Frame the decision: write the brief before the model
Most analytics projects that fail to land were never tied to a decision. A brief written first forces the conversation that decides whether a model is needed at all, and what it has to beat.
Name the decision in one sentence
Use a verb and an object: “Choose which markets to enter first”, “Decide whom to approve at which price”. If the sentence is about a prediction (“forecast churn”), keep going until it names the action the prediction would trigger.
Name the owner and the cadence
One accountable person, and how often the decision is made. A weekly decision and a once-a-year decision need different tools, different evidence and different tolerance for error.
List the real alternatives, including doing nothing
What are the options on the table today and what is the current rule of thumb? That is the baseline the model has to beat, not zero.
Price the two kinds of wrong
What does a false yes cost, and what does a false no cost? They are rarely equal, and the ratio sets the threshold more than any metric does.
Separate hard constraints from preferences
Constraints (policy, regulation, capacity) screen options out. Preferences (weights, trade-offs) rank what is left. Mixing them hides the trade-off the owner is actually making.
Write the stop rule
What evidence would make the owner choose differently, and what result would end the project? Agree it now, while nobody is attached to an answer.
Checklist
Common failure modes
- The brief describes the model instead of the decision.
- Several owners, so nobody can say what “better” means.
- Success is defined after the results arrive.
See it applied
Open conversations