Test an AI use case in one business process

You have a specific workflow, an owner and a way to test results on a limited set of tasks. AI might help draft lead responses, identify a reason to repurchase, suggest an additional offer, support partners, process feedback or assist payment follow-up.

Start with one use case. If the problem is still unclear, begin with a consultation.

Pilot brief: the problem, users, workflow boundaries, owner and intended outcome.

Test plan: a baseline, test tasks, quality criteria, effort and human review rules.

Conclusion: observations, errors and limitations, with a decision on the next experiment. Sales growth or savings are not guaranteed: the pilot tests a hypothesis.

An initial discussion needs only a workflow description, without customer databases, passwords or trade secrets. Use anonymised or synthetic examples where possible.

Before using working data, agree on its scope, lawful basis, storage, access, deletion and permitted AI providers. A WhatsApp message does not itself authorise using company data in AI.

1. Problem. Clarify the workflow, owner and usefulness criteria.

2. Test design. Agree on data, baseline, use case and quality checks.

3. Limited test. Compare AI-assisted work with the current approach; record observations and errors.

4. Decision. Review evidence and choose the next step. Timing, price and scope are agreed after defining the pilot boundaries.

Before testing, agree on what counts as useful and which errors are unacceptable. Afterwards, compare quality, speed, human review effort and risks. If evidence is insufficient, state the uncertainty and define a further test.

Use case: AI drafts a reply to an inbound request; a team member reviews and sends it.

The brief records: request type, permitted sources, prohibited promises, completeness criteria, preparation time and corrections. This illustrates a pilot structure; it is not a reported client result.