How to choose an AI automation agency: six questions to ask

What should you ask any agency?

Put these to every agency on your shortlist, in writing, and compare the replies side by side.

1. Who owns the code and the accounts? A good answer: it is built in your accounts, and you own the code and the data. A weak answer: "we host it for you", with no word on how you would leave.
2. How is the price fixed? A good answer: a written scope with a price, agreed before the build starts. Be wary of an exact quote from someone who has not yet looked at how your business runs.
3. What happens after launch? A good answer names training, documentation and a support option with a notice period. A weak answer goes quiet after the word "launch".
4. How are AI answers limited and tested? A good answer: the AI only answers from information you approved, says when it does not know, and is tested on awkward cases before a customer meets it.
5. Where does a person stay in the loop? A good answer lists the decisions the AI never makes alone, such as money and complaints.
6. What proof can you show, and what kind is it? A build they made, a result the client reported and a result somebody measured are three different things. A good agency tells you which one you are looking at.

Our guide to hiring someone to build an AI system goes deeper on the first three.

How does Agentryx answer the same six questions?

Here are our own answers, so you can hold us to each one.

Question	Our answer
Ownership	Everything is built in your own accounts. You own the code and the data.
Price	A paid audit comes first. Before it we give a range, never an exact price. After it you get a written scope and price, paid 50% upfront and 50% on launch by default.
After launch	We launch in stages, train your team and hand over documentation. Ongoing support is optional, as a monthly retainer.
AI limits and testing	AI systems are limited to approved information, say when they do not know, hand over to a person and tell people they are automated. We test real cases, edge cases and failure paths before anything touches a customer.
Person in the loop	A person reviews money, legal commitments, messages sent in your name, complaints and regulated advice.
Proof	MRM Group is build details only, with no results claimed. TMMB Academy is a client-reported result. We label which is which.

On proof, judge us on what we built and on how we scope. Our approach page sets out each stage, and our AI automation agency page covers what we build.

Why do the two AI questions matter so much?

Because your business answers for what its AI says. In Moffatt v Air Canada (2024), a British Columbia tribunal held the airline liable for incorrect fare information its website chatbot gave a customer. The airline argued the chatbot was responsible for its own actions, and that argument was rejected.

So ask to see the test plan, not a demo. A demo shows the AI on its best day. Our guide on how to test an AI agent before customers lists the cases any agency should run.

What are the red flags?

• It lives in their accounts. If you stop paying, the system goes with them.
• An exact price on the first call. Nobody can price a build before they know your tools and where work gets stuck.
• A guaranteed result or payback time. An honest agency shows you a sum with your numbers, not a promise.
• "It can answer anything." An AI with no limits is the kind that ends up in a tribunal ruling.
• Proof with no label. A big figure with no word on who measured it is marketing.
• No handover plan. If nothing is written down, you depend on them for good.

What is the problem worth? Do the sum before you buy

A good agency does this sum with you before it quotes. The three AI solutions we build each pay in a different way.

• Conversational AI and chatbots answer written enquiries from approved information at any hour and hand over to a person. Lever: more clients.
• Voice agents answer the calls you would otherwise miss, ask your qualifying questions and book. Lever: more clients.
• AI knowledge assistants answer your team's repeat questions from your own documents. Lever: time back, which has to end in pounds.

Take the chatbot. The leak, in an owner's words: "People message us at night and at weekends, and nobody replies until Monday." The system answers from approved information, takes the person's details, books or hands over, and logs the conversation for your team.

Line	Example figure	Running total
Out-of-hours enquiries a month	15
Share you lose today because the reply came too late	1 in 5	3 clients a month
Value of a client	£800	£2,400 a month
Months in a year	12	£28,800 a year

Example figures. Swap in your own. We made these numbers up to show the method. They are not a result, a benchmark or a forecast.

For your own number, count the enquiries that arrived outside opening hours last month and how many of those people never became clients. Then multiply by what a client is worth, or work out your own figure with our ROI calculator. Bring it to a call with any agency: our audit report prices each problem for a year.

When you do not need an agency at all

• The sum is small. If the leak costs less each year than a build, leave it.
• Your current software already does it. Check the settings of your CRM and booking system before paying anyone.
• The process is not settled. AI on top of a process that changes weekly gives you a faster mess.
• You have a capable person in-house. One simple automation between two tools does not need an outside team.
• Your information is not written down. An AI can only answer from what you give it. Write the answers first.

What does it cost with us?

The paid audit is £3,000 plus VAT (US $5,000). You get a written report within 10 working days, and the fee is credited against the build when the build deposit is paid within 14 days of the report. Our AI audit pricing page explains what the report contains.

Scoped builds run from £5,000 to £20,000 (US $10,000 to $30,000), with the timeline agreed in the written scope. Run & Improve support is optional at £1,000, £2,000 or £4,000 a month. Hosting, phone minutes and AI usage are paid directly to the providers from your own accounts.