Your Next AI Vendor Wants to Bill You Per Outcome. Define the Outcome First.

Enterprise software is quietly changing its unit of sale. For two decades the meter was a seat. For the past two years it has often been a token. This week brought fresh evidence that a growing group of vendors wants to bill for something else entirely: a finished piece of work.

CIO Dive reported on Monday that outcome-based billing has moved far enough into enterprise contracts that IT leaders are rebuilding vendor management around it — and that vendors expect to be paid more, not less, as customers hand work over to AI agents (Patrick Thibodeau, CIO Dive, August 31, 2026).

That reporting is about CIOs at universities, insurers, and hardware companies. But dealership technology is bought from the same software market, and the pricing convention that wins there tends to arrive in auto retail a year or two later — in phone AI, CRM, service scheduling, and BDC tooling. When it does, the hard part of the negotiation will not be the rate. It will be the definition.

Myth vs. data. The myth is that outcome pricing has already taken over and that it shifts risk to the vendor by default. Gartner's numbers say otherwise. Only 19% of services buyers and 13% of service agreements on the seller side use outcome-based arrangements today, and Gartner projects that fewer than 25% of tech services contracts will use them through 2031. Gartner VP analyst Tom Coshow's read: the increase in outcome-based pricing is "more buzz than reality." (Gartner, via CIO Dive, August 2026)

What the vendors are actually metering

The category is real, but it is not one thing. Vendors are choosing very different billable units, and each choice moves risk to a different party.

Sources: CIO Dive, August 31, 2026; Salesforce Flex Credits pricing announcement, May 2025.

Notice that only two of those four are outcome pricing in the strict sense. Salesforce meters actions, which is closer to a token model wearing better clothes: an agent that takes six actions to fail still bills six actions. HP's version keeps the seat and bolts a promise onto it. Faisal Masud, who leads HP's platform effort, told CIO Dive that most early adopters will not see those options until mid-to-late 2027.

The backdrop is a repricing of the whole category. Gartner estimated in July that up to $234 billion of enterprise application software spend is exposed to agentic AI between now and 2030, and that price adjustments will account for roughly 20% of enterprise SaaS spending by then (Gartner, July 2026, via CIO Dive).

The 90/10 clause is the entire negotiation

The most instructive detail in the CIO Dive piece is not a price. It is a counting rule.

Zendesk president and CRO Chris Donato described how his company decides what it can charge for: "If the AI resolves 90% of the problem, but 10% goes to a human agent, we don't count it." Automation has to resolve, in his words, 100% of that interaction.

That is a genuinely buyer-friendly rule, and it is worth understanding why a vendor can afford to offer it. A support ticket usually has one job in it. Either the refund got processed or it did not. The unit is clean, so a strict test is safe to write.

A phone call to a dealership is rarely that tidy.

Consider one inbound call on a Saturday morning. The caller asks whether a specific used unit is still on the lot, asks what the payment might look like, and books a Monday test drive. An AI agent that confirms the vehicle and books the appointment but hands the payment question to a salesperson has done most of the commercially useful work. Under a strict end-to-end rule, it has done none of it.

That is not an argument against the rule. It is an argument for writing your own, in your own contract, before a vendor writes one for you.

What counts as one outcome at a store

There are four plausible units for dealership AI, and they are not interchangeable.

• A booked service appointment. The cleanest unit in the building, because the DMS already records it. It is also the one most likely to be double-counted if a customer books online and then calls to confirm.
• A service appointment that actually shows. Better aligned with revenue, worse for vendor cash flow, and dependent on no-show behavior the vendor does not control. Expect to pay a premium for this definition, or to be told no.
• A contained call. Any inbound call fully handled without a human. Easy to measure, but it rewards deflection rather than revenue — a store can hit a great containment number while quietly losing sales calls.
• A qualified sales lead delivered to a person. Closest to how dealers already think about marketing spend, and the hardest to define, because "qualified" is exactly the word both sides will fight about.

