AI-Enabled Services: The Business Model and Its Margins

An AI-enabled services company sells a finished piece of work, such as a filed claim, a closed book of accounts or a resolved support ticket, and uses AI to do most of the production while a smaller team of experts handles judgment and accountability. The catch is margins: they tend to sit between a traditional services firm and a software company.

Definition: An AI-enabled services company (also called an AI services company or "services as software") is a business that charges for completed outcomes in a services market, delivering them with AI systems plus a reduced human team, so that revenue grows faster than headcount.

The pitch is easy to like. Services budgets are often far larger than software budgets. Sequoia partner Julien Bek estimates in a March 2026 essay that for every dollar spent on software, six are spent on services. General Catalyst's August 2025 essay on the future of services puts US services revenue at more than $6 trillion a year against a software market of about $370 billion.

Big market, though, is the easy part. Whether it turns into venture returns depends, in our view, on the unit economics. The multiple gets earned with evidence. Calling yourself "services as software" doesn't earn it.

What an AI-enabled services company is, and three ways to build one

The first decision is who you sell to: the professional, or the professional's customer. Bek draws the line cleanly, and we think it is one of the more useful distinctions in this category:

Copilots sell a tool to the professional, who stays responsible for the work. A legal AI product sold to law firms is a copilot.
Autopilots sell the finished outcome directly to the end customer, bypassing the professional firm. A company that drafts NDAs for businesses, rather than selling drafting software to lawyers, is an autopilot.

Autopilots are the AI-enabled services model. We like starting with work customers already outsource. The budget line exists, the buying habit exists, and you can move toward work they do in house as the models improve.

There are three common build paths:

Software (copilot). Sell seats or usage to existing firms. Typically the highest gross margin, but you capture the tool budget, not the work budget.
AI-native services (autopilot). Start a new firm that sells outcomes, with AI doing most of the production from day one. You capture the work budget but carry delivery risk and human costs.
AI-enabled rollup. Buy existing services firms and re-engineer them with AI. General Catalyst, which invests behind this strategy, targets 30 to 40 percent margins with 10 to 20 percent growth, a standard it calls the Rule of 60. You get customers and revenue immediately, but you also inherit legacy staff, culture and contracts.

AI-enabled services margins: how they compare with software and services

The whole argument lives in the margin gap between services and software. So it helps to anchor on real numbers from filings.

Company (fiscal year)	Model	Revenue	Gross margin
Accenture (FY2025, ended Aug 31, 2025)	Professional services	$69.7B	31.9%
Salesforce (FY2026, ended Jan 31, 2026)	Software, whole company	$41.5B	about 78%
Salesforce subscription and support	Software only	$39.4B	about 83%
Salesforce professional services	Services inside a software firm	$2.1B	about minus 16%

Accenture's fiscal 2025 results show revenue of $69.672 billion, cost of services of $47.438 billion and a 31.9 percent gross margin, with about 779,000 people, roughly $89,000 of revenue per employee. Salesforce's fiscal 2026 results show $41.525 billion of revenue and $32.255 billion of gross profit. Its professional services line brought in $2.137 billion at a cost of $2.474 billion.

That last line is the interesting one. By these figures, even a software leader runs its services arm at a loss to support subscription sales.

AI-enabled services companies are trying to live in that gap. Three costs largely decide where you land:

Human delivery cost: the experts who review, sign off on and fix the AI's output.
Inference and tooling cost: model calls, which can be heavy for agent workflows. The AI gross margins guide covers how to model this line.
Exceptions: the share of jobs the AI cannot finish, which fall back to humans.

Here's the test we'd run. A founder who can show revenue per employee rising, human hours per job falling and inference cost per job under control has a credible path to margins well above a traditional firm. A founder who can't show those trends may have a services firm wearing a software pitch deck.

