'd charge for a human login stopped matching the thing actually doing the work, and several vendors have begun pricing accordingly. Intercom's Fin now charges per resolved customer conversation. Salesforce's Agentforce bills per AI conversation rather than per licensed user. Zendesk's AI add-on charges only when a ticket is actually resolved, not for every attempt. This is a real restructuring of how a software bill gets built, and it is happening inside contracts businesses already have, not just new ones.
The economics behind the shift are worth understanding, because they explain why it is not a passing pricing experiment. A per-seat fee was calibrated to a fairly predictable thing: one person, one login, roughly consistent usage across a workday. An AI agent's usage pattern looks nothing like that — it might handle five customer conversations in an afternoon or five hundred, and each one carries a real, variable compute cost to the vendor regardless of how many humans are on the account. A vendor that keeps a flat per-seat price on an AI feature is either overcharging a business that barely uses it or losing money on one that uses it heavily. Usage-based and outcome-based pricing solves that mismatch for the vendor. It also, done honestly, aligns the vendor's incentive with the customer's result in a way a flat license fee never did — a vendor charging per resolved ticket only gets paid when the tool actually resolves the ticket.
The part that matters for a business budgeting next year's software spend is more mundane: the total on that bill can now move up and down with usage rather than sitting flat with headcount, and usage is a number most finance teams have never had to forecast for this line item before. A helpdesk bill that used to be predictable because it tracked a stable number of support staff can now swing with call volume, seasonality, or a single bad product launch that spikes support conversations. That is a manageable variable once you know to model it. It is a budgeting surprise if the renewal notice is the first place it shows up.
There is a definitional catch worth checking before signing anything billed on outcomes, and it is easy to miss in a sales conversation. "Resolved" is the vendor's term, defined by the vendor's own criteria, and it does not automatically mean the customer's problem actually went away. A conversation the system logs as resolved because the customer stopped responding is not the same as a conversation that ended with the customer's issue fixed, and a vendor whose revenue depends on the resolution count has a built-in incentive to define it generously. Before agreeing to outcome-based pricing, ask for the exact definition in writing, ask to see a sample of conversations logged as resolved, and specifically ask what happens to a conversation that gets escalated to a human afterward — whether that still counts as a billable resolution is a real contract term, not a technicality.
Most of what's actually being sold right now is not pure outcome pricing but a hybrid: a base subscription fee plus a metered component on top, which is easier for both sides to plan around but still needs the same scrutiny. Ask specifically whether a failed or incomplete task is billed at all — a well-structured vendor charges only for a completed resolution, the way Zendesk's model is described; a less disciplined one charges for every attempt, successful or not, which quietly reintroduces the same mismatch outcome pricing was supposed to fix. And negotiate a ceiling. A metered price with no cap turns a single unusually busy month into a bill nobody planned for, and a cap is a normal, biddable term at renewal, not something a vendor volunteers unprompted.
The practical step belongs at the next renewal, not before. Any CRM, helpdesk, or scheduling tool already in use that has added an AI agent feature in the past year is a candidate for this shift showing up in the next contract. Before renewing, ask directly whether that feature is moving, or has already moved, to usage or outcome billing, and ask for several months of actual usage data logged under the current flat fee. Negotiating a metered rate from real numbers the vendor already has on file is a very different conversation than negotiating from the vendor's own usage estimate, and it is a conversation worth having before the new pricing is presented as the only option on the table.
None of this is a reason to be wary of vendor AI features, several of which are genuinely worth what they cost and priced more fairly than the license fees they're replacing. It is a reason to read the next renewal more carefully than the last one. The line item that has sat flat and predictable on the budget for years is being rebuilt underneath a business that isn't watching for it — and the businesses that notice before the renewal call, not during it, are the ones who end up negotiating the terms instead of accepting them.
- saas pricing
- ai agents
- vendor management
- budgeting
- procurement