How do you choose the first AI use case to build with a customer?

30-second answerSay your answer out loud first, then reveal.

Score candidates on business value (volume × time or cost saved, revenue impact, strategic importance) and feasibility (data availability and quality, task suitability for AI, integration complexity, risk of errors, stakeholder support). Pick a high-value, high-feasibility, low-risk use case with a clear metric and a motivated champion as the first win. Save risky or complex ones for later, once trust exists.

A value versus feasibility quadrant placing contract clause extraction and support ticket triage in Start here, autonomous loan approvals in Plan carefully, an internal FAQ bot in Quick wins and demand forecasting via LLM in Avoid.

Good first use case characteristics

High volume, repetitive, language-heavy work (reading, summarising, extracting, drafting, routing).
Human-in-the-loop friendly: the AI assists and a human approves, so errors are cheap.
Measurable baseline: you know the current time per task or error rate.
Data accessible within weeks, not months.
Visible champion who will use it and advocate for it.

Red flags for a first project

Fully autonomous high-stakes decisions (credit, medical, legal outcomes).
No clear owner or metric ("let's see what AI can do").
Data locked behind a 6-month access process.
Tasks that really need deterministic logic or classical ML (forecasting), not an LLM.

PreviousHow do you run a discovery session with a new customer?

This is what real progress feels like.

How do you choose the first AI use case to build with a customer?

Good first use case characteristics

Red flags for a first project

Related