Token pricing is a cost-allocation mechanism. Resolution as a Service (RaaS) is the pricing and architectural model in which enterprise software is priced on problems solved rather than the number of users who log in. These are not two versions of the same idea. They solve different problems for different parties, and confusing them is now the most expensive mistake a seat-based vendor can make on the path to the transition.
The market is converging on a consensus that consumption-based AI pricing is the mature successor to seats. The consensus is half right. Consumption pricing does solve the seat problem. It does not solve the outcome problem. A vendor who bills per token or per API call has moved the unit of commerce from headcount to compute. They have not moved it to value delivered. The customer is still absorbing all the utilization risk. The vendor is still paid whether the problem was resolved or not.
That distinction is not semantic. It is the dividing line between a pricing model that holds through renewal and one that does not.
The Asymmetry That Token Pricing Preserves
Under seat-based pricing, the vendor wins when the customer hires. Under token-based pricing, the vendor wins when the AI runs. Under Resolution as a Service, the vendor wins only when the customer’s problem is solved.
The first two models share a structural feature: the vendor’s revenue is decoupled from the customer’s outcome. Token billing makes this decoupling more visible, not less, because it surfaces exactly what the vendor is actually selling: compute capacity. It is access pricing with a more sophisticated meter.
This matters at renewal in a way it did not three years ago. Enterprise buyers are entering 2026 renewal cycles equipped with AI-generated usage analytics that their predecessors did not have. A procurement team that could not previously quantify what the platform did can now run its own resolution-quality analysis. Enterprise AI pilots often look economically manageable before they reach production. Once they scale, organizations frequently discover that integration, orchestration, governance, and operating costs were materially underestimated. That is not a complaint about token pricing being expensive. That is a complaint about token pricing being opaque on value.
When the procurement team asks what they actually got for that compute spend, a token count is not an answer.
What the Token Bill Cannot Prove
The core commercial obligation a vendor takes on when it moves to outcome-based language, even informally in sales conversations, is attributability. The claim is: our platform resolved your problem. To defend that claim at renewal requires three things, which together define what Crown Point Advisory Group calls an Atomic Resolution.
The resolved outcome must be verifiable, meaning a problem was solved, not merely attempted. A ticket marked closed is not a resolution if the issue recurs in 48 hours. The resolution must be attributable, meaning the platform’s AI execution can be demonstrated to have caused the result, distinguishing its contribution from a concurrent manual process or an ambient change in the customer’s environment. And the resolution must be finite, with a defined start and endpoint that prevents open-ended agentic loops from generating unbounded compute cost under what the customer understood to be a fixed rate.
Token billing satisfies none of these three criteria. A high token count can mean a complex problem was resolved effectively. It can equally mean the model hallucinated across fifteen attempts before producing a usable output, or that an agent entered a reasoning loop and consumed compute without completing any meaningful task. The token bill records the compute. It does not record the outcome.
Under consumption billing, the vendor is paid for every inference it runs, including failed attempts. Under resolution billing, the vendor is paid only for successful resolutions. That difference in exposure is not incidental. A vendor with a 65 percent resolution rate and 1.5 average inference attempts per resolution incurs a failure-mode cost equal to approximately half the direct inference cost for every resolution it bills. Token pricing transfers that cost to the buyer. Resolution pricing holds the vendor accountable for it.
Token pricing protects the vendor from delivery uncertainty. Resolution as a Service requires the vendor to absorb it. That accountability gap is what outcome alignment actually means.
The Results-as-a-Service Confusion
The market has generated a competing framing worth naming precisely because it is close enough to the right answer to be dangerous. Results as a Service frames the AI pricing transition as a billing problem. Change the invoice structure from seats to outcomes, and the transition is complete.
Resolution as a Service is not a billing model change. It is a business transformation architecture. The distinction is the argument.
A vendor who contracts to deliver results without the underlying architecture to prove what was resolved is exposed at exactly one moment: renewal. The customer contracted for accountability. The invoice arrives without evidence. The procurement team, now equipped with its own AI analytics, does not renew at the same rate.
The architectural requirement that Results as a Service skips is the High-Fidelity Repository: a graph-structured institutional knowledge architecture that gives AI agents the domain context, audit trails, and attribution records needed to execute and prove resolutions at scale. Without a High-Fidelity Repository, outcome-based billing is an assertion. With it, it is a documented record. The Repository is also the primary competitive moat in the RaaS era, because every resolution executed against a mature Repository enriches it in ways a general-purpose AI agent cannot replicate.
