What scoring a discovery call involves
Scoring a discovery call means rating each discovery skill the rep showed on one call against written levels, with a quote from the call behind every rating. A front-line manager or enablement lead does it, and it takes about as long as the call itself plus time to check quotes once the rubric is familiar.
- 01Write down what the call was for
- 02Listen to the whole call onceWithout scoring
- 03Mark each skill's momentA skill whose moment never came is N/A
- 04Climb the ladder, whole levelsStop at the first no
- 05Quote every score, then check it
- 06Choose one skill for coachingThe lowest one this deal depends on
Discovery is where a deal gets its facts: the problem behind the stated one, what it costs, who decides, and what happens next. A scored discovery call tells a manager which of those facts the rep came away with and which skill let the others slip. For the wider picture, see our AI call scoring guide, which covers how the same scoring works when a model does it.
The method below borrows from the research behind that guide. Between August and October 2026 we ran more than 5,000 tests and more than 250,000 scoring decisions against a 20-skill capability scorecard on enterprise discovery roleplays, with blind marking by PeerLab's trained assessors. The final setup ranked calls in the same order as the assessors at 0.89 (field guide). Several things that made the model accurate apply to a human scorer too, and they are built into the steps.
The 20-skill rubric at a glance
Each skill is scored in five whole levels, where 1 is absent and 5 is expert. The full rubric, with descriptors at every level, is the discovery call scorecard template.
| Phase | Skills |
|---|---|
| Opening | Agenda and purpose · Framing · Credibility |
| Situation | Current-state discovery · Problem discovery · Follow-up questioning · Listening and acknowledgement |
| Impact | Impact quantification · Urgency and timing · Personal stakes |
| Decision | Stakeholder mapping · Decision process · Decision criteria · Budget and commercial context |
| Value | Linking capability to the problem · Differentiation · Handling objections · Emotional control · Negotiation |
| Close | Gaining commitment |
Twenty sounds like a lot for one call. Few calls trigger all of them, and the skills whose moment never came are marked not applicable.
How to score a discovery call in seven steps
1. Write down what the call was for
Before opening the recording, write one line: what should this call have produced for this deal? For a first discovery call that is usually a confirmed problem, a sense of its size, the people involved in the decision, and an agreed next step. Good looks like: a line specific to the deal ("confirm whether the Q1 system review includes dispatch, and who runs it"), so the scoring has a target.
2. Read or listen to the whole call once without scoring
Get the arc. Emotional control shows in how the rep handles the second push-back, which only makes sense next to the first. In our research, scoring single turns out of context dropped ranking agreement to between 0.57 and 0.63, against 0.89 for the whole-call method. Good looks like: you can say in two sentences what the buyer gave the rep and what they held back.
3. Mark where each skill's moment came
Go through the transcript and tag the moments that call for a skill: the buyer names a problem, mentions a colleague, raises price, pushes back. A skill whose moment never came is not applicable. Leave it out of any average. Good looks like: every score you give later points to a moment you tagged here.
4. Climb the level ladder for each skill
For each skill ask in order: is this at least level 2? At least level 3? Level 4? Level 5? Stop at the first "no". Our final scoring method used the same structure, 80 yes/no questions per call (four per skill), and it was the steadiest setup in our tests. Good looks like: each "yes" rests on something the rep said or did, and each "no" on something missing.
5. Score in whole levels
There is no 3.5. In one production engine we tested in our call-scoring research, 37% of scores were fractional on a scale defined in whole levels, and rounding alone added 16 points of exact agreement with the assessors. A half level is the scorer hedging. Good looks like: you can say which descriptor the call met and which it missed.
6. Attach a quote to every score, then check it
Copy the words from the transcript that justify the level. Then check each quote is there. With cheap automated extraction, a fifth of the quotes a model produced were not in the call. Human scorers paraphrase from memory too, so check theirs the same way. Good looks like: a rep reading your scorecard can find every quote in their own call.
7. Choose one skill for coaching
Pick the lowest-scoring skill that this deal depends on right now. A level 2 on differentiation matters less than a level 2 on stakeholder mapping when the buyer has just mentioned a CFO the rep has never met. Good looks like: one skill, one reason tied to the deal, one thing to try on the next call.
Worked example
This example is illustrative; the company, people and quotes are invented.
