How to build a sales skills scorecard reps and CROs both use
Key takeaways
- A 0–100 score per skill with a benchmark line at 70 is easier for reps, managers and CROs to act on than adjective bands like Developing or Strong.
- Score each skill from three to five observable behaviors, and show no score until there are at least three scored instances.
- Draw the benchmark by cohort, because an organization-wide average hides the team that is struggling.
- Score per speaker, not per call, or team selling will inflate the rep's scorecard.
- Skills that come up less than once a week should be scored in roleplays first, not waited for on live calls.
A sales skills scorecard is only useful if two very different readers use it. The rep needs to know what to do on the next call. The CRO needs to know which teams are short of which skills, and whether closing the gap moves quota. Most scorecards serve neither, because they use adjective bands that nobody can tell apart. This guide shows how to build one that serves both.
What is a sales skills scorecard?
A sales skills scorecard is a regular record of how well each rep performs a small set of defined selling skills, based on evidence from real calls and scored practice. It is not a list of completed courses. It is the behavior level of the four levels of evidence, made visible per rep and per team. The glossary has the short definition.
Why does a 0–100 score work better than rating bands?
A number works better because people can compare it, trend it and try to beat it. Most assessments start with descriptive bands such as Developing, Consistent, Strong and Standout. In practice the same feedback comes back again and again: reps do not know what "Developing" means for them, managers cannot compare two people, and leaders cannot see a trend.
A score from 0 to 100 per skill fixes all three. Add a single index across skills for the overview, and draw a benchmark line at 70. Reps see their own numbers. The line gives them something to beat.
The concession: a number invites gaming. That is why call evidence sits underneath it. Faking a behavior across ten real calls is much harder than passing a quiz.
What are the five layers of a scorecard?
Build the scorecard from the bottom up. Each layer depends on the one below it.
| Layer | What it shows | Who reads it most |
|---|---|---|
| 5. Quota link | Skill score against attainment, by cohort | CRO, sales leadership |
| 4. Benchmark by cohort | Share of each team at or above 70 | CRO, second-line managers |
| 3. Score per skill | 0–100 for each skill, visible to the rep | Rep, frontline manager |
| 2. Call evidence | Which behaviors showed up in which real conversations | Rep, frontline manager |
| 1. Raw activity | Completions and practice attempts | Enablement |
Raw activity is necessary but never sufficient. The quota link at the top is what makes the CRO keep reading.
How do you score a skill from 0 to 100?
Define each skill as three to five observable behaviors, rate each behavior on each call or roleplay, and convert the total to a percentage. Observable means two people listening to the same call would agree whether it happened.
Rate each behavior 0 (absent), 1 (partly done) or 2 (done fully). Then:
Skill score = points earned ÷ points possible × 100, averaged over the rep's last 10 scored instances.
Show no score until a rep has at least three scored instances. One call is a hint, not a measure.
Example: the skill "get the buyer to name the cost of the problem" has four behaviors:
- Asks the buyer to estimate the cost in their own numbers.
- Plays the number back and confirms it.
- Ties the number to a date or event.
- Puts the number in the follow-up email.
A call that earns 2, 2, 1 and 1 scores 6 of 8 points, or 75. Across the last five instances the rep scored 75, 50, 63, 88 and 75, so the skill score is 70. With a four-behavior rubric, the benchmark of 70 means roughly six of eight points: most behaviors done fully, none ignored.
The line at 70 is our working benchmark, not a law. After two quarters, check where your top-attaining reps sit and move the line if the evidence says so.
Where should the evidence come from?
Evidence should come from wherever the behavior happens: live calls for frequent skills, scored roleplays for rare ones, and email or documents for behaviors that never reach a recording.
Count the moments before you rely on live calls
Some skills do not appear on enough calls to measure or learn live. Example arithmetic: a rep has about 5 customer calls a week, so 30 calls in a six-week cycle. If the moment for a skill appears in 1 call in 10, the rep gets 3 real chances in six weeks. Three attempts are not practice; they are three exams.
Our rule: if a skill comes up less than once a week, score it in a roleplay or simulator first and use live calls to coach the moments that do happen. Your own call data will give the real frequency per skill, and it varies a lot. AI roleplay makes the practice layer cheap enough to run weekly.
Score per speaker, not per call
Call scores usually attach to the call, and the call belongs to the rep. When a manager or specialist joins and asks the best discovery questions, the rep's scorecard gets the credit. The rep may have spoken less than a fifth of the words. Their score rises while the gap stays put, and nobody finds out until their next solo call.
Credit each behavior to whoever demonstrated it, and show talk share by person. Fix speaker labels first; mislabeled speakers turn a per-speaker score into a new kind of noise.
