Rehearse a Discovery Call with AI and a Checkable Scorecard

An AI role-play can give you a place to practice asking a clear question, listening to the answer and choosing a useful follow-up. It cannot tell you how a real prospect will react or prove that you are ready for every sales conversation. The value comes from a well-defined fictional situation and a scorecard that points to what you actually said. A flattering summary from a simulator is much less useful than one missed question you can practice again.

This guide builds a short discovery-call rehearsal for a fictional service business. You will separate the prospect's private scenario from the caller's preparation, keep the simulated facts consistent and review a transcript against observable behaviors. The exercise supports focused conversations associated with The 7 Minute Phone Call site. It is original companion material, not an excerpt from the book, a new edition or an assertion that a particular published chapter teaches this AI routine. All businesses and dialogue in the example are fictional.

Choose one skill and one conversation objective

Do not begin with “make me better at sales.” Choose a behavior you can observe in a short exchange. Useful examples include asking an open question about the current process, clarifying a vague consequence or confirming what a useful next conversation would cover. A rehearsal with one target is easier to diagnose than a simulation that tries to measure rapport, qualification, persuasion, pricing, negotiation and closing at once.

Write the conversation objective separately from the skill. In the fictional example, the objective is to understand how a small service company handles appointment changes and decide whether a deeper workflow conversation is relevant. The skill is following up on the prospect's actual answer before switching topics. A caller can meet the skill target even if the appropriate conclusion is that another service or no further meeting is needed. That possibility should remain available in the simulation.

Keep the conversation's length as an exercise constraint rather than a promise that every real discovery call should end at a particular minute. Seven minutes can be a useful practice window because it forces choices, but some real conversations need more time or should end sooner. The exercise should help you identify the next useful question. It should not teach you to rush a person or force an answer to fit a timer.

Write a fictional prospect brief with stable facts

The simulator needs a small private scenario. Give the fictional company a simple service, team structure and current workflow. For example, a six-person maintenance company receives appointment changes through a shared email account. One coordinator checks the inbox each morning and afternoon. A manager approves changes that affect an already promised arrival window. The company has not measured missed messages, and the prospect does not know whether scheduling software would solve the issue.

Add a reason for the conversation that does not predetermine the sale. The fictional prospect might have agreed to talk after a colleague suggested reviewing the intake process. They have no approved budget and no obligation to book another meeting. Their concern is that people sometimes ask the manager for the same decision twice. These facts give the caller something concrete to explore without turning the role-play into a guaranteed success story.

List what the prospect knows, what remains unknown and what they should reveal only when asked. Known facts include the shared inbox and the approval role. Unknown facts include the number of missed messages and the financial effect of delays. A role-player should say they do not know those numbers rather than inventing them. Reveal-on-question facts might include that the coordinator is away on Fridays. That detail rewards a relevant coverage question without requiring the caller to guess it.

Keep the caller brief separate from the hidden scenario

The caller should receive only the information they would reasonably have before the call: the company offers maintenance services, someone suggested discussing appointment changes and the contact has agreed to a brief introductory conversation. Do not show the caller the hidden Friday absence or the answer to every discovery question. Otherwise, the rehearsal can become a performance of knowledge the caller would not possess in a real conversation.

If you are practicing alone in one chat, create a clear split between the scenario setup and the role-play. You may still remember the hidden facts, so acknowledge that limitation. A colleague can prepare the scenario or choose among several fictional variants when you want a less familiar exercise. The purpose is to practice a behavior under a stated setup, not to manufacture the appearance of a blind evaluation.

Avoid loading a real prospect's private emails, recorded calls or personal details into an unapproved tool. Fictional scenarios are sufficient for practicing question structure and follow-up. If your organization uses actual recordings for training, follow its normal permissions and handling procedures. This guide does not determine recording requirements or replace those policies. There is no need to introduce real customer data merely to make a first rehearsal feel realistic.

Give the simulator a role with clear limits

A starting instruction can read: “Act as the fictional prospect described below. Answer one caller turn at a time using only the scenario facts. Do not reveal hidden details unless a relevant question asks for them. When the scenario does not provide an answer, say you do not know. Do not coach or score the caller during the role-play. Do not agree to a next meeting automatically. End the simulation when I type END PRACTICE.” Then supply the private scenario.

Ask the simulator to keep invented details out of the conversation. A model may otherwise add a revenue figure, software contract or urgent deadline to make the dialogue flow. Those additions change the exercise. If a new fact is needed, pause and label it as an explicit scenario revision before continuing. A realistic-sounding answer is not a reliable simulation fact unless it came from the agreed brief or a documented revision.

Do not ask the model to imitate a named real customer or to reproduce a person's voice. Use a generic fictional role with an ordinary communication style. You can vary that style later: concise, uncertain, detail-oriented or distracted. Style should change how information is expressed, not secretly change the underlying facts. If a concise prospect suddenly becomes a different company with a budget because the caller struggled, the simulation has stopped testing the intended skill.

