Results as a Service: What It Is and How to Buy AI by the Result

A price for finished work

Results as a Service, sometimes shortened to RaaS, is a way of buying AI in which you pay a fixed price for each piece of finished work that passes an agreed check. If an item fails the check, or someone on your team has to finish it, that item costs nothing. Most business software is still sold per seat or per unit of usage, such as credits or tokens, and neither of those bills goes down when the work goes wrong.

Customer service is where the model first turned into a published price. Zendesk announced outcome-based pricing for its AI agents in August 2024, and several support vendors have followed since. Our plain guide to what Results as a Service is covers the mechanics. This post is about where the idea came from and how a buyer can tell a genuine per-result offer from usage pricing under a new name.

Where the idea came from

The investor case appeared in April 2024, when Joanne Chen and Jaya Gupta of Foundation Capital published "AI leads a service as software paradigm shift". They sized the opportunity at $4.6 trillion, roughly half in salaries and half in outsourced IT and business process services. Their central point was that "responsibility for achieving the desired outcome sits with the company selling the service."

Sequoia used similar language six months later. In "Generative AI's Act o1", Sonya Huang and Pat Grady wrote that "the AI transition is service-as-a-software" and noted that "Sierra gets paid per resolution." In March 2026 Sequoia's Julien Bek returned to the theme in "Services: The New Software", contrasting a copilot that sells the tool with an autopilot that "sells the work." He estimated that companies spend about $6 on services for every $1 they spend on software.

The current meaning of the phrase owes most to an operator. In March 2025 Dharmesh Shah, co-founder and CTO of HubSpot, wrote "From SaaS to WaaS to RaaS" in his Connecting Dots newsletter. He argued that Results as a Service needs "some objective, measurable result that the customer is paying for," and that it works only in limited cases today, with support ticket resolution as his example. For subjective work he proposed a separate label, Work as a Service.

Nobody owns the phrase

Shah was not the first to use the words. SAS launched an analytics offering called Results as a Service in April 2017, in which its experts delivered answers from the client's own data, and a CIO.com column discussed it that May. When we searched the phrase in October 2026, those 2017 pages still sat near the top of the results.

The phrase has kept spreading since. Bairong, a Hong Kong-listed company, described its business model as Results as a Service in a paid press release in December 2025. In the narrower category of outcome-based pricing, Zendesk's 2024 release calls the company the first in customer service to offer it, while Fin's site says Fin was "first to introduce outcome based pricing". Both are vendor claims, and since nobody owns the phrase, the label on a proposal tells a buyer very little about the billing terms underneath it.

Why the short form causes trouble

Search for the letters RaaS on their own and the results have nothing to do with buying AI. In our October 2026 check, much of the first page came from security vendors explaining ransomware as a service. Most of the rest covered robot as a service and recovery as a service.

The label stays loose even when it does refer to AI. Bairong's release describes a "position-based" pricing mode under the same banner, which works like a salary paid for an AI worker, and a salary sits at the far end from paying per result. In a contract or an RFP I would spell the phrase out every time and then define the billable unit in a sentence of its own.

How to tell a real per-result offer from usage pricing

Usage pricing under a new name is easy to spot once you look at what the invoice counts, and Salesforce shows the difference on one page. Its Agentforce pricing sells Flex Credits at $500 per 100,000, and a standard action uses 20 credits, which works out to $0.10 per action whether or not the action helped anyone. The same page lists the Help Agent, announced in June 2026, at $2 per resolution, and a resolution counts only when "the user's question doesn't need a human escalation".

When a proposal says it charges per result, these are the questions I would ask before looking at the price. Most of them have a published answer from at least one vendor to compare against.

• Does each invoice line count a finished unit of work, or activity such as actions or credits? If it counts activity, the bill rises when the agent struggles.
• Is the result defined in writing before you sign? Zendesk publishes its rule: only conversations that were not escalated to a person count, and an LLM then evaluates each transcript.
• What happens to failed work? At Zendesk, failed verifications "do not consume an automated resolution", so a conversation the check rejects costs nothing.
• Do handoffs cost anything? Fin's pricing page says you are not charged when a conversation is "simply passed to your team without an outcome".
• Is a result credited back if it later proves wrong? Fin deducts a resolution when the customer returns "seeking further assistance (even across billing periods)".
• Where does model usage show up? McKinsey's Lari Hämäläinen told Fortune that the same agent task can cost up to 30 times more from one run to another, and some vendors fold that variance into the price per result while others pass it through.

If the invoice counts activity and failed work still costs money, the offer is usage pricing with a new label, whatever the proposal calls it. Our AI agent pricing comparison puts the published per-result prices next to usage-priced products.

Where per-result pricing does not fit

None of this means every AI purchase should be bought by the result. Shah's own test was that the result has to be objective and measurable, and plenty of valuable work fails it. A strategy memo or a design review is a judgment call, and arguing every month over whether each one counted would cost more than the work itself.

Per-result pricing is also uncommon. In Kyle Poyar's 2025 State of B2B Monetization survey of 240 companies, 5% said their primary pricing model was outcome-based, and HFS Research found in July 2026 that 76% of customer experience buyers were open to outcome or gainshare pricing while only 5% had such a contract. A per-result price usually carries a premium, because the vendor absorbs the cost of failures that it is not paid for. Bessemer's pricing playbook lists cost variability among the drawbacks vendors take on, and if your volume is steady and the work is easy for the model, plain usage pricing may cost you less over a year.

How we sell it at QueryNow

We did not coin the phrase, and we use it for one specific contract. The build of one agent workflow costs $10,000, paid only after the agent 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 because the agent runs in your own cloud tenant, model usage and hosting appear on your own cloud bill. Our Results as a Service page has the full terms.