Six months after enabling AI-driven opportunity scoring, a consultant reviewing the pipeline notices that most open opportunities cluster tightly between scores of 58 and 65, giving sellers little practical way to tell which deals deserve priority attention. Investigation shows the model was trained on a modest set of historical opportunities, most of which are closed-won, with very few closed-lost records and little variation in recorded loss reasons or deal attributes. What should the consultant recommend?
Select an answer to reveal the explanation.
Short Explanation
Picture a model trying to tell winning deals from losing ones the same way a person learns to tell ripe fruit from unripe fruit, mostly by seeing lots of examples of both and noticing what differs between them. If almost everything in the training history is a win, with only a handful of losses and barely any explanation of why those losses happened, there just is not enough contrast to learn from, so the model ends up treating most opportunities as roughly the same. That shows up exactly as described here, a pile of scores bunched together with nothing sellers can use to prioritize. The fix has to happen at the source, feeding the model a richer, more balanced history of both outcomes so it has real differences to learn from, not a cosmetic tweak to how the numbers are displayed or how often they refresh. And scrapping the feature entirely throws away a tool that could genuinely help once it has been given something real to learn from.
Full Explanation
The correct answer is D. A predictive scoring model learns to differentiate outcomes by finding patterns that separate wins from losses across a wide range of historical examples. When the training set is small and dominated by closed-won records with little variety in closed-lost outcomes or documented loss reasons, the model has almost no contrast to learn from, so it produces scores that cluster together rather than spreading opportunities across a meaningful range. Enlarging and balancing the historical dataset, then retraining, gives the model the variation it needs to produce scores sellers can actually act on. Option A is incorrect because changing how scores are rounded or displayed is a cosmetic adjustment that does not address the underlying lack of model discrimination; the scores themselves remain uninformative. Option B is incorrect because refresh frequency controls how current the scores are, not how well the model distinguishes strong deals from weak ones, so recalculating more often would just produce the same clustered scores more frequently. Option C is incorrect because abandoning AI scoring discards a capability that could work well once properly trained, and reverts the team to the inconsistent manual prioritization the feature was meant to improve.