None of these is the right answer for every store. The point is that the answer belongs to you, and it should be in the agreement rather than in a dashboard the vendor configures after go-live.

Three lines that decide your invoice

The completion test. Write down what has to be true for a unit to be billable, including what happens when a human touches the interaction. Borrow the Zendesk standard if it fits, or define a partial-credit tier — but define it.

The attribution window. If a customer speaks to AI on Tuesday and books through the website on Thursday, who gets paid? Marketing has fought this war for fifteen years. AI vendors are walking into it fresh.

The failure clause. This is the one most likely to be missing. Coshow's test is the sharpest sentence in the CIO Dive story and belongs on the wall of every dealer's tech review: "If the vendor isn't taking on the risk, why are you bothering with outcome-based pricing?"

Price it against numbers you already have

Dealers are unusually well equipped for this negotiation, because franchised stores already benchmark cost per outcome in another department. NADA's most recent data gives the reference points.

Source: NADA data reported by Inside Radio, May 2026.

Two things fall out of that table. First, the average store already pays roughly $48,900 a month for advertising, and accepts an implied $739 cost per new unit sold without a completion test of any kind. A per-outcome AI contract with a written definition is, on that comparison, a more accountable purchase than most of the ad budget.

Second, those 276 million repair orders across 16,990 franchised dealerships work out to roughly 16,000 repair orders per store per year, or about 1,350 a month (Carbuki calculation from NADA's published totals). That is the denominator to bring to any per-appointment quote. A vendor's price per booked appointment is meaningless until it sits next to your own appointment volume, your show rate, and your current cost per appointment across the BDC payroll and the ad line.

The part buyers tend to underrate

Paul Fisher, CIO at Seton Hall University, gave CIO Dive the most balanced assessment in the piece. His institution already pays an outsourced help desk per call or chat, and he expects an outcome model would add work: it "adds complication to contract negotiation as you need to be very specific on what the desired outcomes are." He also thinks it may be worth it, because it "incentivizes the partner to 'get it right.'"

Carmen Li, CEO of Silicon Data, framed the near-term benefit more narrowly. So far, she told CIO Dive, the influence of outcome pricing shows up in transparency and planning rather than in cheaper AI.

For a dealer principal, that is the realistic expectation to hold. Outcome pricing is not a discount. It is a forecasting tool and an accountability mechanism, and it costs a harder negotiation up front to get either one.

The measured read

Nothing here requires a decision this quarter. Gartner's adoption numbers are low, the projected ceiling through 2031 is under 25%, and HP expects most buyers will not see mature outcome options until late 2027. Any vendor telling a dealer that outcome pricing is now the industry standard is ahead of the evidence.

What is worth doing this quarter is smaller. Instrument the outcomes you would want to buy. You cannot negotiate a price per booked appointment without knowing your current appointment volume, your show rate, your after-hours call volume, and what a booked appointment costs you today across payroll and advertising. Stores that walk into a 2027 renewal with those four numbers will negotiate from a different position than stores that walk in with a demo impression.

The vendors are busy defining the billable unit. The dealers who do that work first will be the ones who get to argue about price instead.

If you are evaluating AI phone coverage and want to compare it against your own call, appointment, and show-rate data rather than a dashboard, that is the conversation we prefer to have. You can find us at carbuki.com.

Sources

• Patrick Thibodeau, "Agentic AI is shifting the pricing models CIOs rely on," CIO Dive, August 31, 2026 — ciodive.com
• Paige Gross, "Agentic AI to disrupt $234B in SaaS spending: Gartner," CIO Dive, July 6, 2026 — ciodive.com
• Salesforce, "Salesforce Introduces New Flexible Agentforce Pricing," May 15, 2025 — salesforce.com
• NADA advertising and dealership data, reported by Inside Radio, May 2026 — insideradio.com