How to choose an AI automation agency: questions to ask and red flags to avoid

Searches for "AI agency" and "AI automation agency" have grown quickly, and so has the number of agencies. Some are excellent. Many are a few months old, rebranded from something else, and built around reselling one tool. As a buyer, it can be hard to tell the difference from a website alone.

This guide is written for business owners and managers who are about to hire someone to build automation or AI into their operations. It's the checklist we'd want a client to use on us.

Start with the problem, not the agency

Before speaking to anyone, write down one process you want to improve. Keep it concrete:

• What happens today, step by step, and who does each step?
• How often does it happen, and roughly how long does it take?
• What goes wrong — missed follow-ups, typing errors, slow replies?
• What would a good result look like in three months?

This does two things. It makes every agency conversation more useful, and it gives you a way to judge them. A good agency will ask about these details before talking about technology. One that jumps straight to "we'll build you an AI agent" without understanding the process is telling you how the project will go.

Questions to ask an AI automation agency

1. Can you show me systems you've built that are live today?

Not mock-ups, not a demo video — real systems, running for real businesses, that you can visit or that the client will talk about. Ask what the agency built specifically, as opposed to what came from an off-the-shelf tool.

For example, our own case studies link to live systems: an internal workflow portal for Ladhawala & Associates, a compliance firm, and a full ordering and admin platform for Kathiyawadi Kitchen, a London tiffin and catering service. Our own product, TaxPilotDesk, is clearly labelled as a working demo rather than presented as a customer success story. That's the level of clarity you should expect from anyone.

2. How will you work out what to build?

Listen for a discovery step: mapping the current process, identifying which steps are rule-based and which need judgement, and recommending the simplest thing that works. If the recommendation is AI for everything, ask why each step needs it. Often, plain workflow automation is cheaper and more reliable — we explain the difference in AI agent vs chatbot vs automation.

3. What will I get in writing before work starts?

You should receive a written scope covering what will be built, what won't be built, how it will be tested and what "finished" means — plus a price. Vague scopes lead to arguments about what was included and invoices that keep growing.

4. How is it priced, and what are the running costs?

There are two separate costs and you need both. The build cost is what the agency charges. The running cost is what you pay each month for the platforms and AI usage the system depends on. Ask for an estimate of running costs at your expected volume, and ask whose account they're billed to. If the agency adds a markup by billing AI usage through its own account, you should know that upfront.

Some agencies charge a fixed one-time price for the build; others charge a monthly retainer. Neither is wrong, but understand what happens if you stop paying a retainer: does the system keep running, and do you keep it?

5. Who owns the code, the accounts and the data?

This is the question most businesses forget to ask until it's too late. At the end of the project you should know:

• Where the code lives, and whether it's in a repository under your account.
• Whose name the hosting, automation platform and AI provider accounts are in.
• Where your data is stored and whether you can export it.
• Whether you'd need the agency's permission — or payment — to change or move the system.

We build on accounts in the client's name and hand over the code, access, documentation and training, so the system is the client's to keep. Whoever you choose, get the ownership terms in the contract.

6. Where won't AI work for my business?

This is the best single test of an agency's honesty. Every experienced builder knows where AI is unreliable: decisions with legal or financial consequences for people, tasks that need guaranteed accuracy, and processes where the input data is a mess. If an agency can't name a single case where they'd advise against AI, they're selling, not advising.

7. How do you handle mistakes and monitoring?

Automations fail — an API changes, a password expires, an input arrives in an unexpected format. AI agents occasionally get things wrong. Ask how failures are detected, who gets alerted, and how errors are logged. A system that fails silently is worse than no system, because you assume work is being done when it isn't.

8. How do you handle personal data?

Most business processes involve personal data: customer names, emails, phone numbers, documents. If you serve UK customers, the UK GDPR applies; if personal data is processed in India, the Digital Personal Data Protection Act 2023 applies. A competent agency should be able to explain, in plain terms:

• Which third-party services will process your data, including the AI provider.
• Where that processing happens, and how international transfers are covered.
• Whether the AI provider uses your data to train its models (many business plans don't, but you should check).
• What access the agency itself keeps after handover.

You remain responsible for your customers' data, so you need these answers whoever builds the system.

9. What happens after launch?

Ask what's included at launch (monitoring, fixes for defects, training) and what ongoing support costs. Support should be optional and agreed in writing, with clear response times — not a condition of the system continuing to work.

Red flags to watch for

Does it matter where the agency is based?

Less than it used to. Many agencies, including ours, work remotely with clients in other countries. What matters more is overlap in working hours for calls, clear written communication, familiarity with the regulations your business operates under, and experience with the tools common in your market — for example, local tax and compliance workflows, or the accounting and CRM systems popular in your country.

If the agency is in another country, check how they handle international data transfers under the data protection law that applies to you, and confirm the contract is clear about governing law, currency and payment terms. Whatever the location, check that they understand the practical realities of your sector.

A simple scoring sheet

After each conversation, score the agency from 1 to 5 on each of these. The total matters less than any score of 1 or 2, which is usually a deal-breaker.

1. Understood my process before proposing a solution.
2. Showed live systems they built.
3. Offered a written scope and a clear price.
4. Explained running costs and whose accounts they'd be on.
5. Confirmed I'd own the code, accounts and data.
6. Named cases where they wouldn't recommend AI.
7. Had a clear plan for monitoring and mistakes.
8. Answered data protection questions clearly.

What a good first project looks like

The safest first project is small: one process, one clear outcome, a fixed scope and a few weeks of work. It lets you judge the agency on real delivery before committing to anything larger. When Ladhawala & Associates started with us, the work focused on the firm's public site and the internal portal it uses for client work — and the firm's own estimate is that the workflow automation now saves the team 20–30 hours a week. That came from understanding one firm's process well, not from a large AI programme.

If you're comparing agencies and would like to include us, book a discovery call. Bring one process; we'll tell you honestly whether it's worth automating, what we'd build and what it would cost — and if we're not the right fit, we'll say so.