Most vendors attempting the transition discover this gap at the wrong moment. The Vendor Transition Playbook identifies skipping Phase 2 of the Three-Phase RaaS Transition Roadmap as the most common failure mode. Phase 2 is the Hybrid Pilot, months 6 through 18, in which the platform is instrumented to track resolutions per customer and cost-to-serve per resolution type against existing seat contracts, before a single customer has been moved to outcome pricing. Vendors who go directly from recognizing the problem to repricing contracts have no measurement infrastructure to know whether they are pricing above or below their actual cost to serve. The result is a second wave of margin compression, worse than the first, because it occurs on a rising, unmeasured cost base.
The Economic Frame That Makes Accountability Viable
Resolution as a Service requires a specific economic discipline to be commercially sustainable, and that discipline is what gives outcome alignment its teeth.
Resolution Contribution Margin (RCM) is the financial discipline that makes outcome pricing defensible. RCM is the margin remaining after subtracting all costs directly attributable to delivering a single Atomic Resolution, including AI compute, data retrieval, human oversight where applicable, and verification overhead. It is not a fixed ratio applied to compute cost. It is a per-resolution margin analysis: each resolution class must be priced such that its RCM supports the gross margin baseline institutional investors require.
The execution risk is concrete. Intercom’s Fin provides a useful directional example. Publicly discussed resolution-based pricing shows how vendors can begin moving away from seats, but CPAG estimates suggest that early resolution pricing can run ahead of resolution-level margin discipline if cost-to-serve is not modeled carefully. Margins improved as inference costs declined, but the pricing structure required ongoing adjustment. Without RCM discipline embedded in pricing from the start, early-mover advantage compounds losses rather than accelerates them.
Token billing makes RCM analysis structurally impossible. A token count tells you aggregate inference cost. It does not tell you cost-to-serve per resolution class, which customer segments are generating negative margins at volume, or where pricing needs adjustment before a renewal conversation forces the issue.
The Sequence That Separates Survivors from Phase-Skippers
The vendors navigating this transition successfully share one characteristic: they built measurement infrastructure before they changed the invoice. ServiceNow is a useful example of a vendor protecting seat-based predictability with usage-based metering rather than moving directly to pure outcome pricing. HubSpot has begun moving parts of its AI monetization away from pure seat-based pricing, including outcome-based pricing for Breeze Customer Agent and AI credit structures. That is directionally important, but it is not yet enough by itself to prove full RaaS economics.
Enterprise buyers are increasingly receptive to outcome-based pricing, but only when the vendor can provide evidence that the outcome was achieved and attributable.
The practical sequence runs through Phase 1 of the Three-Phase RaaS Transition Roadmap: a Revenue Audit that maps ARR exposure to seat risk and calculates the Ghost Seat Rate for the top accounts. Ghost Seats are licensed seats paid for but no longer actively used because AI has absorbed the workflow. A Ghost Seat Rate above 15 percent in any major account is a compressed transition timeline, regardless of how the current billing relationship is denominated. Knowing that number before the renewal conversation is the difference between negotiating from position and negotiating in panic.
The question of what architectural and governance conditions are required before AI accountability claims can actually be enforced, and what happens institutionally when they are not, is examined in the current series at middlewayinai.com. The commercial framing here is the operational surface of a deeper institutional problem.
Prescription
The immediate intervention is a measurement audit, not a pricing audit. Before any conversation about repricing customers to an outcome-based model, answer three questions per your top ten accounts by ARR. First: can your platform produce a verifiable record of every resolution it delivered in the last 90 days, including which were completed autonomously and which required human intervention? Second: for each resolution type you intend to bill on, can you pass all three Atomic Resolution tests, verifiable, attributable, and finite? Third: do you know your cost-to-serve per resolution class, including AI compute, data retrieval, human oversight, and verification overhead, with enough precision to confirm each resolution type is priced to a positive Resolution Contribution Margin?
If the answer to any of these is no, the billing model is not the problem. The measurement infrastructure is. Token billing is not a temporary waystation toward outcome alignment. It is a structural obstacle to it, because it produces a cost record where you need an outcome record.
The full build sequence is in the RaaS Vendor Transition Playbook.
If your largest enterprise customer’s procurement team requested a verifiable record of every resolution your platform delivered last quarter, including attribution evidence and quality gate outcomes, what would you send them?