Evaluating a workflow¶
A workflow is a segmentation of your range: every decision splits products into groups that get different treatment. Workflow evaluation answers the question behind the question — do these splits actually separate products that behave differently? — using your own sales history.
Segmentation quality¶
The evaluation panel scores the workflow's decision tree against historical performance:
- Segmentation quality — an overall score for how well the tree divides your products into commercially distinct groups.
- Per-decision contribution — each decision node shows how much explanatory power it adds to the tree. A decision that adds nothing is a split your data says doesn't matter — a candidate for removal or a sign the threshold is in the wrong place.
- Branch splits — visual comparisons of the populations on each side of a decision, so you can see whether a split produces genuinely different groups or two halves of the same behaviour.
How to read it¶
- A strong decision separates products with materially different price-performance behaviour — treat them differently with confidence.
- A weak decision may still be right for policy reasons (contractual tiers, brand rules). The score tells you the data doesn't distinguish the groups; only you know whether the business does.
- A weak tree overall suggests the trigger population is too homogeneous to need branching — or that the range should be segmented on a different attribute than the one the tree uses.
Evaluate before you widen
Evaluation is most useful when extending a proven workflow to a broader population. If quality drops sharply as scope widens, the policy doesn't generalise — segment the new population separately instead of stretching the old logic over it.