There are two conversations about AI. The conference one — autonomous agents, total transformation, five-year promises. And the one inside real businesses: overloaded teams, customer enquiries left waiting, documents to process, unexploited data. Only the second one interests us.
The test of a good use case
A good first AI project has four traits: the task is repetitive and frequent; it follows rules a human can explain; errors are recoverable; and the gain is measured in hours or in customers served. Answering frequent questions, sorting inbound requests, extracting data from documents, summarizing files: that's the terrain.
- Repetitive and frequent — otherwise the gain doesn't justify the effort
- Explainable — if nobody can describe the task, AI won't guess it
- Recoverable — sensitive decisions keep human validation
- Measurable — a gain you can't quantify doesn't exist
Integrated, or ignored
The second condition of success: AI must live where the work already happens — in the CRM, on WhatsApp, in the inbox, in the business tool. Every additional app to open divides adoption. The AI projects that succeed are invisible: the team simply notices the task is done.
Good enterprise AI doesn't get noticed. It gets measured.
Starting small isn't thinking small
An assistant answering frequent questions on WhatsApp, automatic sorting of enquiries, data extraction from invoices: each of these projects is scoped in days, built in weeks and measured from the first month. That short loop — scope, ship, measure, extend — is what builds a real AI capability inside a company. Not the spectacular six-month project that ends as a demo.
