AI consultant vs. in-house hire: the math.

Should you hire an AI person or engage a consultant?

Below about 100 employees, engaging is usually the better math; a full-time AI hire starts paying for itself only when you have continuous build work — several new workflows a quarter, not several a year. Above roughly 150–200 employees, or with a genuine data team already in place, hiring wins.

That is the summary. The rest of this guide is the arithmetic behind it, using real numbers on all three options — including the one nobody prices: doing nothing. We sell the middle option, so read the numbers, not our conclusion.

What does an in-house AI hire actually cost?

Plan on $165,000–$280,000 a year, fully loaded, for someone who can genuinely build — an automation or machine-learning engineer, not a prompt hobbyist. That is a $130,000–$200,000 salary plus the standard 25–40% for taxes, benefits, and overhead, and it repeats every year.

The salary is only the visible cost. Add three to six months of ramp before the first workflow ships. Add tooling and infrastructure. Add the market reality that this person is being recruited weekly by companies with deeper pockets, so factor replacement risk into any three-year plan. And note the structural problem no salary fixes: one person is a single point of failure holding systems only they understand. If they leave in month fourteen, you own software with no author.

None of this means the hire is wrong. It means the hire is a factory, and a factory only pays when there is a production line to run.

What does engaging a consultancy cost?

At our published prices, a complete first build costs $15,500–$53,500, one time: an AI Readiness Audit at $3,500–$8,500, plus an Implementation Sprint at $12,000–$45,000, fixed in writing before work begins. There is no salary, no ramp, and no replacement risk — and no meter running, because the price cannot move after the start.

What you get for that is a working system in about 90 days per workflow, your team trained to run it, and ownership of everything at handoff. The trained-team part matters for this comparison: our model deliberately ends with your people running the machine, which is what makes the engage-then-own path viable for a small firm. If you want ongoing coverage as models and vendors shift, Managed Services is a monthly retainer you can cancel any month — a fraction of a salary, and only for as long as it earns its keep. The full price list is published; what the audit deliverable looks like is published too.

What does doing nothing cost?

More than zero, and it compounds. Our ROI calculator models the drag at roughly $4,600–$6,400 per employee per year depending on industry — repetitive manual work plus the revenue capacity it eats — which puts a 25-person firm at $115,000–$160,000 a year, every year.

Those are modeled assumptions, stated in the open, not your books — the calculator shows every number it uses. But even if your real figure is half the model, doing nothing is the most expensive option on this page within two years. It just bills you in a currency that never shows up on an invoice: evenings, errors, and the slow leak of hours your best people spend re-typing what one system already knows into another. The whole model — every industry rate, one worked example, and what breaks it — is published in the ROI math guide.

Where is the break-even?

Divide the loaded cost of a hire by the cost of a sprint: $180,000 a year buys roughly four maximum-size sprints, or a dozen small ones. So the hire breaks even only when you need at least four significant new builds a year, sustained — a pipeline that in our experience appears somewhere north of 100 employees, or in unusually systems-heavy smaller firms.

There is a second lens: headcount. At the calculator’s modeled $4,600–$6,400 of recoverable drag per employee, a 30-person firm has perhaps $150,000 a year on the table — enough to justify several engagements, nowhere near enough to feed a salary that consumes the entire recovery before it starts. The hire begins to make sense when the opportunity pool is several times the loaded cost, which again points somewhere past 100 people. And if neither hiring nor engaging feels right yet, check whether the foundations are the real blocker: the not-ready signs are published.

Here is the same comparison laid flat:

Hire figures are typical 2026 US market ranges; engage figures are our published price list; do-nothing figures are the ROI calculator’s modeled assumptions, not your books.

And the disclosure the table deserves: if you are 200 people with a data team, hire — that is the right answer even though we sell the other row. The worst outcome on this page is not choosing wrong between hiring and engaging; it is defaulting into row three without ever pricing it.

What about a fractional AI hire?

A fractional AI leader — a consultant on a $2,000–$10,000 monthly retainer, a day or two a week — buys you judgment without a salary, and it is the right answer for some firms. Just be clear about what it is not: hands that build.

Fractional arrangements shine when the problem is direction — which vendors, which workflows, what to say no to — and when someone in-house will do the building. They disappoint when the deliverable everyone actually wants is a working system, because advice compounds slowly and retainers do not expire. Run the same test you would run on any consultant: what ships, by when, and who owns it? A fractional advisor who cannot point at something that will exist in 90 days is a subscription, not a strategy. Many buyers end up combining models — fractional judgment plus fixed-price builds — which is roughly what our audit-then-sprint sequence packages into one engagement.

Is there a middle path?

Yes, and it is the one most of our clients actually take: engage for the builds, then put a part-time internal owner — promoted from operations, not hired from a lab — in charge of running them. You get consultant-grade construction without consultant dependency, and an owner without an engineer’s salary.

This works because of how the handoff is structured. Every build ends with your team trained and everything owned by you; the internal owner inherits documented systems, not mysteries. Later, if the pipeline of new builds thickens, that owner becomes the justification — and the hiring manager — for the real in-house hire. The sequence for most firms under 100 people: run the readiness checklist, audit, build once, train, then decide about ongoing coverage with real usage in front of you. Whoever you engage for it, vet them with the ten questions first, and walk in knowing what the market charges.