Product strategy
Applied AI starts with the work, not the model
A striking demo can be built in hours. A product that improves a real operation begins when the team understands the work precisely enough to redesign it.
01
The model is a component. The work is the system.
Starting with “which model should we use?” skips the question that most determines the result: what work must change. A model can classify, summarise or generate, but it does not inherently know when to act, which source counts as evidence or who owns the decision.
The design unit should therefore be the operational decision: its trigger, required information, constraints, consequences and review path.
02
Map before automating
Follow a real case from beginning to end. Record where context arrives, where it is lost and where someone must reconstruct it. Volume is not the only opportunity signal: an infrequent step may deserve more care if it concentrates risk.
- 01What starts and ends the work?
- 02Which sources are authoritative?
- 03Which exceptions interrupt the normal path?
- 04Who can approve, correct or stop an action?
- 05What evidence must remain attached to the result?
03
From experiment to product
An experiment proves that a task can be assisted. A product proves that assistance can be repeatable, observable and integrated. That requires quality criteria, review, useful telemetry, failure handling and an explicit way to update rules.
Applied AI begins where people repeatedly reconstruct context the operation should already know—returning structure to the work so they can make a better next decision.
Sources consulted