Share with your CIO
Agentic AI vendors are selling the same capability under three fundamentally different pricing structures, and the contract terms, not the technology, are where enterprise buyers either win or get taken. Brian Jackson at Infotech maps the landscape through a disarmingly simple analogy: buying inflated balloons. Salesforce’s Agentforce has cycled through per-conversation, per-resolution, and flex-credit models in under a year. Horizon Media’s consumption-based platform triggers dozens of pricing disputes per week per client. The pattern across both is that “outcome” is a contractual definition vendors write, not a natural fact.
What this means for your business
The vendor who defines “done” controls the invoice. That’s the actual exposure here, and it falls hardest on organizations moving agentic AI out of pilot and into production workflows where volume makes ambiguous definitions expensive. If your current AI contracts were written when usage was low and predictable, the definition of a resolved interaction or a completed task probably got less scrutiny than it deserved. The question isn’t whether you’re on the right pricing model; it’s whether you control the measurement.
Salesforce’s per-resolution pricing for Help Agent is the clearest illustration of how “outcome-based” is a vendor-constructed category. Salesforce won’t charge if a customer escalates to a human, but Salesforce also caps voice interactions at 10 minutes before the meter ticks up regardless of escalation. That’s not outcome-based pricing in any pure sense; it’s consumption pricing with an outcome-flavored name on the label. Jackson, writing as an Infotech analyst who sells advisory services to the same IT leaders he’s advising here, understandably frames this as a negotiation opportunity rather than a structural flaw in vendor incentives, but the underlying observation is correct: the definitional gray zone is where buyer leverage actually lives.
Horizon Media’s situation reveals the downstream version of the same problem. When clients do their own work on a consumption model and perceive no value in a given query, they dispute the invoice. The agency absorbed real inference costs regardless. This “value perception gap” won’t stay confined to advertising platforms; any enterprise deploying agentic AI for internal knowledge work or customer service will face it the moment a knowledge worker runs a prompt that returns a useless answer and asks why they’re paying for it. The CFO reviewing that line item won’t accept “the user needed more training” as an answer more than once.
The renewal you should be renegotiating is the one where your vendor wrote the success criteria. Infotech’s survey shows 36% of IT leaders still prefer per-user subscriptions, which means most organizations haven’t yet confronted consumption or outcome pricing at scale. That comfort won’t last. The contracts being signed now for agentic capabilities will set precedents for how “resolution,” “conversation,” and “action” get defined across your entire AI stack. I’d revise the urgency here if vendors started publishing standardized outcome taxonomies, but until they do, the definitional language in your contract is the only protection you have.
Based on reporting from What buying party balloons taught me about buying agentic AI, originally published 2026-11-17 17:10:00.

