AI Pricing in 2027: Buyers Will Pay Only for Finished, Verifiable Work
By 2027, AI vendors led by Salesforce, Zendesk, Intercom and Sierra will bill for narrow, verifiable completed milestones — with invoice-level traceability and hybrid pricing still attached.
By Nathan Brooks
4 min read

What's News
- Sierra described its outcome-pricing model in 2024; Zendesk introduced automated resolution tiers in May 2026.
- In 2027 vendors are forecast to add billable outcomes in sales qualification, IT service and back-office reconciliation.
- Intercom's March 2026 change makes configured Procedure handoffs billable while routine human escalations are not.
- METR's February 2026 update complicated its earlier finding that experienced developers worked slower with AI.
- Hybrid pricing — outcomes plus seats, platform and consumption fees — is forecast to persist through end of 2027.
By 2027, AI vendors will bill for completed milestones rather than interactions — and the milestones will be narrow enough to count, dispute and trace back to a single invoice line. That is the core forecast from Unite.AI's analysis of outcome pricing, which tracks Fin/Intercom, Salesforce, Zendesk and Sierra as the cohort defining how enterprises will buy agentic software.
Outcome pricing is not new. Sierra described its model in 2024. Salesforce sells Agentforce through several units — actions, conversations, licenses and help-agent resolutions. Zendesk introduced automated resolution tiers in May 2026. These are materially different units, even when each vendor pitches the same promise: aligning spend with value.
What does the buyer actually pay for?
A buyer doesn't wake up wanting more AI interactions, the analysis notes. "They want the service request resolved, the qualified lead handed over or the accounts reconciled." The forecast: in 2027, vendors will add newly billable outcomes in sales qualification, internal IT service and back-office reconciliation — domains with recognizable, disputable milestones.
The product problem is finding a unit small enough to attribute and valuable enough to price. "Improve customer relationships" is too broad to settle a charge; "validate eligibility and route a qualified prospect according to these rules" is billable.
Fin's current documentation shows why definitions matter. It separates resolutions, configured Procedure handoffs, qualifications and disqualifications. A configured handoff can be billable; a routine escalation to a human cannot. So "outcome" doesn't mean "the customer's whole problem was solved." In its March 2026 explanation, Intercom argues an agent can perform useful work before human confirmation — coherent, but buyers must contract for either autonomous resolution or a completed subtask.
Will billing become inspectable?
Sooner or later, a customer opens an AI invoice and asks why a specific case appears on it. The analysis predicts vendors will let buyers inspect individual billable events, see how they were classified and challenge mistakes. The needed artifact: an event record connecting the invoice to the work — which request was handled, what triggered the charge, which policy classified it, what happens if the result is corrected later.
Zendesk's resolution unit covers requests resolved without human escalation; Fin's FAQ counts configured handoffs as outcomes. Comparing headline prices without comparing definitions obscures the purchase. Sierra's 2026 discussion of outcome pricing explicitly acknowledges the operational and contractual complexity of attribution — evidence, the author argues, of unfinished work rather than a settled industry definition.
What about the whole workflow's economics?
The hardest forecast: stronger customer evidence in 2027 will expose full workflow economics, not just headline automation rates. A high automation rate doesn't reveal whether remaining cases got harder; lower cost per conversation can coexist with more conversations per resolved problem.
The evidence base is cautionary. The NBER working-paper version of Generative AI at Work found higher issues resolved per hour among support staff using an assistant, with substantial variation by experience — but that measured humans with assistance, not autonomous agents. METR's early-2025 developer study found a slowdown among experienced developers; its February 2026 update says later measurements became hard to interpret due to selection, task choice and changing work patterns.
The prescription: measure the complete operating loop. The cost numerator includes agent fees, platform expense, human review, escalations and rework; the denominator counts cases meeting a stated completion standard. This aligns with the revised 2026 Measuring Agents in Production study on reliability challenges, and τ²-Bench's reminder that completing a support task can depend on coordinating with a user who changes the environment.
Will outcome fees replace subscriptions?
No — and the analysis is highly confident on this. Hybrid bills will remain common through the end of 2027: outcome charges alongside seats, platform access or consumption. Intercom's pricing already separates helpdesk seats from Fin usage; Salesforce exposes multiple purchasing units; Zendesk combines platform packaging with resolution allowances and extra capacity.
The practical consequence: a cheaper advertised outcome may not mean a cheaper completed case, and a higher unit price may pay for itself by removing downstream work. The forecast would fail only if vendors absorb supporting costs into a single simplified outcome unit to cut procurement friction.
What does this change for practitioners? Procurement, finance and agent operators should agree on a shared definition of completion before scaling, sample billed events, and judge pilots on full operating cost — because the strongest AI products will make the buyer's definition of success visible inside the service, awkward cases and correction queues included.
Source: unite.ai
More from Nathan Brooks
Show full bio
News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.
2,368 articles