A sales coordinator receives the same product questions every week. The answers are spread across a product sheet, a stock system and a colleague’s memory. ‘Use AI in sales’ is an ambition. ‘Draft a reply from approved product information, then have the coordinator check it’ is a task you can examine.
For a Singapore SME choosing a first experiment, that distinction is useful. You need a task with a recognisable input, a result someone can judge and an owner who can stop the process when something is wrong. The distributor example in this article is illustrative, not a client case study.
Write down the work before naming the tool
Ask the people doing the task to describe one recent example. What came in? What did they have to find, write or decide? Where did the work slow down? Sometimes the problem is missing information rather than the effort of writing an answer. A better prompt cannot make an outdated product sheet reliable.
Make a shortlist of three recurring tasks. For each one, record the input, the expected output, the current owner and the person who could check an AI-assisted result. Keep the task specific enough to demonstrate with an approved example.
Use four questions to compare the shortlist
- Usefulness: if this task improved, what would be easier for the team or customer? Name the friction, not an unmeasured savings claim.
- Information readiness: is there a reliable source the team is permitted to use? If the source is incomplete, record the gap first.
- Checkability: can a qualified person compare the output with the source and spot a wrong answer? Include the effort of that review.
- Consequences: what happens if the answer is wrong? A draft for internal review is different from an instruction automatically sent to a customer.
Do not hide a serious concern inside an average score. A task may look useful and easy to demonstrate, yet still be unsuitable because nobody can verify it or the information cannot be used in the chosen tool. Treat those as reasons to pause or redesign the trial.
Turn the choice into a one-page pilot brief
Task: draft replies to a defined set of product questions. Input: an approved, current product sheet and a sanitised enquiry. AI step: prepare a draft and flag missing information. Human step: check product facts, stock and commercial terms. Owner: the named coordinator. Boundary: no automatic sending. Review: record corrections, checking effort and unanswered questions.
Set a review point before starting. Keep examples of acceptable and unacceptable outputs, including cases where AI assistance was not useful. If you measure time, count preparation, review and rework as well as drafting. Decide from that evidence whether to refine the approach, stop or consider a wider scope.
What to bring to the first conversation
Bring your shortlist, the process owner and a description of the information involved. Do not email confidential customer records to demonstrate the problem. Synthetic or approved non-sensitive examples are enough to discuss a starting point. EverX’s AI Readiness & Opportunity Workshop is designed to turn that conversation into an opportunity map and a first-pilot action brief.
Further reading
IMDA’s enterprise AI guidance includes identifying use cases and scoping deployment. The questions and example above are EverX’s practical discussion format, not an official assessment or an indication of funding eligibility.