A consultant is enhancing the Sales Qualification Agent for a subscription-box retailer. Marketing wants leads who abandon checkout after adding a promo code to be scored higher than leads who abandon without one, because promo-code abandoners convert at twice the rate once re-engaged. The retailer's e-commerce platform is not connected to Dataverse, and the abandonment event lives only in that platform's own event log. What should the consultant do first to make this signal usable by the agent?
Select an answer to reveal the explanation.
Short Explanation
Think about it this way: the qualification agent can only weigh what it can actually see in Dataverse, and right now this abandonment detail is stuck inside the retailer's own storefront system, invisible to everything else. The fix is not to ask people to remember and retype it, and it is not to swap in some other signal that is only vaguely related just because it happens to already be there. The real move is to pipe the actual event, complete with its promo-code detail, into a Dataverse field the scoring logic can read directly. That turns a one-off insight sitting in a separate system into a durable, structured input the model can lean on every time, rather than something that depends on a rep noticing and typing it in, or a proxy that only sort of tracks the behavior marketing actually cares about.
Full Explanation
The correct answer is D. The Sales Qualification Agent scores leads using signals available in Dataverse; a promo-code abandonment event that lives only in the e-commerce platform's event log is invisible to the model until it is integrated. Writing the event and flag into a Dataverse field the scoring model can reference makes the signal a durable, queryable input rather than something the agent has to infer indirectly. Option A is incorrect because manual flagging by reps is unreliable at volume, introduces inconsistent tagging, and does not scale as a data pipeline the way an integration does. Option B is incorrect because email engagement is a proxy at best; substituting a loosely correlated signal for the actual promo-code event discards the specific business insight marketing identified and produces a noisier score. Option C is incorrect because a Copilot Studio topic that interrogates the rep during a conversation does not solve the underlying data-availability problem for the scoring model itself, and it adds friction for something that should be captured automatically from the e-commerce event rather than recalled by a human.