Price outcomes, not hours

We'd be wary of billing by the hour for work AI does. Hourly billing punishes efficiency. If AI cuts a 10-hour job to 2 hours, an hourly firm loses 80 percent of the revenue. Price the outcome instead, and the efficiency becomes margin.

Sierra, the AI customer agent company, is a public example. Its December 2024 pricing post describes charging when the agent completes a task, such as a resolved support conversation, a saved cancellation or an upsell, with no charge in most cases if the conversation is unresolved or escalated to a human. Revenue moves with the customer's result.

Four structures are common:

Per outcome: a fee per resolved ticket, filed claim or completed return.
Fixed fee per engagement: a flat price for a defined deliverable, such as a monthly close.
Share of value: a percentage of savings or recovered revenue, common in collections and revenue cycle work.
Hybrid: a platform fee plus outcome fees, which smooths revenue while keeping the incentive.

The pricing worksheet walks through anchoring a price to customer value, the first step for any of these.

Worked example: a traditional firm vs an AI-enabled services firm

The numbers below are illustrative assumptions, not market data, and real figures vary widely. Both firms sell $15 million a year of the same work.

Traditional firm: 100 people

Delivery: 85 staff at $105,000 loaded cost, plus $300,000 of delivery tools: cost of revenue $9.225M.
Gross margin: 38.5 percent. Revenue per employee: $150,000.
Sales commissions at 5 percent of revenue ($750,000) give a contribution margin of 33.5 percent ($5.025M).
Other overhead (15 sales and admin staff at $110,000 plus $1.2M of other costs): EBITDA $2.175M, a 14.5 percent margin.

AI-enabled firm: 40 people

Delivery: 24 experts at $130,000 loaded cost, $1.05M of inference (7 percent of revenue), and $300,000 of tools: cost of revenue $4.47M.
Gross margin: 70.2 percent. Revenue per employee: $375,000.
The same 5 percent commissions give a contribution margin of 65.2 percent ($9.78M).
Overhead of 8 engineers at $200,000, 8 sales and admin staff at $120,000 and $1.2M of other costs: EBITDA $6.02M, a 40.1 percent margin.

Now the valuation question. Does the market attach a software-style multiple or a services multiple?

If a buyer pays 1.5 times revenue for the traditional firm, that is $22.5M, or about 10.3 times its EBITDA. Apply 10 times EBITDA to the AI-enabled firm and it is worth about $60.2M, or 4.0 times revenue. If investors treat it as software and pay 6 times revenue, it is worth $90M. If they treat it as a services firm and pay 1.5 times revenue, it is worth $22.5M despite nearly three times the profit.

Same company. Four times the price, depending on what investors believe.

And the AI-enabled numbers are more fragile than they look:

Inference doubles to $2.1M: gross margin falls to 63.2 percent and EBITDA to $4.97M (33.1 percent).
Automation disappoints and delivery needs 36 experts, not 24: gross margin falls to 59.8 percent, EBITDA to $4.46M (29.7 percent), and revenue per employee to about $288,000.

Our read: in this example the valuation gap comes from the multiple investors choose, and that choice tends to follow evidence that the margin is structural. The business models and unit economics guide explains how to present gross and contribution margin, and the seed valuations guide shows why the multiple matters so much for venture returns.

What the early evidence shows about AI replacing delivery work

We'd treat company anecdotes as proof of mechanism, not as benchmarks. Four patterns from 2026 are worth knowing.

A revenue channel can stop scaling with people. At Eight Sleep, when the head of a two-person email marketing team left, a co-founder built AI bots in about three days to run the channel. Email marketing now runs with no dedicated employees and brings in close to $100 million. Counting internal AI agents, the company describes itself as three to four times larger than its human headcount. That is, arguably, the promise of AI services in miniature.

Process time falls sharply when engineers sit with the business. Uber paired a pod of 30 of its best AI engineers with business teams to rebuild processes one at a time. A weekly pricing allocation process went from 15 hours to 2, forecasting from 8 hours to 2, and marketing QA from two weeks to two days.

