"Services as software" means buying finished work, such as closed books, resolved tickets or drafted contracts, from a company that does most of that work with AI, and paying per result instead of per user seat or per billable hour. For buyers, it shifts the question from "which tool do we license?" to "which outcome do we pay for?"

Picture a finance lead renewing two contracts in the same month. One is the accounting software, priced per user. The other is the outside firm that closes the books each month, billed by the hour. A new vendor offers a third option: send us your bank feeds and invoices, and we will deliver a reconciled month-end close for a fixed price per close. There is no seat to license and no hours to approve. The finance lead has to decide whether that is cheaper, riskier or both.

What does "services as software" mean?

Services as software is a business model in which a company sells the work itself, done mostly by AI systems with people checking the parts that need judgement, rather than selling a tool that helps a professional do the work. The customer buys an outcome, not access.

The phrase draws on Services: The New Software, an essay by Julien Bek published by Sequoia Capital on 5 March 2026. It opens with the line "The next $1T company will be a software company masquerading as a services firm," and draws the distinction in two sentences: "A copilot sells the tool. An autopilot sells the work."

• A copilot is AI placed in a professional's hands; the professional is the customer and decides what to do with it.

• An autopilot is AI that delivers the result; the customer is whoever used to pay for the work.

The essay's argument is about budgets. It states that "for every dollar spent on software, six are spent on services," and gives an example: "A company might spend $10K a year for QuickBooks and $120K on an accountant to close the books." An autopilot competes for the second budget, not the first.

Why are vendors moving from seats and hours to outcomes?

Seats and hours both measure inputs, and AI breaks the link between inputs and results: if software does the work, fewer people log in and fewer hours are billed, so a vendor priced on either gets paid less for doing more.

Each model rewards something different:

• Per seat: The buyer pays for: people with access. The vendor is rewarded for: more users, whether or not the work gets done. Breaks when: the software does the work and fewer people need a login.

• Per hour: The buyer pays for: time spent. The vendor is rewarded for: more hours, which can mean slower work. Breaks when: AI does in minutes what used to be billed in days.

• Per outcome: The buyer pays for: a defined, finished result. The vendor is rewarded for: results that meet the definition. Breaks when: the outcome is hard to define, verify or attribute.

An outcome is a unit of finished work that both sides agree on in advance and can check afterwards: a ticket resolved without a person, an invoice matched and posted, a lead qualified against agreed criteria. Outcome pricing only works when that unit can be written down.

What does outcome pricing look like in practice?

Customer service is where outcome pricing has gone furthest, because a resolved conversation is easy to count. Two well-known vendors publish how they do it.

Zendesk: pay per automated resolution

On 28 August 2024 Zendesk announced outcome-based pricing for its AI agents, under which "customers will only incur costs for issues that are resolved autonomously by AI." Its help centre explains that a large language model reviews each conversation when it ends "to confirm that the customer's request was satisfactorily resolved", and only verified resolutions count. Zendesk added resolution tiers, priced by how complex the resolution is, on 18 May 2026.

Intercom Fin: pay per outcome type

Intercom's Fin pricing page (checked 31 October 2026) prices Fin at $0.99 per outcome. A resolution, defined as "No further help is requested after Fin's last answer," is one outcome; others include a procedure handoff, where Fin completes a configured handoff to a person, also at $0.99, and a qualification, where Fin matches a prospect to criteria you defined, at $9.99. You pay for one outcome per conversation at most, and the page says you are not charged "when a conversation is simply passed to your team without an outcome."

The pattern in both is the same: a narrow outcome, a written definition of it, and a rule for who decides whether it happened.

Which work is ready to be bought as an outcome?

Work is ready to be bought as an outcome when it is already outsourced, follows rules rather than taste, and produces a result someone can check. Work that depends on judgement, relationships or accountability stays with people, even if AI helps.

The Sequoia essay gives a useful test. "If a task is already outsourced, it tells you three things. One, the company has accepted that this work can be done externally. Two, there's an existing budget line that can be substituted cleanly. Three, the buyer is already purchasing an outcome." It separates intelligence, following complex rules, from judgement, deciding what to do next: "Writing code is mostly intelligence. Knowing what to build next is judgement."

Checklist: is this work a candidate?

1. It is already outsourced, or it could be without anyone objecting.

2. The finished result can be described in one or two sentences.

3. Someone can check the result without redoing the work.

4. The volume is steady enough to price per unit.

5. A wrong result can be caught and fixed before it does lasting harm.

6. The judgement calls in the process can be pulled out and kept with a named person.

Back-office work such as invoice matching and reconciliation often passes most of these; our post on AI in the finance back office shows where the human checks belong.

What should a buyer check before signing an outcome-priced contract?

Check how the outcome is defined, who verifies it, what happens when it is wrong, and what you pay when volume or quality moves. An outcome contract is only as good as its definition of a finished result.

1. The definition. Write down exactly what counts as one outcome and what does not, with examples of borderline cases.

2. Verification. Agree who decides whether an outcome happened: the vendor's system, your team or a sample review. If the vendor's own AI judges its own work, ask to see how and audit a sample.

3. Quality, not just count. Pair the price with a quality measure, such as reopened tickets or corrections found later, so cheap but wrong results cost the vendor.

4. Minimums and caps. Check for monthly minimums and set a ceiling, so a spike in volume does not produce a surprise bill.

5. Escalation and liability. Agree what the vendor owes when a result is wrong, and who handles the customer or regulator when it matters.

6. Your data and your exit. Make sure you keep the inputs, outputs and logs, and can move the work back in-house or to another vendor.

This is a practical checklist, not legal or financial advice; have your own counsel review the contract. Before you sign, run the vendor on your own past cases, the same way you would test an AI agent before it goes live.

How we approach it at Vizio AI

We start every AI engagement by writing down the unit of work: what a finished result looks like, how we will check it, and what it costs today in people's time and outside fees. That baseline is what any new system, or any vendor, has to beat.

We then split the process into the steps that follow rules and the steps that need judgement, as described in What Is an AI Agent, Really?. The rule-following steps go to workflows and agents with a measurable output; the judgement stays with a named owner on the client's side. Verification, logs and the data stay with the client, so the work can be measured per outcome whether we run it, they run it or another vendor does.

This runs across our three studios: Venture Studio for new products, Growth Studio for revenue work and DX Studio for operations.

Sources

• Sequoia Capital, Julien Bek, Services: The New Software (5 March 2026): https://sequoiacap.com/article/services-the-new-software

• Zendesk, Zendesk First in CX Industry to offer Outcome-Based Pricing for AI Agents (28 August 2024): https://www.zendesk.com/newsroom/articles/zendesk-outcome-based-pricing/

• Zendesk help centre, About automated resolution tiers: https://support.zendesk.com/hc/en-us/articles/9570369117338-About-automated-resolution-tiers

• Intercom, Fin pricing (checked 31 October 2026): https://fin.ai/pricing