Include evidence that never reaches a recording
A call-only scorecard rates by silence: no evidence of a problem reads as no problem. Several deal-losing behaviors live elsewhere. Was the proposal walked through live or only emailed? Were concessions traded or given? Did the follow-up end with a dated ask? Is the buyer's own number in the business case? Start with three such behaviors and a human reviewer.
How do you keep managers' scores consistent?
Calibrate every quarter. Ask all managers to score the same five calls with the rubric, then compare. A biostatistics guide to interrater reliability notes that many texts treat 80% agreement as the minimum acceptable, and suggests that a kappa below 0.60 indicates inadequate agreement (>McHugh, 2012). Below 80% agreement on a behavior, rewrite that behavior's definition before you blame the managers.
Keep self-ratings out of the score. A meta-analysis of training transfer found that relationships were consistently inflated when the same source rated transfer in the same context (>Blume et al., 2010). A rep scoring their own call, or a manager scoring only the reps they coach, is that kind of source. Mix in scores from the AI rubric or a second reviewer.
Example: the cohort view a CRO reads
Example: one skill, four teams, end of quarter.
| Cohort | Reps | Median score | Share at or above 70 | Change vs last quarter |
|---|---|---|---|---|
| Team A | 25 | 74 | 64% | +12 points |
| Team B | 30 | 58 | 30% | +2 points |
| Team C | 20 | 71 | 55% | +9 points |
| Team D | 25 | 69 | 48% | +6 points |
| All reps | 100 | 67 | 49% | +7 points |
The organization median of 67 looks close to the line. The cohort view shows the real story: Team B has barely moved. That is where coaching time should go next quarter, and it is invisible in the org-wide average.
The quota link is the top layer. Example: reps at or above 70 reached 91% of quota on average, and reps below 70 reached 78%. Treat that as a correlation. Tenure and territory also drive both numbers, so compare reps of similar tenure before you claim a link. Kirkpatrick Partners makes a related point: the results that matter should be agreed with stakeholders in advance, not found afterward (>Kirkpatrick Partners).
Who sees what?
| Reader | What they see | What they do with it |
|---|---|---|
| Rep | Own scores, the line at 70, evidence clips | Picks one behavior for the next call |
| Frontline manager | Team scores, each rep's weakest skill | Runs one coaching conversation per rep per month |
| CRO | Cohort share at benchmark, trend, quota link | Moves coaching and program time to the weakest cohort |
The manager row connects directly to coaching coverage. The CRO row belongs on the monthly enablement report.
Common mistakes
- Using bands nobody can tell apart. Use a number and one line.
- Scoring too many skills. Four to six at a time; ten spreads coaching too thin.
- Crediting the call, not the speaker. Team selling then inflates every rep's score.
- Publishing individual rankings. Show cohorts to leaders and own scores to reps.
- Using scores as a promotion verdict. They are an input to coaching. Once pay depends on them, the evidence gets worse.
The limit: a scorecard measures behaviors you can observe. Judgment that happens before any call, such as account research and prospecting choices, mostly stays invisible. Keep managers' judgment in the loop.
If you want help sizing the work of building and running a scorecard like this, the free 20-minute Capacity Check is a good first step.
Frequently asked questions
What should a sales skills scorecard include?
Four to six skills that matter for the current plan, each defined by three to five observable behaviors, scored 0 to 100. Add a benchmark line, a cohort view for managers and leaders, links to the call or roleplay evidence behind each score, and a quarterly view of score against attainment.
Should sales reps see their own skill scores?
Yes. A score the rep cannot see becomes a judgment made about them, not a tool for them. Show each rep their own scores, the benchmark line and the evidence clips. Do not publish rankings of individual reps.
Why use a number instead of rating bands?
Bands like Developing, Consistent and Strong are hard to tell apart, hard to compare across people and hard to trend. A number shows movement, a benchmark line gives the rep something to beat, and leaders can average it by cohort.
How many skills should a sales scorecard track?
Four to six at a time. More than that spreads coaching too thin and makes every score noisier. Retire a skill from active coaching once most of the cohort sits above the benchmark for two quarters.
Can AI score sales calls accurately?
AI can score observable behaviors on calls reasonably well when the rubric is specific and speakers are labeled correctly. Check it the same way you check managers: score a sample of calls by hand and compare. Treat any score from a single call as a hint, not a verdict.
Should skill scores be used in performance reviews?
Use them as an input to coaching, not as a promotion or compensation verdict. Once scores decide pay, reps start optimizing the score instead of the skill, and the evidence gets worse.
