A sales manager complains that leads graded Warm under Dynamics 365 Sales predictive scoring rarely convert for her team, while the underlying numeric scores actually track conversion reasonably well. She asks the consultant to fix the mismatch between the label and reality without retraining or replacing the model. What should the consultant do?
Select an answer to reveal the explanation.
Short Explanation
The number the model produces and the word stamped on top of that number are two different things, and this situation is really about the word, not the number. If a Warm label is landing on leads that behave more like Cool leads, the fix is to slide the boundary line that separates those two buckets, not to touch the math that produces the score in the first place. Retraining would be solving a problem that does not exist here, since the manager already says the numbers track conversion fine. Having sellers hand-edit labels one lead at a time just creates a permanent chore instead of a permanent fix, and it would not help the next lead that comes in tomorrow. And there is no need to throw out the whole feature, since the boundary between labels is exactly the kind of setting built to be adjusted as a team's real-world conversion pattern becomes clear. Think of it as recalibrating where 'warm enough to call' actually starts, not repainting the whole scoring engine.
Full Explanation
The correct answer is B. The Hot, Warm, and Cool labels are just named ranges applied on top of the numeric score, and those ranges are configurable independently of the underlying model, so the consultant can move the boundary between Warm and Cool upward until the label reflects where this team's leads actually stop converting well. Option A is incorrect because retraining addresses the accuracy of the numeric score itself, which the manager already says is fine; the problem here is purely how that score gets labeled, and retraining would not touch the band configuration at all. Option C is incorrect because manually overwriting each lead's grade is a one-off patch that does not scale, has to be repeated indefinitely, and leaves the systemic threshold problem in place for every new lead that arrives. Option D is incorrect on its premise: band thresholds are reconfigurable while the model stays live, so there is no need to abandon predictive scoring in favor of a manual rule just to fix a labeling boundary. The right move is a settings change, not a data change, a manual workaround, or a feature replacement.