AI Agent Outcome-Based Pricing: $1.50 to $2.00 per Resolution in 2026

Outcome-based pricing in AI service desk software is a meter that charges per successful resolution by the AI agent, with no human involvement. Zendesk popularised the model when it launched its AI Agent product on a per-resolution rate of $1.50 with an annual commitment, or $2.00 on a pure pay-as-you-go basis. The contractual logic is that buyers are paying for a unit of value delivered (a closed ticket) rather than a unit of capacity provisioned (a seat or a chat session).

The model has spread quickly into the broader CX category. Intercom Fin charges a similar per-resolution rate. Forethought charges per-deflection. Even vendors who launched on per-seat models are introducing outcome-based options as a procurement-friendly alternative. By April 2026 the per-resolution rate has become the default new-deal structure for CX-leaning AI service desk software, particularly for mid-market buyers under 2,000 seats.

What buyers actually get for $1.50 varies. A resolution is typically a customer or employee interaction that the AI handled end-to-end without a human agent taking ownership of the ticket. If the user re-opens the conversation within a defined window, or opens a related new ticket, or escalates explicitly, the resolution either does not count or is clawed back from the next bill. The exact mechanic lives in the master services agreement and is the single most important contract clause buyers should review before signing.

Compare this to the older per-seat AI uplift model used by ServiceNow Now Assist. A 5,000-seat organisation on ServiceNow Pro Plus pays roughly $175 per user per month for the base platform, plus a Now Assist uplift typically in the 25 to 60 percent range. That is $44 to $105 per user per month additional, or $264,000 to $630,000 additional per year, irrespective of how many tickets the AI actually resolves. If your org runs 10,000 tickets a year, you are paying ten times more per resolution than the Zendesk rate. If you run 500,000 tickets a year, the per-seat model is dramatically cheaper per resolution. The pricing model only makes sense in the context of your ticket volume.

Every outcome-based contract turns on a single sentence: what counts as a resolution. Zendesk defines it as an automated interaction that closes a ticket with no human agent ownership and no related re-open within a defined window. Intercom Fin uses a similar formulation. The problem for buyers is that the definitions sit inside the master services agreement, the windows are not always disclosed in marketing material, and the audit mechanism for verifying the count is rarely automated.

Three traps recur in procurement reviews. First, the re-open window. If a ticket re-opens after seven days, does the original resolution still count? Many contracts say yes. That undercounts re-engagement rates and overstates net deflection. Second, the related-ticket definition. If the user opens a new ticket on the same root cause, is that a resolution claw-back? Most contracts treat it as a separate ticket. Third, the partial-resolution accounting. If the AI provides 80 percent of an answer and the user finishes with a search-engine query, is that a resolution? Vendor accounting usually says yes.

The discipline buyers need is to negotiate a contractual reconciliation right. Specifically, the right to audit a representative bill period against ticket data and adjust the next quarter's billing. Vendors selling on outcome-based models in 2026 are generally willing to grant this when asked. They are less willing to grant it when not asked. Make it a procurement requirement before signing.

Pair the contractual safeguards with quality KPIs. Customer or employee satisfaction on AI-resolved interactions should be tracked and surfaced; a healthy deployment is within five points of human-agent CSAT. Re-open rates within 72 hours should be under 10 percent. First-contact resolution should improve, not regress. These numbers belong in the vendor's quarterly business review deck, and they belong in the contract as service-level targets with consequences for breach.

Outcome-based pricing optimises for the buyer who is uncertain about volume or maturity. The model breaks down in three buyer profiles where per-seat economics return as the better deal.

The first is the very-large enterprise with high ticket throughput. A 15,000-employee organisation processing 250,000 internal IT tickets per year and achieving 55 percent deflection produces 137,500 paid resolutions. At the Zendesk committed rate that is $206,250 per year, but enterprise contracts at that volume rarely stay on the rate card; the negotiated outcome rate typically sits in the $0.80 to $1.20 range. The per-seat ServiceNow Now Assist alternative on Pro Plus might land at $700,000 to $1.2 million per year, but the AI uplift is bundled with deep platform capabilities the buyer is already paying for. The marginal cost of AI deflection on top of platform spend can be lower than the outcome-based deal once volume and discounting compound.

