A sales call scorecard should score five to eight criteria for any one call type, drawn from a larger skill model, and the manager should coach on one of them per conversation.
What other guides recommend
Published guides land in a similar range. Hyperbound recommends "between six and ten scored criteria per call type". Quotain quotes an enablement leader who preferred "four to six scored behaviors, each with a short reason". When managers skip rows or mark them N/A often, Quotain's advice is to split the rubric by call type or cut the low-value rows.
The count is the easy part. What decides whether a scorecard is useful is how each criterion describes its levels, and whether the manager turns the result into one thing the rep does differently on the next call with that buyer. The AI call scoring guide covers the full system.
Three numbers that get confused
The skill model. This is every skill your team expects across the sales cycle. It can be long. The capability scorecard in our research program carried 20 skills, from problem discovery to negotiation (field guide). Nobody scores all 20 on one call.
- The skill model20 skills in our research
- The call-type scorecard5 to 8 skills
- The coaching ask1 skill
The call-type scorecard. This is the five to eight skills that a given call should show. A first discovery call is judged on discovery, impact and stakeholders; a pricing call on value and negotiation.
The coaching ask. This is the one skill the rep works on before the next conversation. A second skill splits the rep's attention on a live call with a buyer who will not follow the plan.
Why five to eight rows per call type
Five to eight is roughly what a manager can find evidence for in one call. Every row asks the manager to locate a moment, pull the buyer's or the rep's words, and match them to a level. On a 40-minute discovery call that is a real job, and it gets done badly once the list runs long.
Too many rows fails in a predictable way. The rows the call was never going to show get marked N/A, or worse, get a polite middle score so the sheet looks complete. Managers rush the last few. The average drifts toward 3, and the one row that decided the call sits in a total nobody can read.
Too few rows fails differently. With three rows, a call that nailed the problem and never asked who else is involved scores the same as one that did both. The rep gets a number that cannot tell them what went wrong.
A simple test settles most arguments about a row: can you name the minute of a typical call where it would show up? If you cannot, it belongs on another call type's scorecard, or on none.
What matters more than the count
Level descriptions carry the weight. In our research (August to October 2026, more than 5,000 tests and more than 250,000 scoring decisions, with blind marking by PeerLab's trained assessors), the biggest single gain in agreement with human scores came from giving our AI scorer marker reasons, the assessors' short written notes on why a real call sat at a level. They lifted its average fit with human scores from 0.62 to 0.72. A different model, GPT-6 Sol, gained nothing from the same reasons: its rank agreement stayed at 0.73. Textbook examples added nothing.
One rewrite went backwards. When we turned the scorecard into "observable checks", its rank agreement with the assessors' order of calls (a separate measure from the fit above) came out at 0.62, worse than the original scorecard (field guide).
So when a scorecard produces inconsistent scores, adding or cutting rows is the wrong first move. Write down why your best managers gave each level, and put those reasons into the descriptions. Even trained experts disagreed on 18% of scores in our study, and a short scorecard with vague levels will do worse than that.
How to write the levels for one criterion
Each level should describe something a listener could point to on the recording. Here is one row, impact, written that way. The wording is illustrative; your method will have its own terms.
| Level | What the rep did on the call |
|---|---|
| 1 | Did not ask what the problem costs the buyer |
| 2 | Heard the buyer describe a cost in general terms ("it costs us a lot") and moved on |
| 3 | Asked a follow-up and got a consequence in the buyer's words, with no size |
| 4 | Got a size in the buyer's own units (hours, missed shipments, money) and who carries it |
| 5 | Got the size and the owner, tied it to a date that makes it urgent, and played it back for the buyer to confirm |
This is different from the observable-checks rewrite that scored worse in our research: that turned levels into yes/no boxes and dropped the marker reasons. Our checklist model keeps the graded levels and asks a yes/no question to climb each one; the failed rewrite replaced the levels. Keep the graded levels and add the reasons.
Two things make this row scorable. Each level adds one observable step to the one below, so two managers arguing over a 3 or a 4 are arguing about a fact (did the buyer give a size?) and can settle it from the transcript. And level 2 names the common miss in the buyer's likely words, which makes it easy to spot on a fast listen.
Then add the marker reasons. Next to each level, write a sentence a manager gave when scoring a real call at that level: "Scored 3: she asked what the delays meant for the team, got 'late nights for dispatch', and did not ask how many." Assessors' reasons like these are what lifted our AI scorer's agreement with human scores in our call-scoring research. They show the next scorer where the line sits on a call that does not look like the textbook.
Building the call-type scorecard, step by step
- List the call types you run. A typical B2B list is cold call, discovery, demo, proposal or pricing, negotiation, and renewal. Each gets its own short scorecard.