Marcus sells scheduling software. His first discovery call is with Elena, head of clinic operations at a group of physiotherapy clinics. The line for step 1: "find out whether no-shows are a priority this year and who signs off on a new system."
| Skill | Level | Quote | Ladder note |
|---|---|---|---|
| Problem discovery | 4 | Elena: "It's the no-shows, but really it's that we can't backfill the slot in time." | Got past the stated problem to the cause; did not ask who else feels it (no 5) |
| Impact quantification | 2 | Marcus: "That must be expensive." Elena: "It adds up." | Asked about cost, accepted a vague answer (no 3) |
| Stakeholder mapping | 3 | Marcus: "Who else would look at a system like this?" Elena: "Our finance director, and the clinic leads." | Asked who; did not ask how they decide (no 4) |
| Gaining commitment | 2 | Marcus: "I'll send over a case study and we can go from there." | Proposed a step the buyer did not confirm (no 3) |
Handling objections was marked not applicable: Elena raised none.
Step 7 picks impact quantification. The deal has a real problem and named stakeholders, but a finance director will compare the cost of the system with the cost of the problem, and there is no number for the second. The coaching for the next call is one question in Elena's own terms: "How many slots a week go unfilled after a late cancellation?" A follow-up step that Elena agrees on the call would lift gaining commitment too.
Where it goes wrong
Counting in place of judging. Rubrics that score "at least three open-ended questions" are quick and easy to game. A rep can ask three open questions and learn nothing. Score what the buyer gave the rep.
Scoring the outcome. The deal moved forward, so the call must have been good. Sometimes the buyer was ready regardless. Score the behavior on this call against the descriptor.
Scoring from notes. Notes taken during the call carry the scorer's memory of it. Score from the transcript, which is where step 6 catches paraphrase.
Averaging across skills. A 3.2 overall hides a 1 on stakeholder mapping in a deal with five approvers. Report skill scores; use the average only to compare calls of the same type.
One scorer, no check. Trained experts with a shared scorecard still disagreed on 18% of scores in our call-scoring research. Have a second manager score one call in ten blind and compare. Our guide on calibrating AI call scores covers the same check for a model.
Scoring every rep's discovery calls
A manager with eight reps running five discovery calls a week has 40 calls to score, on top of the job. That is where an AI scorer earns its place, provided it has been measured: the final method in our research ranked at 0.89 with the assessors and changed 1.7% of scores on a re-run, at about $0.007 per call. The questions to ask any scorer are in Can AI score a sales call?.
Elena's deal shows the other half of the job: one question about unfilled slots, rehearsed before the follow-up. With Caty, our AI sales expert for live deals, the rep brings the deal on a call, talks through what the buyer said, and Caty works the weak skill on that deal with the team's own method and scorecard as the standard. To see one of your own calls scored first, talk to us. The scorecard structure itself is explained in The sales call scorecard.
FAQ
How do you score a discovery call?
Score each discovery skill against written levels, one skill at a time, using the full transcript. Mark where each skill's moment came, climb the ladder ("at least level 2? 3? 4? 5?") until the answer is no, score in whole levels, and attach a checked quote to every score. Then choose one skill for coaching based on what the deal needs next.
What should a discovery call scorecard include?
It should cover the skills that produce a deal's facts: problem discovery, impact, urgency, stakeholders, decision process and criteria, linking the product to the problem, objection handling and gaining commitment. Each skill needs a description of every level. Our template has 20 skills in five whole levels, and skills whose moment never came are marked not applicable.
How long does it take to score a discovery call?
Scoring a discovery call takes roughly the length of the call plus time to check quotes, once the rubric is familiar. That covers one full read of the transcript and the level ladder for each skill. The first few calls take longer. An AI scorer that has been measured against your managers does the same work for about $0.007 per call in our research setup.
Should a discovery call be scored on a 1 to 5 scale or yes/no?
Use five levels, and reach them through yes/no steps. Asking "is this at least level 2? level 3?" in order gives the consistency of yes/no decisions and the detail of a five-level scale. The most consistent method in our call-scoring research used that structure, 80 yes/no questions per call across 20 skills, and changed 1.7% of scores on a re-run.
What if the buyer never raised an objection?
Mark handling objections as not applicable and leave it out of any average. Scoring a skill whose moment never came punishes the rep for the buyer's behavior. If a skill comes back not applicable on most calls, either it is rare at this stage or the descriptor is too narrow, and both are worth knowing.
Can AI score discovery calls accurately?
AI can score discovery calls accurately with a measured setup. In our research the final method ranked discovery calls in the same order as trained assessors at 0.89, and changed 1.7% of its scores on a re-run, while two experts marking the same calls disagreed on 18% of scores. Single models with default prompts did worse, and newer models marked reps more harshly than the assessors.