An AI call scorecard is a short rubric, applied automatically to every cold call, that grades the behaviours which actually produce meetings. The hard part is not the AI. It is defining "great" from your own best calls instead of from a generic checklist. Here is how we build one, what to put in it, and how to coach from it without turning your managers into auditors.

What an AI call scorecard is, and what it is not

A scorecard is four to six criteria, each with two or three rating levels, scored on 100% of calls by a model that reads the transcript. The output per call is a score, a line of evidence per criterion, and one suggested improvement.

It is not a talk-time dashboard. Talk ratio, filler words and call length are easy to measure and weakly linked to meetings. It is not a script compliance tool either. Reps who recite the script word for word are the ones prospects hang up on.

And it is not a replacement for listening to calls. The score tells a manager which three calls to open this week. The manager still explains why.

Step 1: find your real top performers

Rank reps by meetings booked per connect, not by dials. In our benchmark of 1,000,000 B2B cold calls in France, the average connect rate is 11.3%, and the gap between reps on that number is small. The gap on conversation-to-meeting is huge: trigger-based openers reach a 7-10% meeting rate, pitch-first openers 2-3%.

So your top performer is not the rep who makes 200 dials. It is the rep who turns 20 conversations into two meetings while the team average is 0.6. Pull two or three of those reps. If you have a team of four, pull one.

Step 2: pull 20 calls and extract the decisions

Take 15-20 recent connected calls from those reps, including the ones that did not book. Listen for decisions, not phrasing:

• The first 15 seconds. Do they name a trigger (a job change, a hiring post, a website visit) or do they explain what they sell? 65% of reps still do the latter.
• The first objection. Do they answer it, or do they ask a question first? Asking "What do you mean?" before responding lifts conversion by 197% in our data.
• The discovery question. Questions that start with "how" convert 190% better than closed ones.
• The close. Do they ask for the meeting with two concrete slots, or do they end on "I'll send you an email"?

Write each decision down as a behaviour a stranger could observe. "Builds rapport" is not observable. "Names a specific trigger before stating the company name" is.

Step 3: turn behaviours into criteria a model can score

Each criterion needs a definition and rating levels with examples. Here is the rubric we see work for French B2B teams, in five lines:

1. Trigger opener. 0: pitches the product. 1: mentions a generic reason. 2: names a specific event about the prospect.
2. Permission and context. 0: talks over the prospect. 1: asks for time. 2: asks for time and states why the call is relevant now.
3. Objection handling. 0: argues or folds. 1: answers correctly. 2: asks a clarifying question first, then answers.
4. Discovery. 0: no question. 1: closed question. 2: at least one open "how" question that the prospect answers for more than ten seconds.
5. Next step. 0: "I'll send an email". 1: asks for a meeting. 2: proposes two slots and confirms one.

Include the objections your market actually produces. In France that means "envoyez-moi un mail", "on a déjà un prestataire" and "ce n'est pas le moment", not the objection list from a US sales book.

Then test the rubric on 30 calls the model has not seen, and compare its scores with a manager's. If they disagree on more than one call in five, the criterion is written badly, not the model.

Step 4: coach from the score, not from the whole call

Once every call is scored, the weekly routine changes. The manager opens the team view on Monday, sees that criterion 3 is the lowest for two reps, and picks one call per rep where they scored 0 on it. That is a 20-minute one-to-one with a single ask: this week, ask a question before you answer the objection.

One criterion per rep per week. Reps who are told to fix five things fix none.

Track the criterion score and the meeting rate side by side. If objection handling goes from 0.8 to 1.6 average and meetings per connect do not move in four weeks, the criterion is not the lever. Replace it.

Three mistakes that kill scorecards

• Scoring what is easy. Talk ratio and call duration are seductive because they are precise. They do not book meetings.
• Freezing the rubric. Your objections change when your pricing changes. Review the criteria every quarter, and every month during a product launch.
• Using it as surveillance. If reps only hear about scores when they are low, they will game the rubric. Share the top-scored call of the week with the whole team first.

How Sonik approaches call scorecards

Sonik records and transcribes every call made through the dialer, in French and English, and scores it against a rubric you define. The default rubric is built from the behaviours that moved meeting rates in our 1,000,000-call benchmark: trigger opener, question before the objection, open discovery, concrete next step.

Managers get a team view by criterion, a list of the week's best and worst calls per criterion, and clips they can send to a rep in one click. The score and summary are written back to the CRM so the next person who opens the contact sees what happened.

We do not score talk ratio by default. You can add it, but in our experience it distracts from the four behaviours above.

Conclusion

A scorecard is only as good as the definition of "great" behind it, and that definition lives in your top reps' calls, not in a template. Pull 20 calls this week, write down the five observable behaviours that separate your best rep from the middle, and score 30 calls by hand before you automate anything. Then coach one criterion per rep per week and watch meetings per connect, not the score itself.