Short answer: AI sales call scoring checks each conversation against a scorecard: weighted yes/no criteria grouped into categories, with a different scorecard per call type. The AI reads the transcript, marks each criterion, quotes the evidence and adds up the weights. It helps coaching when criteria describe observable facts and weights reward what moves the deal.
Quarterly review. The sales manager opens the call evaluation sheet: the whole team scores between 8 and 9.5 out of 10. The same sheet shows discovery-to-proposal conversion fell from 40% to 25%. One of those numbers is wrong. We hear the same complaint from sales leaders all the time. One of them put it this way: "Now it looks like we all get a 10."
Quick summary:
- AI call scoring works in four steps: transcript, criterion-by-criterion evaluation with evidence, weighted score, and patterns across many calls.
- A scorecard that always scores high is useless for coaching. The usual causes: vague criteria, wrong weights and one scorecard for every call type.
- Below are four copy-ready templates (cold call, discovery, demo, negotiation) with binary criteria and weights that add up to 100.
- Calibrate before you trust it: score five calls by hand, compare and adjust.
- Samu scores every call with the scorecard for its type, with a quote per criterion, and shows each rep's Main Gap on a coaching dashboard.
How does AI call scoring work for sales coaching?
The mechanics are simple:
- Capture and transcribe. The call is recorded (video call, phone, in-person meeting) and transcribed with speaker labels.
- Evaluate each criterion. For each yes/no question in that call type's scorecard ("Did the buyer give a number that sizes the impact?"), the model looks for evidence in the transcript and marks it met or not met, citing the passage.
- Weight and add up. Each criterion carries a weight; the call's score is the sum of the criteria met.
- Aggregate. Across dozens of calls, the system shows which criterion each rep misses most. That becomes the topic of the next one-on-one.
The model is only as good as the scorecard. For a broader view of what AI coaching can and can't do, read AI for sales coaching: what to expect.
Why does everyone on the team score a 10?
When a scorecard scores the whole team high, the problem is almost always in its design.
1. Vague criteria
"Ran a good discovery." "Handled objections well." "Built rapport." Each evaluator reads these differently, and when in doubt, almost everyone says yes. A useful criterion describes something that did or didn't happen on the call and that two people would mark the same way. "Did the buyer give a number that sizes the problem (hours, money, customers, percentage)?" leaves little room. That's also why 1-to-5 scales drift: no fact separates a 3 from a 4, and over time everything converges on 4.
2. Wrong weights
If "introduced themselves and their company" weighs the same as "agreed on a dated next step," a rep can hit every easy criterion, miss the one that decides whether the deal moves and still score 85. A practical rule: the criterion most tied to the deal advancing should weigh at least three times the lightest format criterion.
3. One scorecard for everything
A four-minute cold call and an hour-long negotiation can't be judged with the same list. With a single scorecard, half the criteria don't apply on any given call, evaluators mark them as met or skip them, and the average climbs.
What does a scorecard that doesn't discriminate cost?
Unfocused coaching: if everyone scores 9, the one-on-one goes back to "how's it going?" Lost credibility: when the score never moves, the team stops looking. And no early warning: if a rep's discovery calls never reach the buying process, a good scorecard tells you weeks before the forecast does.
Want to see your scorecard run on your team's real calls? Book a demo and we'll review it with your data →
Sales call scorecard templates by call type
Weights add up to 100 in each and reward what moves the deal forward. Adjust them after calibrating.
Cold call (SDR)
Goal: book a meeting with the right person.
| Category | Criterion (yes / no) | Weight |
|---|---|---|
| Opening (20) | Said who they are and why they're calling within 30 seconds? | 10 |
| Opening | Gave the prospect an out or asked permission to continue? | 10 |
| Problem (35) | Named a specific problem that similar companies have? | 15 |
| Problem | Asked how they handle it today? | 10 |
| Problem | Prospect confirmed or ruled out the problem in their own words? | 10 |
| Objections (15) | Asked a question before answering the first objection? | 15 |
| Close (30) | Proposed a meeting with a specific day and time? | 20 |
| Close | Confirmed who attends and how the invite will arrive? | 10 |
Discovery (AE)
Goal: understand the problem, its impact and how the customer buys.
| Category | Criterion (yes / no) | Weight |
|---|---|---|
| Opening (10) | Agreed on agenda and time at the start? | 5 |
| Opening | Asked what the buyer wanted from the meeting? | 5 |
| Situation (10) | Asked how the process works today and what tools they use? | 10 |
| Problem and impact (35) | Buyer described at least one specific problem in their own words? | 10 |
| Problem and impact | Asked at least two follow-up questions about that problem? | 10 |
| Problem and impact | Buyer gave a number that sizes the impact? | 15 |
| Buying process (20) | Asked who else is involved and who signs? | 10 |
| Buying process | Asked about formal steps (procurement, legal, security)? | 10 |
| Urgency (10) | Asked why solve it now, or what happens if they do nothing? | 10 |
| Close (15) | Summarized and the buyer confirmed? | 5 |
| Close | Meeting ended with a dated next step? | 10 |
Demo
Goal: show how the problems the buyer described get solved. More in how to run a B2B sales demo.
