A consultant investigating why every lead captured through the company's public website contact form receives the same middling predictive score, regardless of company size or engagement, discovers that the web-to-lead integration leaves several fields the model weights heavily, such as industry and employee count, blank at the moment of capture. What should the consultant do?
Select an answer to reveal the explanation.
Short Explanation
When a whole batch of leads all land on the same middling score no matter how big the company or how engaged they've been, that's usually not the model being lazy, it's the model not having anything to work with. If industry and employee count are consistently blank for every lead coming in through the website form, there's nothing for the model to differentiate those records on, so of course they all land in the same generic bucket. The real fix is upstream, get that form integration actually capturing those fields when the lead comes in, so the model has the same raw material for web leads that it has for everything else. Carving web leads out and scoring them by hand forever just accepts the gap instead of closing it. Stripping those fields out of the model altogether to make things consistent would fix nothing for web leads while making scoring worse for every other source that was populating them fine. And just rescoring more often won't help either, because refreshing the score doesn't summon data that was never captured to begin with.
Full Explanation
The correct answer is C. When fields the model relies on are consistently null for an entire lead source, the model has nothing to differentiate those records on, so they cluster around a generic score regardless of how promising any individual lead actually is. Fixing the integration so industry, employee count, and similar fields populate at the moment of capture addresses the root cause and lets the model score web leads using the same information it uses for leads from other sources. Option A is incorrect because it works around the symptom rather than fixing it, permanently losing the efficiency benefit of predictive scoring for an entire lead source when the underlying data gap is fixable. Option B is incorrect because removing those factors from the model to fix one source's data gap would degrade scoring quality for every other lead source that does populate those fields correctly, trading a source-specific problem for an organization-wide one. Option D is incorrect because rescoring more often does not create data that was never captured in the first place; refresh frequency has no bearing on fields that remain blank at the source, so the clustering would persist no matter how often the score recalculates.