Estimate ML Effort Before You Commit

A machine learning estimate is never only about model development. The effort includes problem discovery, data access, labeling, evaluation, integration, operations, and the people needed to keep the capability healthy after launch.

This is an original companion to Part 8 of the AI & PM Insights ML strategy series on effort and likelihood of success. The purpose is to replace false precision with a plan that is honest about uncertainty.

Estimate the whole system

Break the work into the decisions that must be made before the product can create value. Include data rights and quality, domain expertise, technical implementation, product design, integration dependencies, evaluation, and operational ownership.

This gives leaders a clearer picture of what they are funding. It also surfaces dependencies that might otherwise appear as late surprises.

Separate certainty from assumption

For each major workstream, identify what the team knows, what it is assuming, and how it will learn. A range is more useful than a single confident date when the range is tied to the assumptions that drive it.

For example, a data audit may reduce uncertainty about collection and labeling. A small end-to-end prototype may reduce uncertainty about integration. These are investments in learning, not delays before the real work starts.

Make likelihood visible

A credible plan acknowledges both the expected benefit and the conditions required to achieve it. Review technical feasibility, data readiness, user adoption, delivery capacity, and operational risk together. Then decide whether to proceed, narrow the scope, sequence a discovery milestone, or choose another approach.

The strongest estimate gives a team a way to act under uncertainty. It should not create the illusion that uncertainty has disappeared.