Begin with a permission-based opening and a neutral question

A practice opening might be: “Thanks for making a few minutes. I understand you wanted to look at how appointment changes reach the team. Is that still useful to discuss?” The aim is to confirm the topic and allow correction. The fictional prospect might answer, “Yes, though I am not sure we need another system.” A useful next question explores what currently happens, rather than arguing that software is necessary.

For example: “Could you walk me through the last ordinary appointment change, from the first message to the person who confirms it?” That question invites a sequence of observable work. It is more informative than “Are you frustrated with your current process?” which suggests the emotional answer the caller wants. The role-play should let you notice when your wording smuggles a conclusion into the question.

In the fictional response, the prospect explains that messages arrive in a shared inbox, the coordinator reviews them and the manager approves changes to promised windows. A relevant follow-up is: “What happens between the coordinator seeing the request and the customer hearing the decision?” A less useful response is to immediately describe your product's notification feature. The distinction is visible in the transcript, so the scorecard can point to it rather than making a vague judgment about listening.

Practice clarifying uncertainty without manufacturing pain

Suppose the prospect says, “Sometimes things get lost.” You do not yet know whether that means an unread email, an unclear approval, a duplicate question or a customer who never receives confirmation. Ask: “When you say lost, what would we see happening in one example?” The fictional prospect might describe a request being forwarded twice because nobody knows whether the manager has answered. That is more concrete than assuming a revenue loss.

If you ask how often it happens and the prospect does not know, preserve the unknown. A useful response is: “Would it be worth looking at a small sample before deciding what needs changing?” An unhelpful response is: “So this must be costing you thousands.” The second sentence turns missing evidence into a sales claim. Your simulator should not reward it by inventing a large number that makes the caller's argument look correct.

You can still explore consequences through specific questions. Ask who has to stop work, what gets repeated or how the team notices an unresolved request. Keep the answer at the level the prospect can support. A person may report inconvenience without having measured its cost. That is a real distinction a discovery conversation should preserve. The role-play is valuable when it helps you remain curious without pushing the speaker into a dramatic problem statement.

Use a scorecard based on observable turns

A practical scorecard can contain five behaviors: confirmed the topic, asked about the current process, followed up on a stated detail, kept unknown quantities unknown and checked whether a next step was useful. For each, record present, absent or unclear, plus a transcript turn identifier. A label without evidence is only a judgment. The turn reference lets you reread the actual wording and decide whether the behavior occurred.

Define each behavior before the simulation. “Followed up” means the caller asked a question tied to something the prospect had just said, not merely that the caller asked a second question. “Kept unknown quantities unknown” means the caller did not convert an unmeasured issue into a fabricated cost or frequency. “Checked a next step” means the prospect had a meaningful chance to decline or suggest a different focus. These definitions help prevent the review from grading charm alone.

Do not make an overall score look more authoritative than the rubric. Four observed behaviors out of five is a count of rubric items in one fictional exchange. It is not an eighty percent probability of closing a sale or proof of professional competence. Keep the missed behavior and its example visible. The most useful result may be a single line showing where you interrupted an explanation or ignored an unknown.

Review a short fictional transcript

Consider this exchange. Caller turn C1: “How do appointment changes reach the team today?” Prospect P1: “They go to a shared inbox. Our coordinator checks it twice a day.” Caller C2: “Our system would let everyone see messages instantly. Would you like a demonstration?” The caller has asked about the process, but the next turn moves directly to a solution. The transcript does not yet show whether twice-daily review is a problem or what happens after review.

A revised C2 could be: “After the coordinator reads a change request, how does the team decide who confirms it?” That follow-up uses a stated detail and explores the handoff. If P2 says the manager approves promised-window changes, C3 might ask what happens when the manager is unavailable. The hidden Friday coverage fact could then become relevant. The caller is uncovering decision information rather than proving that the chosen product is necessary.

At review, mark the original C2 as evidence that the caller switched topics before clarifying the handoff. Do not rewrite the transcript and then score the rewritten conversation as the original performance. Keep both versions, label one attempted and one revised, and practice the revised question in a fresh scenario. The comparison should help you hear the difference in behavior. It should not conceal the mistake that gave the exercise its value.

Ask for feedback without inviting invented authority

After ending the role-play, use a separate review prompt: “Review the transcript using only the five defined behaviors. For each, provide present, absent or unclear and quote a short relevant turn. Do not infer sales outcomes, diagnose personality or invent prospect reactions. Identify one follow-up question I could have asked, explaining which actual prospect statement it responds to.” Then inspect the proposed review yourself.

If the model cites a turn that does not exist, mark the feedback as unsupported. If it says the prospect was impressed, ask what observable statement supports that. A simulator's self-generated praise is not independent evidence about a real person's reaction. You can use the feedback to locate candidate issues, but the final review should remain tied to the transcript and the agreed rubric.

For a more grounded comparison, ask a colleague to review the same exchange without first reading the model's judgment. Compare disagreements about specific behaviors. If the rubric cannot distinguish a relevant follow-up from a topic switch, revise the definition. The goal is a usable practice method, not agreement for its own sake. Different interpretations can reveal that your question or scorecard is ambiguous, which is a problem you can fix.

