Red flags when hiring an AI automation agency

The clearest red flags when hiring an AI automation agency are these: they talk about tools before they ask about your process, they cannot name a fixed deliverable, they have no maintenance plan, they keep ownership of the accounts they build on, and their case studies cannot survive one follow-up question. Any one of these should slow you down. Two or more should stop the conversation.

I build automations for UK B2B service firms every week, so I see the aftermath of bad engagements: half-finished workflows, credentials nobody holds, and invoices that kept arriving after the value stopped. This piece is the checklist I wish those buyers had used. It sits alongside the broader AI Automation for B2B: what actually works guide, which covers what to automate in the first place.

Why is "tools first" the biggest warning sign?

If the first call is about which platform they love — n8n, Make, Zapier, some proprietary wrapper — rather than which of your processes bleeds time or leads, you are watching a solution hunt for a problem. Tools are interchangeable. Your bottleneck is not.

A competent agency starts by asking what happens when a lead comes in, who touches it, where it stalls, and what a stalled lead costs you. Only then does the tool question make sense. When an agency cannot describe your process back to you in plain language, then it cannot automate it — it can only decorate it.

What should the deliverable look like?

Vague scope is where budgets go to die. "Ongoing AI transformation" is not a deliverable. "A lead-routing workflow that moves every form submission into the CRM, assigns an owner, and triggers a first-touch email within five minutes" is a deliverable. You can test it. You can point at it when it breaks.

Ask three questions before signing:

• What exactly will exist at the end that does not exist now?
• How will we both know it works — what number moves?
• What happens at the end of the engagement — handover, documentation, training?

If the answers are fuzzy, the invoices will not be. Fixed-price, fixed-scope work forces the agency to think before it builds. Open-ended retainers with no defined output reward the opposite. This is one of the five usual suspects I cover in why automations fail — unclear ownership of outcomes kills more projects than bad code does.

Who owns what you are paying for?

This is the quiet trap. Some agencies build everything inside their own accounts: their Zapier workspace, their OpenAI keys, their sending domains. The work functions, you pay monthly, and the day you stop paying, everything vanishes — along with the data inside it.

Insist on the following in writing: workflows built in your accounts or transferred to them at handover, credentials held by you, documentation that lets another competent builder pick the system up. An agency that resists this is not selling you a system. It is renting you a dependency. The same ownership logic applies whether you build in-house or buy — I set out the full trade-offs in build vs buy: sales systems for service firms.

What does a sound engagement actually look like?

Here is the mechanism a credible agency follows, step by step. When they receive an enquiry, then they run discovery on your current process before quoting. When discovery shows a process worth automating, then they map it — trigger, steps, exceptions, owner — on paper first. When the map is agreed, then they build against a fixed scope with a named acceptance test. When the build passes that test, then they hand over credentials, documentation, and a maintenance option you can decline. When something breaks later, then there is a defined response, not an emergency upsell.

Every "when" has a "then" you can verify. If the agency's process has no such chain — if it goes straight from enquiry to invoice — that is your answer.

Which questions expose a weak agency fastest?

Three questions do most of the work. First: "Show me a system you built two years ago that still runs." Automations rot; anyone can demo something built last week. Second: "What did your last failed project teach you?" An agency that has never had one is either new or lying, and neither is reassuring. Third: "What would you not automate in my business?" A good builder will name several things — judgement calls, pricing, anything customer-facing without review. Someone who says AI can handle it all is selling the hype, not the work.

None of this means agencies are the wrong route. A good one earns its fee many times over. But the burden of proof sits with them, and firms that grew on referrals alone often lack the practice to run this scrutiny — worth reading the maths on referral-only pipelines if that sounds familiar. Vet the process, fix the scope, keep the keys.

---

Next step: the Growth System Audit — £450, seven days, credited against any build — maps where your growth system leaks and what to build first.