How to hire an AI automation expert

First, work out what you are actually hiring for

"AI automation expert" covers at least four different jobs, and hiring the wrong one is the most expensive mistake available to you. The distinctions matter because each has a different rate and a different failure mode.

Most UK small businesses asking this question need the first or second row. If someone in the first row quotes you like the third, or someone in the third row is scoped onto a job the first row could do in two days, you are overpaying by an order of magnitude. A quick read of what AI automation actually is will tell you which row your problem sits in.

The three hiring routes, honestly compared

The in-house column deserves a blunt note. A permanent automation hire only pays for itself when there is a continuous queue of work. Most SMEs have a burst of five or six valuable automations and then a long quiet period of small tweaks. Hiring a salary against that shape is how businesses end up inventing automation projects to justify a headcount. Revisit it once the queue genuinely never empties.

The seven questions to ask

Every one of these is designed so that a competent person answers immediately and a demo merchant changes the subject.

1. "Whose account will the workflows live in?" The only correct answer is yours. Anything else is worth ending the conversation over. This is covered again below because it matters more than price.
2. "What happens when this workflow fails at 3am?" Listen for specifics: error routes, retries, a notification going to a named human, a way to replay the failed run. Vague reassurance means they have not built anything that ran long enough to fail.
3. "Which platform, and why that one for this job?" The reasoning matters more than the answer. Somebody who recommends the same platform for every problem is selling what they know rather than what fits. Our own reasoning is set out in Make vs Zapier and n8n vs Make.
4. "What will this cost to run per month, after you have finished?" They should be able to estimate platform and AI usage. If they cannot, they have not thought about how many steps their design burns per run, which is the difference between a £9 and a £129 monthly bill on the same automation.
5. "Show me something you built that is still running." Not a screen recording of a workflow being built. Something in production, with an owner who can be asked about it. Under NDA, a redacted description of the process and its failure rate is a fair substitute.
6. "What in my list would you tell me not to automate?"The single most revealing question. Anyone who says everything on your list is a great candidate is selling. Real answers sound like "that one runs four times a month, the automation will cost more than the time it saves".
7. "What do I get at the end, besides a working workflow?" Documentation, a diagram, credentials in your own vault, and a handover session. If the deliverable is only the workflow, you have bought a dependency rather than an asset.

Red flags

• The portfolio is all builder screenshots. Anyone can drag modules onto a canvas. Ask what it does and how often it runs.
• They also sell a course on becoming an AI automation agency. Not automatically disqualifying, but ask which is the main business. UK search data for this term is dominated by people wanting to start one of these agencies rather than hire one, and a share of the market is teaching the trade rather than practising it.
• A mandatory retainer attached to a simple build. A well-documented workflow with proper error handling runs unattended for months. Retainers are legitimate when processes change often; retainers on static workflows are rent.
• "AI agents" as the answer to everything. Agents are genuinely useful for open-ended judgement. For "copy this order into the accounting system" they add cost, latency and a new way to be wrong.
• No questions about your data. Where customer records go, which jurisdiction they are processed in, and what the model provider retains are UK GDPR matters. Silence on this is not diligence you want to supply yourself.
• A price before a conversation. A number quoted from a one-line brief is a guess that will be revised upward, or a scope so narrow it will not solve the problem.

The clause that matters more than the price

Automation platform accounts are where the leverage lives. If the Make, n8n or Zapier account is in the contractor's name, then every workflow you paid for sits behind their login, on their card, and your processes stop the day the relationship does. This is common, it is rarely malicious, and it is entirely avoidable.

Insist on all four:

• The platform account is registered to your company, billed to your company.
• The person you hire is added as a collaborator you can remove.
• API keys and credentials for your own tools are created by you and held in your password manager.
• Documentation and workflow exports are handed over on completion, not on request.

None of this is unreasonable, and it costs nothing to arrange at the start. Ask for it after the build and you are negotiating from the weaker position. Everything we build for clients sits in their own accounts for exactly this reason.

What to have ready before you approach anyone

You will get better quotes, faster, from anyone competent if you arrive with these four things. None require technical knowledge.

1. The process written down as it happens today. Who does it, in what order, with which tool at each step. Screenshots are fine. Perfection is not required, and the gaps you discover writing it are frequently the actual problem.
2. How often it runs. Times per day or per week. This single number decides both the platform and whether the automation is worth building at all.
3. The tools involved, and who has admin on each. Access is the most common cause of a two-day build taking three weeks.
4. What the current process costs you. Hours a week multiplied by a loaded hourly rate multiplied by 52. Six hours a week at £25 an hour is a £7,800 a year problem, which tells you instantly whether a £1,500 build is sensible.

If that last calculation does not clear the bar, the correct decision is not to automate it, and anyone worth hiring will say so on the first call. The AI automation cost guide works through the arithmetic in more detail.

How to structure the first engagement

Do not hand a stranger your six most important processes. Structure the first piece of work so that being wrong about them is cheap.

Start with one workflow that matters but is not critical. Something that saves real time and whose failure would be irritating rather than damaging. You learn how they communicate, whether they document, and whether the thing still works in week four, for a few hundred pounds rather than a few thousand.

Agree fixed price on a defined scope. Hourly billing on automation work rewards the slow discovery of complications. Fixed price on a written scope puts the estimating risk with the person who is qualified to estimate.

Define done in advance. Working in production, error handling in place, documented, credentials transferred, owner trained. Write those five words down; they prevent most disputes about whether a project finished.

How we work

We start with an audit rather than a build, because roughly half of what businesses arrive wanting to automate is not worth automating, and finding that out costs less than discovering it afterwards. Work is fixed price against a written scope, built in your platform accounts, documented and handed over. No mandatory retainer.

See what that covers on our AI automation service page, or start with what AI automation actually is if the terminology is still doing more harm than good. If the business is a store, ecommerce automation is the more specific starting point.

Frequently asked questions