- Write the outcome each call type exists to produce. Discovery produces a problem the buyer has sized and a list of who else cares. A pricing call produces an agreed value case and a path to signature.
- Pull the skills that produce that outcome from the skill model. Expect five to eight. If you have more, merge rows that measure the same move, or move a skill to the call type where it shows up most.
- Write the levels for each row, in the observable style above, and add marker reasons from real calls.
- Score five to ten past calls with two managers, separately. Pick a mix of calls the team thought went well and calls it thought went badly.
- Compare row by row. Where the managers disagreed, read the reasons. Different scores for the same reason usually mean a vague level. The same score for different reasons means the row is measuring two things.
- Cut or rewrite. Remove any row that came back N/A on most calls or at the top level on every call. Rewrite any row the managers read differently.
- Re-score the same calls after every change, so you know the edit helped.
Expect it to take an afternoon for one call type, with most of that time going on step 4.
Example rows for each call type
The rows below are a starting point drawn from common B2B methods, for illustration. Swap in your own method's terms; a Gap Selling team will name the current state and future state, a MEDDPICC team will name the economic buyer and decision criteria.
| Call type | Outcome the call exists to produce | Typical rows (five to eight) |
|---|---|---|
| Cold call | A reason to talk again | Opener tied to the buyer's business, permission, problem hypothesis, buyer's reaction explored, objection handled, next step |
| Discovery | A sized problem and a map of who cares | Agenda, problem discovery, impact, current process, stakeholders, decision process, next step |
| Demo | Proof that the product fixes the problem the buyer described | Recap of the buyer's problem, demo tied to that problem, buyer's questions explored, value confirmed in the buyer's words, objections, next step |
| Proposal or pricing | Agreement on value before price | Value recap, price presented against impact, buyer's reaction explored, decision process confirmed, paper process mapped, next step |
| Negotiation | Terms both sides can sign | Positions explored, trades instead of discounts, walk-away known, decision maker present, written next step |
| Renewal | A decision on the next term, with evidence of value | Outcomes since signature, open issues, stakeholders changed, expansion signals, next step |
"Next step" appears on almost every list on purpose. It is the one row that tells you whether the call moved the deal, and it is easy to score: a date in the buyer's calendar, owned by the buyer, or not.
Running a calibration session
Before the scores feed any report, get two or more managers to the same score on the same three calls, then rewrite whichever level caused a disagreement and add the winning quote as a marker reason. Plan an hour, and repeat it when a new manager joins. The full agenda is in how to run a call calibration session.
Turning the score into one coaching ask
The scorecard is finished when the rep knows the one thing to do differently on the next call. Picking that one row is a judgment, and three questions make it quicker:
- Which row, if it had gone well, would have changed what happens next on this deal?
- Which row has the rep missed on more than one recent call?
- Which row can the rep practice before their next conversation with this buyer?
When the answers point at the same row, that is the ask. When they differ, the first question usually wins, because the next call with this buyer is the nearest place the change can show up. Write the ask as a behavior ("ask how many hours the delays cost each week"), set the call it will be checked on, and score only that row when the next recording comes in. The rest of the scorecard can wait a week.
Should criteria be weighted?
Weight a row only when it clearly decides whether the call did its job, and keep the weights visible to reps. Equal weights are easier to explain and harder to game. The risk with a weighted total is that it averages a fail on the deciding row into a respectable number. Whatever you choose, report the row scores next to the total, because the coaching conversation runs on the rows.
A reasonable middle path for most teams: no weights, plus one row flagged as the row that matters for this call type. On discovery that is often impact. A call that misses it is a coaching call regardless of the total.
When a longer list makes sense
A longer list fits jobs other than coaching a single call. Certification is one: a new hire's ramp review may check the whole skill model across several recorded calls, and twenty rows spread over five calls is still a handful per call. Quality assurance in a regulated sale is another, where a compliance checklist of yes/no items sits alongside the skills rows and is scored separately.
The test is the same in both cases. Each call still gets scored on the rows it should show, and the long list is assembled from those short ones over time. A rep should never leave a review holding twenty scores and no idea which one to work on.
