The AI consultant vs in-house hire decision comes down to one question for a 200-person company: do you already know which work to automate? If you know the processes and need someone to run them every day for years, hire. If nobody knows yet where AI earns its place, bring in an outside team first and hire once there is a running system to own.
The reason is the calendar. A technical role takes a median of 76 days to fill from the day it opens, and the new hire then has to learn where your hours go before automating any of it, while our own engagements start the diagnosis on the first call.
The short version
- Hire in-house when you already know which processes to automate and need someone to run them every day.
- Bring in an outside team first when nobody knows yet where AI earns its place, then hire once a running system needs an owner.
- A US AI engineer starts at $175,000 in the middle of Robert Half's range, ~$255,000 a year with average benefits, and a technical role takes a median 76 days to fill. Published consultant retainers run $2,500 to $15,000 a month.
- A solo consultant fits an audit or one defined build. An AI automation company fits work that spans departments and has to keep running.
AI consultant vs in-house hire: the three options compared
A solo AI consultant, an AI automation company and a full-time hire are three different purchases, and each one fails in its own way. Lumping the first two together as "outside help" hides the difference that matters most: one person or a team.
A solo consultant is one person, billed by the hour or by the project: Layer3's 2026 guide puts freelance AI consultants at $75 to $150 an hour and keeps the $500 top of its range for enterprise firms. That fits an audit or one defined build. When the project ends, the knowledge leaves with them unless someone wrote it down.
An AI automation company sends a team that covers engineering, data and process design, and it can stay to run what it built. Published retainers run $2,500 to $15,000 a month, so $30,000 to $180,000 a year, with active development only in the top tier, and our guide to AI automation consulting cost breaks the ranges down by engagement shape. The risk here is a team that sells AI where a plain script would do the job.
An in-house hire knows your company and stays, and nobody outside can match that. At 200 people, the first AI hire is the whole AI function, so the risk is scope: one person ends up covering every department, and when that person leaves, the knowledge goes with them.
| In-house hire | Solo AI consultant | AI automation company | |
|---|---|---|---|
| Time to start | Median 76 days to fill a technical role, then time to learn the company | When the contract is signed | When the contract is signed; ours starts with a diagnosis call |
| Cost (US) | $175,000 mid-level starting salary, ~$255,000 a year with average benefits | $75 to $150 an hour for a freelancer | $2,500 to $15,000 a month on retainer, $30,000 to $180,000 a year |
| What stays afterwards | The person, for as long as they stay | Whatever was written down | The systems, what was written down, and the team if you keep the retainer |
| How it fails | One person covers every department, then leaves | Leaves at the end of the project with the knowledge | Sells AI where a script would do, or writes nothing down |
| Best when | You know the processes and need a daily owner | One audit or one defined build | The work spans departments and nobody knows yet where AI fits |
What an in-house AI hire costs, and how long it takes
An in-house AI engineer costs a salary plus benefits from the first day, and the seat stays empty for months before that.
Robert Half's 2027 salary guide puts the US starting salary for an AI engineer with moderate experience at $175,000, the middle of its $154,500 to $196,750 range. Benefits come on top: for full-time private-sector workers they are 31.5% of what employers pay, per the Bureau of Labor Statistics' June 2026 figures.
By our arithmetic, a $175,000 salary at that benefits share costs ~$255,000 a year. The share is an average across all full-time workers, so treat the total as an estimate rather than a quote.
The calendar matters as much as the money. Ashby's 2026 hiring data, across 247,000 jobs, puts the median at 76 days from opening a technical role to making the first hire, against 56 days for business roles. Add the new hire's notice period, then the time they spend learning which processes exist, who owns them and where the data sits.
My answer: an outside team first, then a hire
The first months of AI work are diagnosis, so I would bring in an outside team first and hire the owner once systems exist.
