A consultant configures predictive opportunity scoring for a client that runs two business units through one Dynamics 365 Sales instance: an enterprise hardware division with nine-month sales cycles and a small-business accessories division with two-week cycles. After scoring is enabled, reps in the accessories division report that almost every open opportunity gets a low score, even ones that close within days. What is the most likely cause and correct fix?
Select an answer to reveal the explanation.
Short Explanation
When you train one scoring model on two business units that behave completely differently, the model ends up judging every deal by the habits of whichever group dominates the history. Here that means a nine-month enterprise sales cycle is setting the baseline, so a healthy two-week deal in the smaller division looks abnormally fast and gets marked down, even though closing quickly is exactly what success looks like for that team. The trap options treat the symptom as something else entirely: one assumes there is simply not enough data to produce a real score and suggests giving up on the feature, one imagines scores go stale on a weekly clock and just need refreshing, and one assumes a single missing field is dragging every score down. None of those match what is actually happening. The real issue is that a single blended model cannot represent two different rhythms of selling at once, so the practical fix is to let each business unit train or calibrate its own scoring rather than sharing one model across mismatched sales cycles.
Full Explanation
The correct answer is B. When one model is trained on the blended history of two business units with very different sales cycles, it learns patterns dominated by the larger or longer-cycle group, and applies those patterns to opportunities that behave nothing like them. That is exactly what is happening here: hardware-cycle norms are being imposed on two-week accessories deals, so fast-moving, healthy opportunities look abnormal and score low. The fix is to separate the training data or scoring configuration by segment so each division's model reflects its own historical pattern of wins and losses. Option A is incorrect because the accessories division is receiving scores, just poorly calibrated ones; the problem is bad calibration, not absence of data, and telling reps to ignore the feature discards a signal that would be useful once fixed. Option C is incorrect because predictive scores are recalculated based on record activity and data changes, not a fixed weekly refresh cycle, so staleness is not the mechanism here. Option D is incorrect because estimated revenue is one input among many rather than a dominant factor, and populating it would not resolve a segment-level calibration mismatch caused by blended training data.