Farhan is the operations manager at a 30-person building supplies distributor. After a team workshop, his whiteboard is full of AI ideas: answer customer enquiries, prepare quotations, summarise meetings, sort supplier invoices.

Everyone agrees AI could help. Nobody agrees where to start.

That's the real challenge with a first AI automation pilot. Pick a task that's too big or too risky, and the pilot stalls. Pick the right one, and you quickly learn whether AI can do useful work in your business with less effort. Here's how to choose.

I. Define the task and its finish line

"Automate customer service" is too broad to test. "Draft a reply to delivery-status enquiries using the order record, for a service colleague to approve" is specific enough.

Write down what starts the task, what information it needs and what "done" looks like. Include the handover: a draft sitting in an inbox hasn't answered the customer.

Also check whether you need AI at all. If a step follows a fixed rule, a form or simple automation may be enough. AI earns its place when the task involves reading varied messages, summarising or drafting. Our beginner's guide to AI automation for SMEs explains how these workflows fit together.

II. Score three AI use cases with five questions

Ask the people who do the work to rate each idea:

  • Frequency: Does it happen often enough to test with real examples?

  • Inputs: Is the information current, easy to access and approved for use with AI?

  • Review: Can someone check the result without redoing the whole task?

  • Risk: Can a mistake be caught before it reaches a customer?

  • Ownership: Who will maintain the instructions and handle exceptions?

Back each answer with evidence. "We get about 12 similar requests a week" beats "this could save lots of time". If there's no owner, or no approval to use the data, don't start, however attractive the idea.

III. Choose a low-risk task over an exciting one

Farhan narrowed it down to two ideas. One was letting AI negotiate customer discounts. The other was turning approved product information into quotation drafts for salespeople to check.

He chose quotations. Every draft gets checked before it reaches a customer, and prices and discounts stay under the existing approval rules.

Before going live, test a handful of varied examples: a simple request, one with a missing product detail and one based on an outdated price list. Decide the right outcome for each first. When information is missing, the AI should flag the gap, not fill it.

IV. Measure the whole task, not just the drafting

Record the time spent preparing inputs, drafting, checking and correcting. Track how many drafts are accepted and what errors reviewers catch.

For example, a quotation that used to take 30 minutes might now need 5 minutes to prepare, 3 minutes to draft and 14 minutes to check. That's a saving of 8 minutes, not 27. These figures are illustrative, so measure your own.

If checking takes longer than expected, look at the source information before expanding the pilot.

V. Write a one-page pilot brief

Fill in these six lines with your team before you start:

  • Task: We will test ___ and keep ___ out of scope.

  • Inputs: We will use ___, maintained by ___.

  • People: ___ owns the workflow and ___ approves the output.

  • Baseline: The task takes ___ today, with these common errors: ___.

  • Continue or pause: We'll continue if ___ and pause if ___.

  • Review date: On ___, we'll decide whether to expand, adjust or stop.

Keep the instructions, sample outputs and exception rules together, so the next colleague can run the workflow too.

VI. Common questions

How long should a first AI pilot run? Long enough to see a good number of real examples. For a weekly task, that usually means a few weeks. Set the review date in your brief.

Do we need new software? Often not. Many SMEs start with AI tools they already pay for, plus a shared template. Add integrations only once the workflow works.

Want help choosing and building your first pilot? Explore our AI for Business Automation programme or book a discovery call.

Farhan is an illustrative example based on a typical SME.