Skip to content
AI-accelerated delivery · You pay when it works
Plano, TX · MunichAccepting Q4 2026 briefs
Blog/AI
October 18, 20267 min read

Outcome-Based Pricing for AI Agents: What Vendors Charge per Result in 2026

Six vendors publish a price per AI result, nearly all of them in customer service. The prices only compare once you line up what each vendor counts as a result.

A row of blue and teal price tags, each hanging from a small chat bubble with a check mark, on a deep navy background

Six published prices

Most AI agent vendors still sell by the seat or by usage, but a small group now publishes outcome-based pricing for AI agents, with a fixed price for each result. I checked the pages below on October 18, 2026. Five of the six are in customer service, and each vendor counts a result differently, so the prices only compare once the definitions sit next to them.

What each vendor counts, and what it leaves out

Fin's definition is the widest of the support group. A resolution counts when the customer confirms it or "does not request more help after Fin answers", and Fin's help article adds a second billable outcome: a Procedure you configured to end in a handoff to a person or workflow. A customer who "explicitly asks to speak with a human agent at any point" is not billed. If the customer comes back for more help, Fin's measurement guide says the resolution is deducted, even across billing periods.

Zendesk applies a stricter test. It considers only conversations that were not escalated to a person, and an LLM then evaluates each transcript. Since a change that rolled out in May and June 2026, it has billed only resolutions marked "Verified". Failed verifications "do not consume an automated resolution", and Zendesk's help article says an email conversation ends 72 hours after the first email.

Gorgias uses a time window from the other side. Its AI Agent bills an interaction that was "fully resolved without a human agent", but if a person on your team replies within 72 hours, the conversation is billed as an ordinary helpdesk ticket.

Salesforce's Help Agent counts a resolution when "the user's question doesn't need a human escalation" and treats actions within a 10-minute window on a call as one resolution. Escalated or abandoned conversations are not billed, and neither is one where the customer says they did not get an answer. Salesforce's Kishnan Chetan told CIO.com, "that's not a resolution in our books."

HubSpot's announcement does not define a resolved conversation at all, although Jon Dick, HubSpot's CCO, summed up the intent as "You pay when it works, full stop." Chargeflow's definition is the plainest of the six: a recovered chargeback is "A dispute we fight and win on your behalf", and the 25% fee excludes the dispute fee itself.

Lining up the numbers fairly

On the list above, HubSpot's $0.50 looks like a quarter of Salesforce's $2. The comparison holds only if the two units mean the same thing, and on the published terms they may not, since one has no public definition and the other excludes any conversation that needed a person. A cheaper unit that counts more conversations can produce a larger bill than a dearer unit that counts fewer.

Minimums change the arithmetic at low volume. Fin's standalone plan carries a minimum of 50 outcomes a month, included in a $49 base plan, and unused outcomes are lost at the end of the month. Salesforce's Help Agent packs start at 1,000 resolutions, which is $2,000 paid in advance, while Zendesk includes five automated resolutions per agent each month on its Support Team and Suite Team plans and ten on Suite Professional. Our AI agent pricing comparison keeps these figures in one table next to the usage-priced products.

How rare pure outcome pricing still is

These vendors are the exception. Orb, which sells billing software, looked at 80 AI agent companies and found 3.8% using outcome-based pricing, against 4.5% in 2025, though it says little about its method. Kyle Poyar's 2025 State of B2B Monetization survey put the share of companies whose primary model is outcome-based at 5%, with 25% expecting it to be by 2028.

Buyers say they want it more often than they sign it. In July 2026 HFS Research found 76% of customer experience buyers open to outcome or gainshare pricing and 5% with such a contract. For AI billed separately from core software, Futurum's 2H 2026 survey found per-seat pricing was still the top preference at 42.3%, with outcome-based pricing at 21.1%. My reading is that finance teams want a ceiling on AI spend before they want anything else.

Several of the best-known vendors do not publish a price at all. Sierra, which set out its approach in a December 2024 post, agrees the outcome criteria with each customer. Decagon offers per-resolution pricing at "a higher fixed rate", but its pricing post says "The vast majority of our customers choose per-conversation pricing." Ada uses resolution pricing for some enterprise customers, and neither ServiceNow nor Crescendo lists its rates.

Outside customer service, prices are negotiated

A few published per-result prices reach into sales. HubSpot's Prospecting Agent costs $1.00 for each lead it recommends for outreach, and Fin charges $9.99 for a qualification outcome. Beyond that the public rate cards run out.

Document work has almost none. EvenUp, which serves personal injury law firms, announced case-based pricing in May 2025 without publishing a rate. In our experience, AI work on invoices or claims is priced inside each contract, so a buyer has no list price to compare against and has to start from what the work costs today.

That cost varies more than most people expect. Ardent Partners put the average all-in cost of processing one invoice at $9.40 in 2025, and its 2026 study found $2.67 for best-in-class teams against $12.50 for everyone else. With a spread that wide, a price per result only makes sense once you set it against your own cost per invoice.

Prices will move

One caveat applies to every number in this post: none of them has been stable for long. HubSpot cut its price from $1.00 per conversation to $0.50 per resolved conversation in April 2026, and Zendesk narrowed what it bills about a month later. Fin renamed its unit from resolution to outcome and widened it to include configured handoffs. Salesforce charged $2 per conversation in 2024 and added Flex Credits in May 2025, and its Help Agent is the first per-resolution price on the Agentforce pricing page.

The costs underneath could push prices up. Gartner forecasts that the generative AI cost per resolution in customer service will exceed $3 by 2030, more than many offshore human agents in B2C, and expects inference costs per agentic workflow to rise more than fivefold through 2028. If you sign a per-result contract, fix the unit price for the term and write down how it can change at renewal, because this year's list price may not be the one you renew at.

Where QueryNow fits

Outside customer service there is no rate card to copy, so we set the price from each client's own numbers. The build of one agent workflow costs $10,000, paid only after it passes acceptance criteria you signed.

After launch you pay a fixed price for each result that passes its checks, set from the acceptance test numbers, with a $1,200 monthly minimum. Work that fails a check or that your team finishes is not billed, and since the agent runs in your own cloud tenant, model usage appears on your own cloud bill, outside the price per result. Our Rockwell Automation work and a Generac assessment ran on results-based terms, and the details are on our pay when it works and Results as a Service pages.

Take action

Ready to ship AI in your organization?

We build one workflow into a working tool in two weeks. You pay $10,000 only after every acceptance criterion you signed off on is met.

One workflow · Two-week build · $10,000, paid on delivery

Q

QueryNow

QueryNow deploys production AI for enterprises on Azure, AWS, or Google Cloud. Founded in 2014, we help pharma, healthcare, manufacturing, and financial services organizations deploy governed AI systems. We build it, you pay when it works.

Learn more about us →

Share this article

LinkedIn →
Tell us the workflow →
Take the next step

Turn these insights into real results

Point at the workflow your team hates. We build the tool that kills it in two weeks, and you pay only when it works.

The two-week build

We scope one workflow with you and sign an agreement on the acceptance criteria. We build the tool in your environment in two weeks. You see it work before you pay.

  • +A fixed scope and acceptance criteria, signed on day one
  • +A working tool, built in your environment
  • +Automated evaluation against your own data
  • +You pay $10,000 only after every criterion is met
$10,000

One workflow tool. Paid on delivery.

One workflow at a time. $10,000 per build, due only after it meets the criteria you signed.

The Monday digest

The three posts worth your time each week, one paragraph each. From the people building production AI.

Keep reading

Related articles

More from AI