AI-enabled services framework

Market attractiveness: the eleven-criterion framework

Eleven criteria across three groups — market structure & demand, delivery economics, and defensibility — for reading whether an AI-enabled services business will be a good one. Click any criterion for the strong and weak signal to score it against, or switch to table view to scan a whole section at once.

Criteria 01–05

Market structure and demand

How splintered is supply, and how weak is lock-in to any single incumbent?

Strong signal

Thousands of sub-scale providers; no vendor holds >5% share; switching is routine; no leading brand name and/or category isn’t yet defined

Weak signal

Oligopoly with entrenched multi-year contracts and deep workflow integration

Even if they see it coming, can incumbents build or buy AI, or are they structurally blocked?

Strong signal

Billable-hour, commission, or offshore-labor models incumbents can't cannibalize

Weak signal

Tech-forward incumbents with capital and incentive to adopt AI quickly

How essential is the service, and how expensive is a mistake, independent of spend share?

Strong signal

Non-discretionary work where errors trigger penalties, stranded goods, or liability

Weak signal

Discretionary, deferrable work that gets cut in downturns; errors are cheap

If AI cuts the price 80%, does volume grow more than 5x, or does the market simply shrink?

Strong signal

Large latent demand currently un-served at human price points

Weak signal

Fixed unit volume; efficiency gains become a pure price war

Is the legacy practitioner base aging or shrinking, leaving customers already underserved?

Strong signal

Retiring workforce, thin training pipeline; buyers actively shopping for alternatives

Weak signal

Abundant, cheap labor supply keeps the status quo comfortable

Criteria 06–08

Delivery economics

What fraction of delivery cost is labor AI can actually do, and where is the irreducible human floor (liability, physical work, relationships)?

Strong signal

70%+ of COGS is process work (document review, reconciliation, data entry); human layer is thin and supervisory, with a path from ~35% to 70%+ gross margin

Weak signal

Delivery requires physical presence, bespoke judgment, or relationship work that caps automation well below half of COGS

Can correctness be objectively measured, and how costly is a mistake? Together these set the depth of autonomous operation.

Strong signal

Output is binary or testable (a vulnerability reproduces or it doesn't); errors are cheap and caught early. Trust compounds fast

Weak signal

Quality is subjective or errors surface late and expensively, forcing human review loops that cap margin and slow scaling

Does the market's pricing convention let you decouple price from cost, or does AI efficiency get competed away?

Strong signal

Outcome-based pricing norms (% of collections, per-finding, bps on premium). Customer never sees your cost structure

Weak signal

Hourly or headcount-based pricing where efficiency gains are visible and quickly passed through to price

Criteria 09–11

Defensibility

Is the work recurring (contractual), re-occurring (predictably repeated), or genuinely one-off?

Strong signal

Always-on operational work (billing cycles, managed services, ongoing administration) with natural contract structures and high logo retention

Weak signal

Episodic, project-based engagements requiring constant resale; revenue resets to zero each period

Does delivering the service generate proprietary data that makes the Nth engagement measurably better?

Strong signal

Each engagement compounds a dataset no competitor can replicate (loss histories, negotiated market terms, building-level operating patterns)

Weak signal

Engagements are independent; defensibility reverts to relationships and switching costs as foundation models commoditize capability

Is licensing a moat once crossed, or does regulation mandate a human in the loop, capping automation depth permanently?

Strong signal

License is hard to obtain but one licensed human can supervise unlimited AI throughput, which slows every follower equally

Weak signal

Per-transaction human sign-off required by statute, a structural ceiling no model improvement removes; or no regulation at all, inviting unlimited entry

The complete framework: eleven criteria across market structure & demand (01–05), delivery economics (06–08), and defensibility (09–11).