AI leverage · Field note
Where AI actually fits in a small business
The best AI project is rarely the most impressive demo. It is usually a repetitive decision, handoff, or task that quietly consumes time every week.
Start with friction, not tools
Most teams begin by asking which AI product they should buy. That reverses the useful order. First list the work that is slow, repetitive, inconsistent, or dependent on one person remembering every detail. Then decide whether AI belongs in that workflow.
Good early candidates include summarising customer conversations, drafting first responses from approved information, classifying enquiries, turning meeting notes into actions, and preparing a first version of recurring reports. These tasks have clear inputs and outputs, and a person can review the result before it affects a customer.
Use three filters
- Frequency: Does the task happen often enough to matter?
- Clarity: Can you explain what a good result looks like?
- Risk: Can a person review the output before anything irreversible happens?
If a workflow scores well on all three, it is a stronger starting point than a large autonomous system. You can measure it quickly and learn without putting the business at risk.
Design the smallest useful system
A practical AI workflow normally needs more than a prompt. It needs a trigger, reliable context, an output format, a review step, and a record of what happened. For example, a new enquiry can trigger a summary and suggested reply, but the reply should use your service information and wait for approval before sending.
Keep the first version narrow. One team, one workflow, one measurable result. Track time saved, turnaround time, error rate, or conversion quality. If the workflow improves the number that matters, expand it. If it only creates novelty, stop.
Protect trust
Do not send sensitive customer data into unapproved tools. Do not let generated claims reach customers without review. Record where AI is used, who owns the workflow, and how a person can correct it. Useful automation should make responsibility clearer, not blur it.
Write down the five recurring tasks your team complains about most. Choose the one with high frequency, a clear definition of done, and a reversible output. That is usually where AI belongs first.