Positioning an AI Feature

For the past two years nearly every client has asked whether AI can be added. Most of the time we suggest waiting and instead finding one specific, repetitive task with a high tolerance for error, then validating it internally first.

Three good starting tasks

Text structuring: turning meeting notes or ticket descriptions into structured fields that a human reviews.

Information extraction: pulling key data from contracts, invoices and résumés to cut re-typing.

Draft generation: support replies, product descriptions and weekly reports, so users do not start from a blank page.

Three situations to approach carefully

Direct automated replies to customers, where errors are expensive and a human fallback is mandatory.

Money calculations and compliance judgement, where the model may assist but never decide.

Tasks without a clear evaluation standard; if you cannot tell good from bad, do not start.

An AI feature matters when it genuinely removes repetitive work from a specific role, not because the technology is advanced. Without that point, skip it.