| Category | Criterion (yes / no) | Weight |
|---|---|---|
| Context (20) | Restated the discovery problems and the buyer confirmed them before showing anything? | 15 |
| Context | Asked whether anything changed since the last meeting? | 5 |
| Demonstration (35) | Showed the end result before the setup? | 10 |
| Demonstration | Tied each feature shown to a problem the buyer mentioned? | 15 |
| Demonstration | Paused at least twice to ask how it would apply to their team? | 10 |
| Attendees (15) | Asked new attendees what they expect from the solution? | 10 |
| Attendees | Clear who makes the decision? | 5 |
| Objections (10) | Explored each objection with a question before answering? | 10 |
| Close (20) | Agreed on what has to happen inside the buyer's company to move forward? | 10 |
| Close | Meeting ended with a dated next step? | 10 |
Follow-up and negotiation
Goal: reach a dated decision without giving away margin.
| Category | Criterion (yes / no) | Weight |
|---|---|---|
| Status (15) | Reviewed both sides' commitments since the last meeting? | 15 |
| Committee (25) | Confirmed who approves and who signs? | 10 |
| Committee | Asked about pending approval steps and their dates? | 15 |
| Value (20) | Restated the impact the buyer sized before discussing price? | 20 |
| Negotiation (20) | If the buyer asked for a discount, asked for something in return? | 10 |
| Negotiation | Avoided offering a discount nobody asked for? | 10 |
| Close (20) | Agreed on a decision date? | 10 |
| Close | Left with a plan of steps, owners and dates on both sides? | 10 |
Reading the result: a discovery call that meets everything except the impact number (15) and the formal buying steps (10) scores 75. That 75 tells you exactly what to work on in the next one-on-one.
How do you calibrate a scorecard in one hour?
- Pick five calls of the same type: two you consider good, two bad and one in between.
- Score them by hand, criterion by criterion, noting the minute each thing happened.
- Compare the score with your judgment. If a call you consider bad scores above 70, an easy criterion is overweighted.
- Find the criteria you hesitated on. If a yes/no took more than ten seconds, rewrite it as an observable fact.
- Have someone else score the same five. Where you disagree, rewrite next.
Recalibrate when the pitch, product or customer profile changes, and leave the scorecard alone in between.
How does Samu automate call scoring?
With ten reps and several meetings each a day, nobody listens to all of them. Samu applies the same scorecard to every call.
Custom Score by call type
In Samu, each call type has its own score, built from the panel without filing tickets: weighted categories, weighted binary criteria inside each one. You can load your methodology (SPICED, BANT, MEDDIC, Sandler or your own) as criteria. Because capture includes phone calls and in-person meetings, field visits can have their own scorecard, separate from a remote demo.
Evidence for every criterion
Each criterion comes with the supporting quote, plus a narrative evaluation of the call. That settles "but I did ask about budget."
Coaching dashboard with Main Gap
The manager dashboard shows a team traffic light, the reps who need coaching and each rep's Main Gap: the criterion they miss most. Analytics add talk ratio, question rate and duration, and downloadable reports show weekly score, by criterion, by rep and weighted.
Level Up and the Library
With Level Up (beta), reps see their weakest criteria and recordings of teammates who scored 100% on them. The best calls go into Library collections for onboarding.
Nubox, in Chile, measures 30,000 to 40,000 interactions a month and coaches 50 reps with Samu. Its CRO, Felipe Gómez, put it this way: "Quality stopped being a feeling and became a data point." (see case studies)
What Samu doesn't do
Samu doesn't decide what a good call looks like for your team. You define criteria and weights, and if they're poorly written, Samu will precisely evaluate an imprecise standard. Calibration is still the manager's job. It also doesn't replace the one-on-one: Samu tells you which criterion fails, on which calls, with which quote. The conversation, the role-play and the decision about what to work on first stay with you.
FAQ
How does AI call scoring work for sales coaching?
The AI transcribes each call, applies the scorecard for that call type, marks each yes/no criterion with a supporting quote and adds the weights into a score. Across many calls, it shows which criterion each rep misses most, which becomes the coaching topic.
How many criteria should a sales call scorecard have?
Six to ten per call type. Past ten, reps don't know where to start.
Should I use a 1-to-5 scale or yes/no criteria?
Yes/no. Two evaluators mark a fact the same way, and the criterion's weight already provides the gradation.
Can I use SPICED, MEDDIC or BANT as a scorecard?
Yes. Each element becomes one or more binary criteria. The "M" in MEDDIC, for example, becomes "Did the buyer give a number that sizes the impact?" In Samu they load like any other criterion.
Can AI score calls in languages other than English?
Samu works in English, Spanish and Portuguese, so teams selling across the US, Latin America and Brazil can use the same scorecard structure.
Conclusion
A scorecard helps when it discriminates: binary criteria that describe facts, weights that reward what moves the deal and a separate list per call type. Calibrating takes an hour. Applying it to every call is what doesn't fit on a manager's calendar, and that's the part you can automate.
Book a Samu demo and we'll build your team's scorecard on real calls, or check plans and pricing.