Repeat one skill with a meaningfully different scenario

After the review, choose a new fictional situation that tests the same behavior. Instead of appointment changes, use a company handling quote revisions or new customer onboarding. Keep the target skill of following up on a stated detail. Change the workflow facts enough that memorizing the previous dialogue will not answer the new case. The caller should have to listen to what is actually said.

Vary one difficulty at a time. A prospect who gives brief answers is a different practice challenge from one who provides several conflicting details. If you change style, complexity, time limit and target behavior simultaneously, it becomes harder to understand why the next attempt felt different. Keep a short note about the changed condition. You are designing practice, not a laboratory study, but clear comparisons still help you learn from it.

Include an appropriate no-fit outcome. In one scenario, the company may already have a clear process and no relevant need for the caller's service. Practice acknowledging that and ending usefully. A role-play that always ends in a booked meeting teaches the caller that persistence must eventually be rewarded. Real conversations can end with a referral, a request for more information or a decision not to continue. Your practice should allow those ordinary outcomes.

Handle common simulator failures during the exercise

The simulator may become too helpful and volunteer every hidden fact in its first answer. Pause, note the failure and restart with an instruction to reveal only what the question reasonably asks. It may also drift into coaching during the conversation. Keep the prospect role and review phase separate. If it cannot maintain the boundary, shorten the exercise or use a human partner for the role-play while retaining the written scorecard.

Another failure is an overly agreeable prospect who accepts every premise. Test this by asking a leading question during a development run. A coherent fictional prospect should be able to say they do not know, disagree or correct the caller using the scenario facts. Do not confuse hostility with realism. You need a prospect whose answers are bounded by the brief, not a theatrical obstacle designed to make the caller feel challenged.

Retain the scenario version, prompt version and actual transcript for any comparison you make. A different model, changed interface or revised brief can alter the behavior of the exercise. That does not make rehearsal useless; it limits the conclusions you should draw from a score change. Treat the simulation as a practice aid whose outputs need checking, consistent with the broader concern about confidently generated errors described in NIST's generative AI risk profile.

Build a simple practice routine around the evidence

A short routine can include two minutes to read the caller brief, several minutes of dialogue and a separate review window. Those times are planning choices, not research-backed requirements. End with one specific practice intention: “I will ask what the speaker means by a vague process problem before proposing a solution.” The next exercise should give you an opportunity to attempt that behavior again.

Keep a small collection of transcripts that illustrate different gaps. One may show a leading question, another a premature proposal and another a useful clarification of uncertainty. A collection organized by behavior is easier to use than a folder of unlabelled conversations. Remove unnecessary personal details if any authorized real-world training material is later included. Fictional cases can remain the default because they are sufficient for many question-design exercises.

When a rehearsal exposes a genuine workflow design question, take that question to the appropriate discussion. The 7 Minute Phone Call site offers conversation resources; Elixir's services may be relevant when the issue concerns how a team owns an intake or follow-up process. A simulator should not decide the organization's operating policy. Its useful contribution is helping a caller ask clearer questions so the people with the actual facts can make a better-informed next decision.

Frequently asked questions

Can AI tell me whether my discovery call was good?

It can suggest observations about a transcript, but “good” needs a defined purpose and rubric. Ask for evidence of specific behaviors and check the cited turns. Do not treat a fluent review as an independent prediction of sales success. A useful review identifies something you can practice, such as a missed clarification, and distinguishes that observation from speculation about how a real prospect would feel.

Should the simulated prospect always be difficult?

No. Difficulty should match the skill you are practicing. A simple scenario can expose a topic switch clearly, while a complicated one can hide it under many unrelated problems. Include ordinary, uncertain and no-fit cases. The prospect should remain consistent with the brief and have room to disagree. Making every response hostile or every conversation successful would both distort the exercise.

Can I rehearse alone if I already know the hidden facts?

Yes, provided you recognize the limitation. You can still practice wording and review whether your follow-up connects to what was said. For a less familiar test, ask a colleague to prepare a fictional variant or choose one you have not read recently. Do not describe a self-prepared scenario as blind testing. The exercise remains useful when its conditions are reported honestly.

What if the model invents a budget or a deadline?

Pause and record the drift. Either remove the invented fact and restart from the last stable turn or explicitly revise the scenario before continuing. Do not let the new fact silently make your proposal appear more appropriate. The scenario's knowns and unknowns are part of the test. An invented number can change the conversation objective and invalidate the comparison with earlier attempts.

Should I memorize the revised questions?

Use them as examples of a behavior, not as a script to impose on every conversation. A follow-up works because it responds to a particular answer. In a new scenario, listen for the actual detail and choose a question that clarifies it. Memorized wording can help you start practice, but the scorecard should reward relevance and accurate understanding rather than repeating a preferred sentence.

Does this require recording real customer calls?

No. The complete exercise can use fictional dialogue entered as text. Real recordings introduce separate permission, handling and organizational requirements that are unnecessary for this first practice routine. If your team already uses authorized recordings, follow its existing procedures. Do not upload private call material merely to make a simulation more realistic when an invented case can test the same question skill.