The second is the organisation that has built deflection into its core operating model and refuses to outsource the meter. Per-seat or annual-contract pricing protects buyers from over-paying when deflection rates exceed expectations. If your org gets to 70 percent deflection across 100,000 tickets, a per-seat contract caps your spend; an outcome-based contract scales linearly with success. Sophisticated procurement teams will negotiate caps and floors into outcome-based contracts to manage this exposure.

The third is the organisation with deep ITSM platform integration where the AI layer is a feature, not the product. ServiceNow Now Assist users get auto-summarise on every incident record, draft resolution notes inside the agent workspace, change-risk prediction, flow generation, and Virtual Agent self-service all bundled into the seat. Pricing per-resolution would only count the Virtual Agent deflections. The platform-embedded AI work delivers measured value that is not in the deflection metric. Per-seat captures that wider value envelope, outcome-based does not.

The honest decision rule: under 40,000 paid resolutions per year, outcome-based wins. Above 100,000, run the negotiation both ways and let the math decide. Between 40,000 and 100,000, the answer is almost always to start outcome-based and renegotiate at year two when the deflection rate has stabilised.

Atlassian Intelligence in Jira Service Management uses a hybrid meter that is worth understanding because it is neither true outcome-based nor true per-seat. Premium and Enterprise customers receive 1,000 assisted conversations per month included. Above that allowance, conversations are metered at $0.30 each. A conversation counts whether or not it leads to a resolution.

The hybrid is procurement-friendly. The included allowance is generous enough that small and mid-market Jira shops will rarely hit it. The overage rate is low enough that exceeding the allowance is not a budget event. The trade-off is that the meter does not align cleanly with deflection: a Virtual Service Agent conversation that fails to deflect still costs the buyer 30 cents above the allowance. Atlassian's implicit pricing argument is that the agent is paid to engage, and the buyer is paid in deflection over time.

For Jira-native organisations, this is usually the cheapest path to AI service desk capability because the underlying JSM Premium spend is already in budget. See the Atlassian Intelligence comparison for the full meter mechanics and how the included allowance scales with Premium versus Enterprise seats.

Buyers entering an outcome-based AI service desk contract in 2026 should run a five-step procurement playbook. None of these steps are exotic; they are simply not consistently applied because the pricing model is new enough that purchasing departments still treat it like a SaaS seat-licence contract.

Step one. Define the resolution metric in your own terms before the vendor defines it. A defensible internal definition is: an interaction that closes a ticket without human agent ownership, with no related re-open or new ticket on the same root cause within 14 days, where the user has not subsequently opened a duplicate or related issue. Bring this definition to the negotiation as a counter-proposal to the vendor's default contract language.

Step two. Negotiate the audit and reconciliation right into the MSA. The vendor should provide monthly granular usage data including ticket IDs, classification confidence scores, and an exportable resolution log. The buyer should have a contractual right to audit any month, identify miscounted resolutions, and adjust the next billing cycle. This protects against silent over-billing as volume grows.

Step three. Set quality service-level targets with consequences. AI-resolved CSAT within five points of human-agent CSAT, re-open rate under 10 percent within 72 hours, first-contact resolution improved by at least three percentage points within six months of go-live. Breach should reduce the per-resolution rate by 10 to 20 percent until remediated. Vendors confident in their accuracy will agree to these clauses.

Step four. Cap and floor the contract. Negotiate a soft cap at 150 percent of forecast resolution volume, with the vendor agreeing to discount further if you exceed it. Negotiate a soft floor at 50 percent of forecast, with the vendor agreeing to credit if you fall short. This insulates both parties from forecast risk while keeping the outcome-based logic intact.

Step five. Run the model both ways. Get a per-seat or annual-contract quote from the same vendor or an equivalent vendor. Build a three-year TCO comparison at three deflection rates (30, 50, 70 percent) and three volume scenarios (forecast, 80 percent of forecast, 130 percent of forecast). Use the resulting nine-cell grid to identify the price point where the alternative model becomes cheaper. That price point is your trigger for renegotiation at year two.