AI consultant vs in-house hire: the break-even arithmetic nobody shows you
Short answer
Hire in-house when AI work is permanent and central — you need new or reworked automations continuously, or AI is part of the product you sell. Use a consultant when the work is a project with an end: an audit, the first several builds, a stalled rollout. The arithmetic: a US-based automation or AI engineer costs roughly $150,000-230,000 a year fully loaded, before the three-to-five months it takes to hire and ramp one. Consultant-built automation at published rates runs a few thousand dollars per workflow plus a support retainer. The break-even is cadence, not headcount maths: below roughly two genuinely new workflows a month, sustained, the salary cannot pay for itself in automation output alone.
The question usually arrives as “should we hire someone for AI?” — asked by a founder or ops lead who has a list of manual processes, a budget line that could become either a salary or a project fee, and no neutral party to ask. Recruiters will say hire. Consultants will say consultant. This article shows the arithmetic and names the cases where each answer is honestly right, including the ones where ours is not.
Last updated: 9 September 2026.
Where these numbers come from
The salary figures are typical US market ranges for automation and AI engineering roles as of 2026 — the kind you can verify in an afternoon on any salary aggregator or by posting the role. They are ranges rather than points because seniority and market move them substantially. The consultant figures are simply our published prices, which are on the AI automation consultant page and repeated below; agencies charge multiples of them, and those market ranges are in the pricing guide.
One bias worth declaring: we are a consultancy. The correction for that bias is built into the article — the hire-wins cases are listed first, and the break-even is stated as a number you can check against your own backlog rather than a vibe.
The three cases where hiring wins outright
Get these out of the way first, because when they apply, no consultant arithmetic beats them.
AI is in the product. If customers pay for something with AI inside it, the work is permanent, competitive, and central. You need people who wake up thinking about it, accumulate context for years, and are in the room when the product changes. Consultants are structurally wrong for this — not too expensive, wrong.
Maintenance alone is a real job. Somewhere around several dozen live workflows across multiple departments, keeping the estate running — integrations breaking on vendors’ schedules, policies changing, new edge cases — becomes a genuine full-time role. At that point a retainer is a worse deal than a salary.
The data cannot leave. Regulated environments where processing, credentials, and context must stay resident make external engagement structurally awkward. A consultant can design and train, but the operating capability has to live in-house.
If one of those three describes you, hire. The rest of the article is for everyone else.
What the hire actually costs
The sticker price is the base salary. The real price has four more lines.
| Cost line | Typical figure | The part people forget |
|---|---|---|
| Base salary | $110,000–180,000 / year (US) | Automation engineers who are also comfortable with AI tooling sit at the upper half of the range. |
| Loaded multiplier | × 1.25–1.4 | Payroll taxes, benefits, equipment, software seats, and the manager hours the role consumes. |
| Hiring | 3–5 months of search and ramp | Paid twice: recruiter or sourcing time, plus the manual work continuing at full cost while the seat is empty. |
| Utilisation | Rarely 100% automation work | In a 30-person company the "AI hire" becomes the default owner of every technical odd job within a quarter. |
| Key-person risk | One resignation | Every undocumented workflow becomes an emergency the day they hand in notice. |
Fully loaded, the realistic annual figure is $150,000–230,000, and the first production workflow ships a quarter or two after you decide to hire — because first you have to find the person.
None of this is an argument against hiring. It is the denominator for the break-even.
What the consultant actually costs
Our prices, since they are published anyway: a $490 audit (credited in full against the first build), workflows from $1,900 each, support at $690 a month including one new or reworked workflow. Agency rates for the same deliverables run roughly five times higher — the ranges are in the pricing guide, and the reasons ours are lower are on the service page.
A representative first year — audit, five builds of mixed complexity, support retainer for the last eight months — lands in the low tens of thousands of dollars. Call it a fifth to a tenth of the loaded hire, delivered starting in weeks rather than after a hiring pipeline.
The catch, stated plainly: that money buys outcomes, not capacity. When the fifth workflow ships, the consultant stops costing money — and also stops being around. If what you need is a person absorbing ambiguity full-time, no per-project price competes with a salary, because that is not what per-project pricing sells.
The break-even is cadence
Strip away the positioning and the decision reduces to one question: how many genuinely new automation workflows do you need per month, sustained?
| Your realistic cadence | Cheaper structure | Why |
|---|---|---|
| A backlog of 3–8 workflows, then quiet | Consultant, project pricing | The work has an end. A salary keeps being paid after the backlog is empty. |
| ~1 new workflow a month, ongoing | Support retainer | $690/month buys maintenance plus that cadence. This is most 10–100 person companies, indefinitely. |
| 2+ new workflows a month, sustained for a year+ | Hire (consultant as bridge) | At this volume, per-project fees approach a salary — and a resident engineer compounds context that a consultant re-learns per engagement. |
| Unknown — that is the problem | Audit first | A $490 audit produces the ranked backlog. Its length answers the hiring question with data instead of a guess. |
Two honest asymmetries to weigh alongside the table. The hire compounds: an engineer who has been inside your business for a year sees automation candidates nobody briefs a consultant on. The consultant de-risks: no ramp, no severance, patterns already learned on other people’s money, and if the first build disappoints, the exit costs nothing.