Uber's lesson on ROI matters for anyone selling to large buyers. Freed hours fill with other work. A more reliable way to capture AI efficiency is a tighter budget, such as holding headcount flat next year if AI makes people 10 to 30 percent more productive. Your customers may judge you on a lower cost line more than on hours saved.

Forward deployed engineering can be a warning light, not a model. When a vendor sends engineers to sit inside the customer and build as the technology evolves, the product often isn't finished. Sierra's first founding engineer effectively worked inside a customer, and its forward deployed team now gets large enterprises live in weeks. Heavy human deployment is normal early. Many investors read it as a sign the workflow isn't yet solved in software, so it helps to show it falling per customer.

Cost cutting alone rarely makes a rollup work. Buying a pre-AI company and simply cutting costs and raising prices is unlikely to be enough today. In our view the acquirer also needs to generate new revenue with AI and re-engineer the company. That is a much harder management job than optimization.

"But services markets are six times bigger than software"

That's the bull case, and it deserves a fair hearing. Services markets are many times larger than software. AI finally lets a firm grow revenue without growing headcount in step. Outcome pricing lets an AI-native firm undercut incumbents and still keep more margin. Bek goes furthest, suggesting the next trillion-dollar company could be a software company that looks like a services firm from the outside.

But.

Jason Lemkin, the SaaStr founder, is the sharpest skeptic. In August 2026 he argued that venture money is unlikely to turn accounting or law firms into the next breakout AI company, that structures built on sister companies with partial ownership are too convoluted, and that the model looks better on a spreadsheet than in practice. He grants it is possible but expects the category to produce no exits. In his view the business may be worth a services multiple whatever it is called.

Then there's execution. TechCrunch reported in September 2025 on research from Stanford's Social Media Lab and BetterUp Labs, surveying 1,150 employees, on "workslop": AI output that looks finished but takes real time to fix. The researchers estimated its cost at $186 per employee per month. If an AI services firm's experts spend their day correcting output, the margin story collapses back to a services margin.

Where we land on the AI services multiple

We think multiples will be earned company by company, based on how much of delivery is actually software. We agree with the bulls on the size of the market and with the bears on the burden of proof.

A big market isn't a moat. What compounds is the data each job leaves behind, which is the same point we make in our take on why AI alone isn't a moat. Show the automation rate climbing and you usually persuade more than any vision slide.

That's our answer, not the answer. A founder in a niche where licensed sign-off is unavoidable may reasonably accept a lower ceiling in exchange for a real business.

Build software, AI-enabled services or a rollup: a decision checklist

Questions worth answering before you choose a model:

Who owns the outcome today? If customers already outsource the work, an autopilot can take an existing budget line. If professionals do it in house and want to keep control, a copilot is easier to sell.
How much of the work can AI finish without a human today? If most jobs still need expert review, expect services-like margins at first and plan the path down.
Is there liability or licensing? Regulated work (legal, audit, medical coding) may require licensed people to sign off and can limit who is allowed to own the practice. That is what pushes some founders into convoluted ownership structures.
Can you price the outcome? If customers insist on hourly rates, your efficiency may become their discount.
Do you need acquisitions to get distribution? A rollup buys customers and trust, but needs large checks, debt or both, plus an integration team.
Can you run a services culture? Services firms live on utilization, quality control and client management. Software founders often underestimate this.
What does your data compound into? Today's judgment work should become tomorrow's automated work as you collect proprietary data. If each job leaves you with data that makes the next job cheaper, you may be building a moat; see AI startup moats.

AI services metrics to track from day one

Investors are likely to ask for most of these, so build them into your reporting early:

Gross margin by month and by customer cohort, with a clear split of human delivery, inference and tooling costs.
Revenue per employee and human hours per job, trended over time.
Automation rate: the share of jobs completed without human rework, and how it has changed.
Inference cost per outcome and sensitivity to model price changes.
Pricing structure and how much revenue is outcome-based versus hourly.
Quality metrics: error rates, rework, customer disputes and liability claims.
Forward deployed or implementation headcount per new customer, and whether it is falling.
For rollups: acquisition prices, integration timelines and the post-acquisition margin of each acquired firm.

