Machine learning for business: start with the decision, not the model

The most common reason machine learning projects stall is that they start with a technique instead of a decision. "We should use AI" is not a project. "We want to know which orders are likely to be delayed so we can act early" is.

Three questions before you build

• **What decision will change?** Name who will act on the prediction and what they will do differently.
• **How will we know it worked?** Choose a measure — fewer late deliveries, less stock written off, faster triage.
• **Do we have the history?** A model learns from past examples, with the outcome recorded. Without that, start by collecting it.

Start simple

A well-chosen rule or a simple statistical model is often good enough and easy to explain. Use it as the baseline any more complex model must beat.

Keep people in the loop

Present predictions as risk scores or recommendations with the main contributing factors, and let staff override them. Their overrides are valuable feedback.

Plan for drift

Models degrade as the world changes. Monitor accuracy after launch and schedule retraining.

Read more about our machine learning and predictive analytics work.