The failure mode on each side
The premature hire. A 25-person company hires a $150k automation engineer off the back of an enthusiastic quarter. The genuine backlog turns out to be six workflows — about two months of specialist work. By month four the engineer is fixing the wifi and rebuilding the CRM fields; by month nine they leave, and the workflows they built are now unowned. Total cost of learning the backlog was short: well over $100,000. A $490 audit would have measured it in two weeks.
The eternal consultant. The opposite trap: a company outsources so long that no internal capability ever forms, pays retainer prices for what has become routine internal work, and treats every process change as a change order. If your consultant’s incentives are healthy, they will tell you when you have crossed into hire territory — the shortlist-length test above is exactly that tripwire, and it is why our audit sometimes ends with “hire, and here is the job description.”
The middle options, briefly
Consultant builds, generalist maintains. Hire an ops-minded generalist (much easier to find than an AI specialist), let a consultant do the initial heavy builds, and hand over with runbooks. The generalist grows into the role on top of working systems instead of a blank page. This is the highest-success pattern we see in 20–100 person companies.
Fractional. A part-time senior automation person one or two days a week. Works when you find someone good; the market for them is thin, and the failure mode is the same key-person risk as a hire with less of the loyalty.
Whatever structure you pick, three non-negotiables protect you: you own the accounts and credentials, every workflow has a runbook a non-expert can follow, and each automation has a named internal owner. Those three make switching structures later a decision instead of a crisis — and anyone, employee or consultant, who resists them is telling you something.
How to decide this week
- List the manual processes you believe are automatable. Cost each one: instances per month × minutes × loaded hourly rate. (The free process audit tool walks through exactly this.)
- Estimate the build backlog honestly. Under ten workflows: that is project work, not a job.
- Ask what happens after the backlog ships. If the answer is “maintain and add occasionally”, that is a retainer. If it is “there will always be more because this is our product” — write the job description.
- Check the three hire-wins cases at the top. Any of them apply? Hire, full stop.
- Still unsure? The audit measures the backlog for $490 and is credited against a build. Its result is also the honest input to the hiring decision — either way you stop guessing.
Frequently asked questions
Should I hire an AI consultant or an in-house AI engineer?
Decide on cadence and permanence. If you need automations built continuously — several new or reworked workflows every month, indefinitely — or AI is core to the product you sell, hire. If the work in front of you is a bounded set: find the expensive processes, build the top handful, keep them running — a consultant delivers that for a fraction of one year's loaded salary, with no ramp time and no single point of failure.
How much does an in-house AI engineer actually cost?
In the US, base salaries for automation and AI engineers typically run $110,000-180,000 depending on seniority and market. Fully loaded — payroll taxes, benefits, equipment, software, management time — multiply by roughly 1.25-1.4, which lands at about $150,000-230,000 a year. Add hiring: three to five months of search and ramp before the first production workflow ships, which is a real cost paid in manual work continuing.
How much does an AI consultant cost compared to a hire?
At our published rates: a $490 process audit credited against the first build, workflows from $1,900 each, and support at $690 a month. A first year covering an audit, several builds and a support retainer typically lands in the low tens of thousands — against $150,000-230,000 loaded for a hire, before recruitment costs. The consultant number scales with output; the salary is paid whether workflows ship or not.
When is hiring in-house clearly the right choice?
Three cases. AI is part of the product you sell, so the work is permanent and competitive. Your automation estate is large enough that maintenance alone is a real job — dozens of live workflows across departments. Or your data cannot leave the building for regulatory reasons and you need the capability resident. In all three, a consultant is at best a bridge while you hire.
What is the risk of relying on a consultant instead of hiring?
Knowledge leaving at handover. Mitigations to insist on regardless of who you use: you own the accounts and credentials, every workflow ships with a runbook written for a non-expert, and someone internal is named as owner of each automation. If a consultant resists any of those, that is your answer about the consultant, not about the model.
Is there a middle option between consultant and full-time hire?
Two. A support retainer — a consultant keeps existing automations running and adds roughly one new workflow a month, which covers most 10-100 person companies indefinitely. Or hire an ops-minded generalist internally and let a consultant do the initial heavy builds; the generalist inherits maintenance with the runbooks. The expensive mistake is the reverse order: hiring a specialist first and discovering there is only three months of specialist work.
Can we start with a consultant and hire later?
It is the standard path, and a good consultant should make it easy: your accounts, documented workflows, runbooks written for the next person. The audit shortlist also doubles as the job description — after a year you know exactly how much automation work actually exists, which beats guessing at headcount from a strategy deck.