A consultant is enabling predictive lead scoring for a startup that migrated to Dynamics 365 Sales only three months ago and has closed fewer than 200 leads to date. Predictive scoring needs enough historical won and lost lead outcomes to train a reliable model, and the client's current volume falls well short of that. What should the consultant recommend?
Select an answer to reveal the explanation.
Short Explanation
Think of predictive scoring like a forecaster who needs a season of games under their belt before their predictions mean anything. A brand-new environment with only a couple hundred closed leads just hasn't played enough games yet, so there's no reliable pattern for the model to learn from. Turning the feature on anyway would hand sellers scores that look official but aren't grounded in real outcomes, which tends to burn trust fast. The better move is to keep prioritizing leads the old-fashioned way, with clear manual criteria or simple rule-based scoring, while the team keeps closing deals and building up that history. Stuffing old CRM data in as loose notes doesn't help either, because the model needs structured win and loss records, not paragraphs Copilot can summarize. And asking sellers to just make up scores for existing leads is even worse, since fabricated numbers aren't real outcomes and would just teach the model the wrong lessons later. Patience here pays off with a model that actually reflects how this client's deals really close.
Full Explanation
The correct answer is A. Predictive lead scoring trains on a tenant's own history of closed won and lost leads, and with fewer than 200 closed records the model has too little signal to learn patterns that generalize to new leads. Recommending manual prioritization or configurable rule-based scoring in the interim keeps the sales team working leads in a sensible order while the environment accumulates the history predictive scoring needs, after which the feature can be enabled with a realistic chance of producing reliable grades. Option B is incorrect because insufficient historical data does not simply produce a rougher model that improves gradually; without enough closed-lead outcomes to learn from, the scores produced would not reflect real conversion patterns at all, which risks eroding seller trust before the feature has a fair chance to work. Option C is incorrect because free-text notes are unstructured content Copilot can summarize conversationally, but they are not the structured won and lost outcome data the predictive model trains against, so importing them this way does nothing to satisfy the volume requirement. Option D is incorrect because manually assigned scores are not actual sales outcomes; substituting them would not give the model real conversion history to learn from and could bias any future training on the same records.