Which workflow should your SME automate first?

The best first AI project is rarely the most impressive demonstration. It is the workflow your team already understands, already repeats, and already wishes would disappear.

Choosing well matters because early projects create the organisation’s evidence about AI. A narrow automation that reliably saves five hours each week builds confidence. A broad assistant that produces inconsistent answers can make every later proposal harder to approve.

The seven-question checklist

1. Does the workflow happen every week?

Frequency creates value and learning. A task that takes three hours once a year is usually a poor automation candidate. A 20-minute task repeated every morning may be excellent. Look for intake, classification, reconciliation, reporting, follow-up and status updates.

2. Can someone explain the current process?

If nobody can describe the steps, decisions and exceptions, automation will expose that ambiguity rather than solve it. Name one process owner who can provide real examples and decide what the correct output looks like.

3. Is there a clear input and output?

Strong candidates have observable boundaries. An input might be an email, form, PDF or spreadsheet row. The output might be a task, approved record, notification or dashboard update. “Help the team work smarter” is not a usable boundary.

4. Is the decision explainable?

AI can classify and extract information, but the business still needs rules for confidence and review. Define which cases can proceed automatically, which require a person, and what should happen when required information is missing.

5. Can you provide representative samples?

A demo built on ideal inputs proves little. Collect normal examples, unusual examples and known failures. Remove personal or sensitive information where possible, and make sure the person testing knows the real operating context.

6. Can the result be measured?

Choose one or two outcomes: handling time, response time, error rate, backlog, completion visibility, or time to produce a report. Capture the baseline before building. Without it, a project may feel modern without becoming valuable.

7. Can it be contained?

A first sprint should touch as few systems as possible. One input, one decision flow and one destination is a good shape. If the idea requires a new company-wide data model, five department approvals and replacement of a core platform, split it.

Three common first projects

• Request intake: extract details from messages or documents, identify category and urgency, and create an owned task.
• Management digest: pull stable data from existing sources and deliver a consistent daily or weekly summary.
• Record control: replace a shared spreadsheet that suffers from duplicate entry, unclear ownership or missing history.

What to avoid in the first sprint

Avoid high-stakes autonomous decisions, vague internal chatbots, and projects whose value depends on every employee changing behaviour at once. Also avoid selecting a workflow only because a vendor has a compelling demo. Start from the operational cost, then choose the technology.

The goal of the first project is not to “adopt AI.” It is to make one piece of the business work measurably better and establish a safe, repeatable way to build the next improvement.