Diagnosis means finding where the hours go, deciding which fixes need a model at all, and agreeing who approves what the system does. My rule for the middle step: deterministic and repeatable means a script, judgment and language means a model. I think most items on a company's AI list are scripts, and that is cheaper to learn in week one than in month six.
Diagnosis goes faster with people who have already done it inside other companies. The hire who comes after inherits written instructions, test cases and a number instead of a blank page, and the job description finally names real systems instead of "do something with AI".
Hired the other way round, the new person has to run the diagnosis alone before anything is built.
What this looked like for one client
Our client, a creator-marketing company, exchanges thousands of messages a week with creators. When we came in, an engineer at a partner firm already owned the software the team works in, and the client's day-to-day lead had their own written instructions for AI messaging. We replaced neither of them.
We took one step of an 8-step checklist: reply to one creator message, ~5 minutes for a person and 1 turn for an agent. The engineer's platform runs it, we own the agent's written instructions, the test set and the number, and a team member presses send on every draft.
The goal is 90% of replies sent with no human edit, and two things must stay at zero: a product claim sent to a creator, and an account getting limited. The rule for autonomy has three rungs: a person sends everything, then spot-checks, then the agent runs alone, moving up only when the test set and last week's number clear the bar.
The plan adds a Friday review where a group from the client, the partner firm and us approves or passes 5 proposed changes. One mistake during the preparation: we asked for a second explanation of a workflow already on a recorded call, so we now prepare from the recordings and written answers and ask only about the gaps.
That is the shape I would copy: the outside team brings method and measurement, the people inside keep the judgment, and the split is written down before anything is built.
Where this answer does not hold
We sell the outside option, so weigh this answer with that in mind. There are three cases where I would hire first.
Hire first when the AI work is your product, because then the knowledge is the business and belongs on your payroll. Hire first when your data cannot leave your own systems, because an outside team's speed depends on access. Hire first when you already have engineers and a short, known list of processes, because the diagnosis is done and what is left is daily ownership.
An outside team also fails in its own way: when nobody inside owns the result. If no one on your side approves the output and watches the number, the system drifts as soon as the outside team steps back.
The salary and rate figures above come from a staffing firm and a consultancy, so use them as a sanity check rather than a market rate.
The question to ask before either
Before you post a job or sign a consultant, write down where the work will live in 12 months and who will own it.
Four answers cover it: where the AI's written instructions live, who keeps the test cases that prove it works, who owns the number it is measured on, and who presses send until that number clears the bar. The test cases matter because nothing should grade its own output, including the person who built it.
A hire who cannot answer those four questions and a consultant who will not are the same risk. Getting the answers down for your own operations is the first thing we do when we work inside a company.
Key takeaways
- Compare the options in the same unit: a hire costs a yearly salary plus benefits, a retainer a monthly fee, and a freelancer an hourly rate.
- Count the calendar as a cost: the weeks a technical role sits open, then a notice period, then the time a new hire spends learning where the hours go.
- Expect most items on a company's AI list to be scripts, and question anyone who prices every item as AI.
- Before anything is built, write down where the AI's instructions live, who keeps the test cases, who owns the number and who presses send.
- An outside team needs an inside owner who approves the output and watches the number, or the system drifts when the team steps back.
Common questions
Should a 200-person company hire an AI engineer or an AI consultant first?
Hire first if you already know which processes to automate and need a daily owner. If you do not, bring in an outside team first: technical roles take a median 76 days to fill, and the first job is diagnosis, not building.
How much does an in-house AI engineer cost compared with an AI consultant?
Robert Half's 2027 guide puts the US starting salary for an AI engineer with moderate experience at $175,000, ~$255,000 a year with average benefits. Layer3's published rates put freelance AI consultants at $75 to $150 an hour and enterprise firms up to $500, with retainers at $2,500 to $15,000 a month.
What should a company own when an AI consultant leaves?
The written instructions the AI follows, the test cases that prove it works, the number it is measured on, and access to every system it runs in. If those live only in the consultant's head, you are renting the work.