The mistakes that tend to hurt in diligence:

Reporting blended margins that hide human cost. Investors often rebuild the cost line from headcount anyway.
Counting forward deployed engineers as sales or R&D when they are really delivery.
Pricing by the hour and giving efficiency gains away.
Scaling headcount with revenue. If people grow as fast as sales, the market is likely to value you as a services firm.
Buying firms before the AI works. A rollup of unautomated firms is a traditional rollup with a higher cost of capital.

Two 1752vc programs fit this category at different moments. If you already have real traction as an AI-native company, the Lightning Round pitch competition is a fast way to learn whether investors read your margins as software or services. If the product is validated but you still need to sell outcomes to skeptical buyers, the 12-week, remote GTM Accelerator is one place to build that sales motion.

The bottom line

AI-enabled services can be a great business and still get a mediocre multiple. The gap closes with trend lines, not labels: fewer human hours per job, more revenue per person, inference under control.

The pitch says software.

The payroll decides.

Key takeaways

AI-enabled services companies typically sell finished outcomes in services markets, using AI for most production and a smaller expert team for judgment and accountability.
Filings show the gap they are trying to close: Accenture reported a 31.9 percent gross margin for fiscal 2025, while Salesforce's fiscal 2026 gross margin was about 78 percent.
Outcome-based pricing, of the kind Sierra publishes, can turn efficiency into margin instead of lost billable hours.
The multiple is contested: bulls point to a services market many times larger than software, skeptics expect rebranded services firms to earn services multiples, and in our view evidence of real automation decides it.
It helps to track gross margin by cost line, revenue per employee, automation rate and inference cost per outcome from day one.

Frequently asked questions

What is an AI-enabled services company?

An AI-enabled services company sells completed work, such as processed claims, bookkeeping or resolved support tickets, instead of selling software to the people who do that work. AI handles most of the production and a smaller team of experts reviews output and takes responsibility. The goal is to grow revenue without growing headcount at the same rate, which lifts margins above a traditional services firm.

What gross margins do AI services companies have?

There is no reliable public benchmark yet, because most AI services companies are private. They typically aim to land between traditional services firms and software companies. For reference, Accenture reported a 31.9 percent gross margin for fiscal 2025 and Salesforce about 78 percent for fiscal 2026. General Catalyst's rollup strategy targets 30 to 40 percent profit margins.

What is the difference between AI services and SaaS?

SaaS sells access to software that the customer's team uses to do the work, usually priced per seat or by usage, with gross margins often near 80 percent. An AI services company does the work itself and sells the result, often priced per outcome. It captures a larger budget but carries delivery costs, human review and liability that SaaS vendors avoid.

Do VCs invest in AI-enabled services startups?

Some do, but opinion is split. Sequoia has argued that services are the next large market for AI, and General Catalyst has committed capital to AI-enabled rollups, while prominent skeptics doubt these companies will earn software multiples. Investors tend to focus on margin trends, revenue per employee and the share of work completed without human rework.

What is an AI rollup?

An AI rollup is a strategy of acquiring existing services businesses, such as IT managed service providers or accounting firms, and using AI to raise their margins and growth. General Catalyst calls it the AI-enabled roll-up. The risk, as we see it, is that cost cutting alone is not enough: the acquirer also needs to re-engineer delivery and generate new revenue with AI.

Sources

Disclaimer: This guide is for general education only and is not legal, tax or investment advice. Laws, market data and program terms change, so it may not reflect the latest developments or fit your situation. Treat it as a starting point, not a source of truth, and talk to a qualified lawyer, accountant or financial adviser before you make decisions.