DECISION ATLAS / THE WORKING METHOD
A large problem is
too large to solve.
Break it down. Segment it. Find where the value or failure is concentrated. Then decide what to do, to whom, through whom, and how you'll know it worked.
Start with the problemWhat are we actually trying to change?
Start with an outcome, not a model, vendor or dashboard. Choose a fictional problem, then challenge how it should be partitioned.
One problem. Several legitimate cuts.
The same population can be partitioned by different dimensions. A good cut isolates different mechanisms and different actions. Every cut here is mutually exclusive and collectively exhaustive within the synthetic dataset.
Where does attention create the most value?
Change the objective and the priorities may move. The estimates are illustrative assumptions, not observed causal effects.
| Segment | Base | Reachable* | Actions* | Net opportunity* | Select |
|---|
* Illustrative only. Formula: eligible × reachable × assumed incremental action rate × net contribution per action, minus contact/handling cost and assigned setup cost. Reachable = base × eligibility × reachability. The estimates can be wrong; no segment is automatically authorised for action.
Which question could change your mind?
Open the right diagnostic instrument for the selected segment. Your original library of 42 defined measures and 13 combinations sits behind this step.
Who does what, to whom, and through which route?
Turn the diagnosis into a focused, falsifiable intervention—not a bank-wide technology proposal.