To choose an AI automation consultant, start with one task you want off your team’s plate, then judge each consultant on how they would handle it. A good one scopes the task in writing, prices it clearly and proves it on a small first project with tests you agree. They also tell you who holds the accounts and the data. Where they are based matters less than how they work, so a remote consultant can be the right choice.
Start with one task, not a tool
A consultant is easiest to judge when you bring a specific job. Before the first call, write down one task your team repeats every week:
- What starts it: a new enquiry, an invoice arriving, a job marked done.
- The steps a person follows today, in order.
- The apps it touches: your inbox, calendar, accounting software or customer list.
- Who checks the result before it reaches a customer.
- How often it happens each week.
Give every consultant the same page. A strong one asks follow-up questions about the steps and the checks before talking about tools. If a proposal could have been written without reading your page, it probably was. Our list of AI automations worth starting with has tasks that suit a first project.
Fixed price, retainer or hourly: how should you pay?
There are three ways to pay a consultant, and they can be combined.
| How you pay | How it works | Suits | Ask about |
|---|---|---|---|
| Fixed price | One price for an agreed scope | A defined task with clear steps | What counts as a change, and what a change costs |
| Monthly fee | A set fee each month for running and fixes | Automations that must keep working as your apps change | What the fee covers, the minimum term and the notice period |
| Hourly | You pay for the time used | Advice, reviews and small changes | A ceiling on hours each month |
A fixed price for the build, then a monthly fee for running and upkeep, is a sensible mix. Both should be in writing before work starts. The AI usage can be a separate cost again. For example, Claude API usage is billed separately from paid Claude plans. Ask whether the monthly fee includes it or whether you pay the provider. Our guide to what AI automation costs breaks down each part.
Why a small first project should come first
Before you commit to a large build, ask for one automation that works on your real work. Agree the tests before the build starts: real examples, the result each one should give, and what must go to a person instead. The automation goes live only when those tests pass.
Start small, then go live on agreed tests
1Write down one task
Its trigger, steps, apps and who checks the result
2Agree scope and tests
In writing, with real examples and the result each should give
3Build a small first project
One automation, working on your real work
4Go live when tests pass
Tie part of the fee to this step
The tests are agreed before the build starts, not after.
Tie part of the payment to the tests. Paying some of the fee at the start and the rest when the agreed tests pass keeps both sides focused on the same result. Our guide to stopping an AI assistant giving wrong answers shows how to write tests for anything that talks to customers.
A first project also shows you how the consultant works. You see how quickly they reply, how clearly they explain problems and whether they keep to dates. That tells you more than a sales call.
Who should hold the accounts and the data?
Agree whose name every account is in, and what happens to your data when the work ends, before anything is built. There are two sound ways to set it up.
- In your own accounts. The AI, automation and app accounts are in your name. You pay each provider and see every charge, but you also manage the bills and logins.
- Hosted by the consultant. The consultant runs the automation on their own systems for a monthly fee. There is less for you to manage, but the automation stops if the fee stops.
Neither is wrong. What matters is knowing which one you have. Your customer and business data should stay yours either way, and come back to you in a format you can use if you leave. Ask for that in writing.
How to give a remote consultant safe access
Remote work runs on video calls, shared documents and access to your tools. Agree a regular call, a reply time for messages and who to contact when something stops. Then set up access with care, because that is where the risk sits.
1. Give each person their own login
Never share your own password. Where your software allows it, invite an outside person as their own user or guest. In Microsoft’s business accounts, for example, guests sign in with their own credentials while you keep control of your data.
2. Give only the access the work needs
This is the principle of least privilege: people get access only to the data and actions their job requires. On the Claude Console, for example, the Developer role can manage API keys and see usage and costs, while only an Admin can manage users and their roles. Keep the Admin role yourself.
3. Keep each project separate
A separate space for each project keeps its costs and access apart from everything else. Claude Console workspaces can each have their own members, API keys and limits, so a consultant’s project can sit in a workspace of its own.
4. Remove access when the work ends
List every login you give out, and take each one away at handover unless the consultant still runs the upkeep. Microsoft’s guidance on access reviews names invited guests who were not removed after an administrative task as something to check for.
Questions to ask an AI automation consultant before you hire
Ask every consultant the same questions, and compare the answers side by side.
- Which part of our task would you automate first, and why?
- What do you need from us, and by when?
- Which steps run on their own, and which wait for a person to approve?
- How will we test it before it goes live?
- What happens when one of our apps changes, or an AI model is retired?
- What do we get when the work ends: our data, the documents and the logins?
- Can we see past work, or speak to a past client?
Take care if a consultant promises time savings before seeing your work, or will not put the scope, tests and running costs in writing. The same goes for anyone who asks for your own password or full admin access to everything.
We are based in Singapore and work remotely with clients worldwide. Our AI systems are set up around your business, then hosted and looked after by us. Half the set-up fee is due to start, and the other half when the tests we agreed together pass. Your data is yours to export at any time. Work that does not fit those systems is a custom AI solution, priced after a free 30-minute call.