Questions to Ask an AI Automation Agency Before You Hire

Before you hire an AI automation agency, ask them to explain in plain words what exactly will be built, who owns it afterwards, where your customers' data goes, what it costs to run each month, and what happens when it breaks. If they can answer all five without jargon, you're in reasonable shape. If any answer is vague, that's the part to push on.

I run an automation business myself, so treat this as a buyer's checklist written by someone who would like to be asked these things.

What exactly will you build, and how will we know it works?

Ask for a one-page scope: what triggers the automation, what it does, what it never does, and who reviews the output. Then ask how success will be measured. A good answer names a number you can check, such as hours saved a week, response time, or leads followed up, not "efficiency".

If you're unsure which process to start with, use a simple scoring method first. Starting with the right small job beats a large vague project.

What will it cost to run, not just to build?

Build cost is the part everyone quotes. Running cost is the part that surprises people. Ask for:

• Monthly fees for the agency's support or retainer
• Software subscriptions and usage-based charges, and who pays them (you or the agency)
• What happens to costs if volume doubles

Then check the maths against your own numbers using the method in how to measure whether an automation is paying for itself.

Who owns it when we finish?

This is the question that most often goes unasked. Get written answers to:

1. Do the workflows, prompts, scripts and configuration belong to you?
2. Are they built inside accounts you own, or inside the agency's accounts?
3. Can you export everything and hand it to someone else?
4. What happens if the agency stops trading or you stop paying the retainer?

The safest setup is that the tools live in your accounts, you hold the admin logins, and the agency has delegated access. It's much easier to hire a replacement later if you can see and control what exists.

Where does our data go?

You are responsible for the customer and candidate data you hand over. Ask:

• Which tools and services will touch our data, and in which countries is it processed?
• Will our data be used to train anyone's models? Ask this about each tool, not just the agency.
• Who at the agency can see it, and is access removed when a project ends?
• Will you sign a written data processing agreement?

That last one is not a nicety in the UK and EU. The ICO's guidance says that whenever a controller uses a processor, there must be a written contract, and that controllers should only use processors that can give sufficient guarantees about security. The UAE has its own personal data law with similar ideas, and free zones such as DIFC and ADGM have separate regimes, so if you handle people's data there, check which one applies to you. This is not legal advice; a short conversation with a lawyer is worth it for anything sensitive. If you want a starting point for your own team, see a one-page AI use policy.

How do you make sure the AI doesn't say or do something wrong?

AI systems make mistakes, and a good agency says so. Ask where a human checks the work before it goes out, what the fallback is when the system isn't confident, and how mistakes are logged. For anything customer-facing, ask to see the failure cases from testing, not only the demo. A review checklist is a good thing to agree on before launch.

What happens when it breaks?

Automations break. A website changes, an account's login expires, a service updates its rules. Ask:

• Who gets alerted, and how quickly will someone respond?
• Is support included, and for how long?
• How will you tell me something has stopped, rather than my finding out from a customer?

Silent failure is the real risk. An automation that fails loudly is annoying; one that fails quietly can go unnoticed for weeks.

Can I talk to someone who uses it?

Ask for a reference you can contact, or a description of a similar project in your type of business. If the agency can't share names because of confidentiality, ask them to describe the situation and the result in enough detail to judge. Be wary of a portfolio that is all screenshots and no outcomes. Look up who is actually doing the work, too: a named person with a track record is easier to hold to account than a logo.

Do you tell me when I don't need this?

A good sign is an agency that sometimes says "this should be a spreadsheet" or "don't automate that yet." Whether you work with us at Just DCA, or anywhere else, notice whether they challenge your idea or just accept it. Our consulting option exists for exactly that: working out what's worth building before you pay to build it.

Limits: when this checklist doesn't apply

• Off-the-shelf tools. If you're simply subscribing to software, you need a vendor review rather than an agency review, though the data questions still apply.
• Tiny, low-risk projects. For a single internal automation with no personal data, a lighter conversation is fine.
• Regulated sectors. Health, finance and legal work often have extra requirements that no generic checklist covers.

A checklist can't guarantee a good outcome, but it makes a bad one much less likely. If you'd like to run these questions past us, get in touch.