Signs your scorecard has the wrong number of rows
| What you see | Likely cause | Fix |
|---|---|---|
| Managers skip rows or rush the last few | Too many rows for one call | Keep the rows this call type should show |
| The same row is N/A on most calls | The row belongs to another call type | Move it; allow N/A only where the call type permits, and leave it out of the average |
| Every rep scores 4 or 5 on a row | The top levels describe something easy | Rewrite levels 4 and 5 so they need buyer evidence |
| Two managers give the same call different scores | Levels describe traits a listener cannot point to | Tie each level to evidence on the call, add marker reasons |
| Reps cannot say what to work on after a review | No coaching ask was set | Pick one row per conversation |
| Scores rise and deals do not move | Rows count what the rep did and ignore how the buyer responded | Add the buyer's response to the level descriptions |
N/A has a fair use. A buyer who ends the call after twelve minutes has not given the rep a chance at stakeholders, and scoring that row 1 punishes the rep for the buyer's calendar. Track the count. A row that is N/A on most calls belongs on a different call type, and a row that is never N/A where it sometimes should be is being filled in by habit.
Common mistakes
One scorecard for every role. SDR cold calls and AE pricing calls share almost no rows. A single sheet forces N/A across half of it for everyone.
Activity rows on a call scorecard. "Logged the call in the CRM" or "sent the recap within an hour" are worth tracking, on a KPI scorecard. On a call scorecard they inflate the total without saying anything about the conversation.
Changing the rows every month. Scores from before and after an edit stop being comparable. Batch changes, re-score the same sample calls, and note the date the scorecard changed.
Using the total to rank reps before managers agree. If two managers cannot reach similar scores on the same call, a ranking built on those scores mostly measures which manager reviewed which rep.
Six rows on a discovery call (illustrative)
The deal and scores here are illustrative. A rep runs a first discovery call with a logistics buyer. The team's skill model has 18 skills; the discovery scorecard pulls six of them. The call scores well on problem discovery and rapport, and at level 2 on impact: the rep heard "the delays cost us a lot" and moved on. The manager picks impact as the coaching ask, because the buyer's finance lead joins the next call and will want a number. Before that call the rep talks the deal through with the coach and leaves with one question that sizes the cost in the buyer's own words.
Here is how the six rows might read on that call, with the evidence a manager would note. The company, Calder Logistics, and the rep, Robin, are fictional.
| Row | Score | Evidence from the call |
|---|---|---|
| Agenda and rapport | 4 | Robin set a three-point agenda and the buyer added one item |
| Problem discovery | 4 | Buyer described late handovers between depots, in their own words |
| Impact | 2 | "The delays cost us a lot." Robin moved to the next question |
| Current process | 3 | Robin learned the handover is a spreadsheet, but not who keeps it up to date |
| Stakeholders | 3 | Finance lead named; operations director mentioned, role unclear |
| Next step | 4 | Follow-up booked with the finance lead, date in the buyer's calendar |
The total looks healthy. The row that matters for the next call is the one at 2, and it is the only one the manager raises with Robin.
Cutting your own list down
Start from the discovery call scorecard template or the general sales call scorecard, keep the rows your call type should show, and spend the saved time on level descriptions. The AI call scoring guide covers the rest of the system. To see a real discovery call scored on a short scorecard, talk to us.
FAQ
What should a sales call scorecard include?
A sales call scorecard should include the five to eight skills that call type is meant to show, a description of each level for every skill, and space for the quote that justifies each score. Add the reasons your best managers give for each level. Leave out activity counts, which belong on a KPI scorecard.
Should every criterion apply to every call?
Every criterion should not apply to every call. Build a scorecard per call type from a shared skill model, so a discovery call is not marked down for skipping negotiation. Missing stages scored as failures add noise to every report built on the scores.
How many levels should each criterion have?
Four or five levels often work, as long as each level describes what the rep did on the call. Our research scored levels 1 to 5 and asked a yes/no question at each step ("at least level 2? 3? 4? 5?"). That design is our checklist model, which changed only 1.7% of its scores on a re-run.
What is the difference between a call scorecard and a rep scorecard?
A call scorecard judges one conversation against the skills that call type should show. A rep scorecard tracks a person over a period: activity, pipeline, outcomes and the trend in their call scores. Keep them separate. Activity counts on a call scorecard inflate the total, and a single call's score on a rep scorecard overstates one good or bad day.
When should a criterion be yes/no?
A criterion should be yes/no only for compliance items, such as a required disclosure or a recording notice. Skills rarely split cleanly into done and not done, so most rows need the graded levels described above, where each level adds one observable step.
Can AI score a call against five to eight criteria?
AI can score a call against five to eight criteria if each level describes what the rep did on the call and the scorer is checked against your managers before anyone relies on it. Compare the tool's best-to-worst order of a set of calls with your managers' order, and score the same calls twice to see how many scores move. The method is in can AI score a sales call.
Who should write the scorecard criteria?
The managers who will score with it, working from the team's sales method, with enablement holding the pen. The marker reasons for each level should come from those managers' own reviews of real calls, because those reasons are what the next scorer needs to see.