The economics.
An AI engineer at the level required to build business-specific systems on top of CCH Axcess, ServiceTitan, or iManage costs $185K-$280K in fully-loaded total comp in most US markets, $140K-$200K in remote-eligible roles. The ramp is six months minimum: the engineer needs to learn your stack, your data, your business's specifics, and build the cultural relationships that make adoption work.
A commissioned build runs $45K-$180K fixed-fee for one specific workflow shipped in 4-7 weeks. The boutique brings the playbook; you bring the data and the partner-group buy-in.
One year of an internal hire's loaded cost ($185K-$280K) is roughly 1-3 commissioned builds. Two years is 3-6. The math favors hire only if the operation has 3-6 distinct AI builds queued for the next two years AND the cultural infrastructure to absorb them.
Where internal hire is the right answer.
Three patterns:
The operation with 3-6 AI builds queued, not one. A multi-year roadmap of distinct workflows justifies the hire's loaded cost across multiple deliverables. One-shot builds don't.
The operation where AI capability is becoming a competitive moat. Some operators (large litigation practices, complex underwriting agencies, multi-brand PE platforms with continuous M&A) genuinely use AI as competitive differentiation, year after year. The continuous evolution requires an in-house owner.
The operation with the cultural infrastructure to absorb a senior IC hire. The Head of AI needs partner-group access, stack admin permissions, and authority to override workflow inertia. Firms without this infrastructure burn the hire in 18 months and re-hire from scratch.
Where commissioned build is the right answer.
Three patterns:
The operation with one or two named workflow problems, not a roadmap. "Our quote turnaround is 6 hours and it's killing our win rate" is a build problem. Hiring a Head of AI to solve it is a $200K-$280K answer to a $90K question. Commission instead.
The operation where AI is an enabler, not a moat. If reclaimed senior capacity is what the operation needs (and what most $8M-$50M firms need is exactly that), commissioning the build for that one bottleneck is the right structure. The system runs; the senior staff get their hours back; nobody needs to manage an internal engineer.
The operation that doesn't want to manage an engineer. Hiring engineers is a competence. Some operators have it; many don't. Firms in the second category should not start having it for the sake of one AI build. Commission.
Three failure modes of internal hire.
The Head of AI hired without queue. Firm hires a strong engineer; engineer ships the first thing in three months; then there's no clear next thing. Engineer leaves within 18 months because the work is repetitive maintenance. Firm starts over.
The Head of AI hired without authority. Engineer is brought in; partners refuse to give them workflow-override authority; engineer's deliverables don't ship because they cannot push past adoption resistance. Engineer leaves frustrated.
The Head of AI hired too senior or too junior. Senior engineers want to ship to many users; one mid-market operator doesn't have the surface area to keep them engaged. Junior engineers can't navigate the partner politics. The right level for a single-firm Head of AI is narrow.
What we recommend.
Most $8M-$50M firms in our segment have one workflow they want to ship in the next 90 days, not a roadmap. Commission the first build. If the operation finds, after two or three commissioned builds, that the queue is genuinely deep and ongoing, hire the Head of AI to maintain and extend. This sequence is more reliable than the reverse (hire first, hope the queue justifies it later).