Output is not outcome
It is easy to count dashboards built, models deployed or reports scheduled. None of those numbers says whether anyone chose differently because of them. Value appears only when a decision, made by a particular person at a particular time, is better than it would otherwise have been.
That suggests a different unit of account: the recurring decision. Pricing a product, replenishing a store, approving a loan or scheduling maintenance are all decisions with owners, frequencies and costs of error.
Start with an inventory of decisions
List the decisions that matter most in a function, estimate how often each is taken and what a wrong choice costs, then ask what evidence the decision maker uses today. Gaps between what is used and what is available point to the highest-value analytics work.
This list also reveals decisions that are made implicitly, with no clear owner. Making an owner explicit is frequently the first and cheapest improvement.
Design for the moment of choice
Evidence delivered after the decision is history. Evidence delivered in a different tool is often ignored. The design question is where the decision is made and what the person needs to see there: the recommendation, the reasoning, the uncertainty and a way to override.
Capturing overrides and outcomes closes the loop and shows whether the analytics or the human judgement was closer to the truth.