← All articles

Plan and connect AI handoffs

How to Design an AI Sales Role-Play That Gives Reps Useful Feedback

Turn one sales skill into an AI-prospect scenario, a behavior-based rubric, and feedback reps can use on their next attempt.

By Niko Minadze8 min read

Editorial illustration for How to Design an AI Sales Role-Play That Gives Reps Useful Feedback

An AI sales role-play becomes useful when a rep knows exactly what skill they are practicing and can see where their response met, or missed, the mark. Start with one call situation, give the AI prospect facts it must keep consistent, and assess a few actions visible in the exchange. Then have a coach test the exercise before assigning it to reps. A plausible conversation alone is not proof that its feedback is right.

Sales simulation can pair a virtual prospect with post-session feedback. The Pitchbase guide, for example, describes feedback on skill breakdowns, key moments and alternative phrasing. Those are useful design possibilities, not a reason to accept every assessment without review.

1. Pick one behavior for one call

Suppose an SDR is making an initial call about Harbor Handoff, a fictional shared board for tracking customer handoffs. The practice goal is not “improve discovery.” It is: find out what the buyer means by an integration concern before claiming the product solves it or asking for a demo.

A successful attempt would sound like the rep asking how the handoff works now, what needs to move between systems, or what “integrated” means to this buyer. The rep should learn the requirement, reflect it back, and avoid promising a capability that has not been established. These are actions a coach can inspect in a practice exchange; the simulated call cannot establish whether the rep would win a real deal.

2. Write the buyer’s facts before the dialogue

Give the prospect enough context to answer different questions, but not so much that the exercise becomes a speech. For this hypothetical call:

  • Buyer: An operations manager at a service business. They are direct and short on time, but will answer relevant questions.
  • Current process: The team records sales-to-delivery handoffs in a spreadsheet. Customer information also sits in a CRM. The buyer sometimes has to reconcile the two manually.
  • Concern: They do not want another disconnected place to enter the same information. They want to know whether a handoff board would fit their existing process.
  • Specific requirement: At each handoff, the customer name and agreed start date from the CRM need to reach the delivery team’s board without someone retyping them. The buyer has not decided whether a CSV import would be workable; they also want to know how a changed start date would be handled.
  • Offer facts: Harbor Handoff is a shared handoff board and supports CSV import. Whether it has live CRM synchronization is not established in this exercise. Neither the buyer nor the assessor should invent an answer.

These are teaching-example facts, not claims about an available product. Keep them fixed while allowing the prospect’s tone and resistance to respond to the rep. Kendo advertises control over role-play personality and call type, as well as the product or offer being practiced. The Pitchbase guide describes persona, resistance and hidden concerns as scenario inputs. Neither description proves that a generated buyer will keep its facts consistent; that is something the coach must check.

3. Instruct the prospect to react, not perform a script

Put the scenario facts into the role-play setup available to you, or use the following as a plain-language brief where a tool accepts one. The setup method depends on the tool. Keep instructions for the prospect separate from instructions for the assessor, so the buyer does not start delivering coaching mid-call.

Act as the operations manager in the scenario above on an initial SDR call.
Open with: “What did you want to discuss?”
Answer the rep’s questions using only the scenario facts. Be concise and direct.
If the rep asks about the current process, explain the spreadsheet and CRM handoff.
If asked what information needs to move, how updates work, or what “connected”
or “integrated” would mean to the buyer, disclose the specific handoff requirement
and the buyer’s uncertainty about whether CSV import would suffice.
If asked how updates work, also explain that the buyer wants to know how a changed
start date would be handled.
Raise the concern about duplicate entry when it fits the conversation, especially
if the rep proposes another tool without learning about the current process.
Do not switch systems, needs, or product facts during the call. Do not assert that
Harbor Handoff has live CRM synchronization; that has not been established.
If the rep makes an unsupported integration claim, ask what they mean by it.
Do not score or coach the rep while acting as the buyer.

The disclosure condition matters: a rep who asks what “connected” means should be able to uncover the same handoff requirement as one who asks what information must move. A rep who asks how updates work should also hear the changed-start-date concern. A fixed opening makes attempts easier to compare; the remaining answers should respond to the rep rather than recite a script. These are testable design instructions, not a guarantee the AI will follow them.

4. Score actions the exchange can show

Use a small rubric tied to the single goal:

  1. Investigates the present handoff: Asks how information currently moves from sales to delivery, rather than assuming the spreadsheet is the whole process.
  2. Clarifies and reflects back the requirement: Asks what information needs to move, how updates work, or what “connected” or “integrated” means to this buyer; learns the requirement from the answer and reflects it back before proposing a demo or claiming the product meets it. An initial question alone does not complete this criterion.
  3. Handles the unknown honestly: Does not present live CRM synchronization as a confirmed capability; if it matters, says the requirement needs checking.

Give the assessor the exchange and the same fixed scenario facts, but a different job from the prospect’s:

Assess only what the recorded exchange shows against the three criteria above.
For each criterion, return: observed, not observed, or unclear; the words in the
exchange supporting that mark; and one concrete response the rep could try next.
Do not infer an action that is absent from a complete exchange. Use unclear when
the record does not reliably show what happened. Do not treat an unanswered
question as a reflected requirement. Flag product claims not established by the
scenario facts. Do not predict a real buyer’s decision or a deal outcome.

For example, imagine the buyer says, “We track handoffs in a spreadsheet, but I can’t add another disconnected tool.” The rep replies, “We can automatically sync your CRM. Can we schedule a demo?” Useful feedback would quote the rep’s reply and mark clarification not observed: the rep proposed a demo without learning and reflecting back what must connect. It would also flag “automatically sync” as unsupported by the exercise facts. One alternative for the next attempt is, “What would connected mean in your handoff, what information needs to move, and what happens when a start date changes?” The rep can then reflect the answer before discussing a possible solution. This exchange is hypothetical; it is not a completed test of an AI assessor.

Unclear is different from missing behavior. If a recording cuts out during the rep’s response, the assessor may lack evidence to mark it either way. If the complete exchange shows a question but no reflection, criterion 2 is not observed. A coach should be able to point to the same words when agreeing with either judgment.

5. Have a coach try to break the exercise

Before using the role-play for rep development, have a coach take several paths: ask what data must move, ask what “integrated” means, ask how updates work, ask how changed dates are handled, pitch immediately, and make an unsupported integration promise. Inspect whether each relevant question reveals the intended facts. In particular, the updates question should surface the changed-start-date concern, not just the fields that move at handoff. Then have the coach assess the recorded exchanges without looking at the AI’s feedback first.

Compare cited moments, not just overall marks. If the AI penalizes a rep who learned and reflected back the requirement before pitching, tighten the rubric or assessor instructions. If the buyer alternates between “manual CSV is fine” and “only live sync is acceptable,” correct the scenario: neither position is a fixed fact here. Resolve disputed judgments with a coach before using the feedback to guide a rep’s development.

The exercise is ready for broader practice when a rep can name the skill being tested, different sensible questions uncover consistent facts, an updates question surfaces the changed-date concern, and a coach can substantiate the main feedback from the exchange. Until then, revise the scenario or assessment, not the rep’s development plan, based on what the trial reveals.

← All articles