A consultant is building a custom Copilot Studio topic for a sales chatbot that answers pricing questions from prospects on a self-service portal. When the topic's confidence in answering a pricing question is low, the design must hand the conversation to a human seller and preserve a record of the exchange for quality review. (Select TWO.) Which two actions should the consultant add to the topic to meet both requirements?
Select all correct answers, then click Submit.
Short Explanation
This one is really two separate asks stitched together, so the fix needs two separate pieces working side by side. One piece has to notice when the bot is unsure and pass the conversation over to an actual person instead of guessing at a pricing answer. The other piece has to quietly save a copy of what was said, so someone can look back later and judge how the bot handled it. Neither job covers the other — routing a conversation to a human does not automatically create a record of it, and saving a transcript does not get a person involved in the moment it is needed. Falling back on the same scripted answer no matter how confident the bot is defeats the whole point, since the situation specifically calls for different treatment when confidence is low. And turning off the safety net that triggers a handoff does the opposite of what is wanted — it leaves uncertain conversations running on autopilot with no reviewable trail and no human in sight.
Full Explanation
The correct answers are A and D. The scenario has two distinct requirements — routing low-confidence conversations to a human, and preserving those conversations for later review — and each requires its own action inside the topic. An escalation node that transfers the conversation once confidence drops below the threshold satisfies the human-handoff requirement, while a create-record action that writes the transcript to a custom Dataverse table satisfies the quality-review requirement; together they cover both halves of the design without either one substituting for the other. Option B is incorrect because returning the same canned answer regardless of confidence ignores the entire premise of the requirement, which is that low-confidence situations need different handling than confident ones, and it does nothing to involve a human seller. Option C is incorrect because disabling the fallback topic removes the very mechanism that would trigger an automatic handoff, which is the opposite of what the scenario asks for — it guarantees that low-confidence conversations are never escalated rather than ensuring they are. Neither C nor D contributes to logging a transcript either, so both fail on both counts.