Call scorecard template — a ready-to-use example to download

Looking for a call scorecard template you can roll out on Monday, not theory to mull over? Below you’ll find a complete call center scoring sheet: 14 criteria in three formats, with weights summing to 100 points and a separate critical-error section. You can download the same template as an XLSX file and adapt it to your own campaign in an hour.

What is a call scorecard?

A call scorecard — in industry literature a call center scorecard, in practice simply a scoring sheet — is a standardized set of criteria against which a reviewer or a system evaluates every call the same way. Instead of the impression “that call was weak”, you get a breakdown: opening 5/5, needs discovery 2/5, recap of agreements — missing.

A good scorecard does three jobs at once. First, it standardizes evaluation: two reviewers scoring the same recording should arrive at the same result. Second, it turns feedback from opinion into fact — an agent will argue with “you weren’t very professional”; “you didn’t recap the agreements in 4 out of 10 calls” ends the discussion. Third, it generates data: results across weeks and months line up into trends per agent and per campaign.

One caveat before you copy the template: a scorecard is always derived from the campaign’s goal. A sales line polices the offer process, complaint handling — accuracy of information, collections — compliance. Our template is written for a sales-and-service call; you can swap the rows, but the structure (areas, formats, weights, critical errors) stays the same.

Three criterion formats in one scorecard

1. Yes/no checklist — did something happen

The format for binary behaviours: the agent introduced themselves or didn’t, disclosed the recording or didn’t, recapped the agreements or didn’t. In a typical scorecard the checklist makes up 50–70% of the items — 8 of 14 in our template. Its strength is zero room for interpretation: there is no such answer as “partially”.

2. Point scale — how well something was done

Where quality of execution matters — needs discovery, objection handling, clarity of explanations — a checklist flattens the picture. A 1–5 scale only works when every threshold is described: what exactly 1, 3 and 5 points mean. Without described thresholds, a scale is an impression dressed up as a number.

3. Critical events — red lines outside the scoring

A separate category for events that invalidate a call regardless of the rest of the score: a promise the company can’t keep, no recording disclosure, misleading the customer. A critical error doesn’t subtract points — it disqualifies the call and triggers an alert to the supervisor. That’s why these rows carry the note “disqualification” instead of a weight. This is how critical error detection in CallSea works: an alert with a transcript quote the same day, not at the monthly review.

Weights: how to distribute 100 points

Weights tell the team what really matters — and this is where scorecards most often break down, because “everything is important” ends with fourteen criteria worth 7 points each. A practical rule: formal obligations and behaviours that directly drive the call’s outcome (recording disclosure, identity verification, recap of agreements, next step) get 10 points each; style criteria and individual supporting behaviours — 5 each. In our template the scored criteria sum to exactly 100, so a call’s score reads like a percentage with no conversion.

Critical errors stay outside that pool deliberately. If “a promise the company can’t keep” carried a weight of 15, a call with that event could score 85/100 and look good. Disqualification plus an alert reflects the real cost of such an event better than any number of points.

An example call scorecard — the template

The full scorecard template: 14 criteria across 10 areas. The Score and Notes columns — empty in the template — are omitted in this preview:

Exactly this table — with empty Score and Notes columns to fill in — is waiting in the file: download the call scoring sheet (XLSX). Swap the rows for your own campaign and make sure the weights still sum to 100.

How to write criteria two people will score the same way

The template is half the work — the other half is how the criteria are worded. The test is simple: if two reviewers could score the same recording differently, the criterion is too vague. Everyone will interpret “the agent opened the call professionally” their own way; “the agent gave their name and the company name in the first 30 seconds” — no one.

A good criterion has three components: an observable behaviour (what exactly must be said in the call), a moment (when) and a pass condition (what counts as fulfilled). The same principle decides whether a scorecard can be automated — criteria written around impressions won’t work for a human or for AI. One constraint is hard: emotion and tone-of-voice criteria for agents must not be scored automatically at all — which is why in CallSea models read transcript text only, and a configuration validator blocks such criteria. We covered the craft of writing unambiguous criteria, with examples of good and bad wording, separately: how to design call scoring criteria for AI analysis.

Rolling the scorecard out in your QA team

A new scorecard deployed overnight causes two predictable problems: reviewers score by old habits, and agents dispute the results. The sequence that avoids both:

1. A pilot on 30–50 recordings. Before the scorecard goes into production, two reviewers independently score the same set of historical calls — deliberately varied, exemplary and problematic ones included.
2. Calibration criterion by criterion. Discrepancies usually sit in 2–3 items, not the whole scorecard. Refine the divergent criterion and repeat the trial on the same set — until the scores converge.
3. Communication to agents before launch. The team gets the scorecard with its weights in advance and knows exactly what they are evaluated on. A scorecard the agent knows is a development tool; a surprise scorecard — a source of conflict.
4. A review after every script or offer change. A scorecard policing a promotion that no longer exists measures noise. A steady review cadence (e.g. quarterly) plus a review triggered by campaign changes is enough.

The last decision is scale. Manually scored scorecards typically cover 1–2% of calls — enough for feedback, but trends and critical errors slip through between samples. Automated scoring against the same scorecard covers 100% of calls with results minutes after the call ends; the scoring formats — checklists, scales, events — are described on the call quality scoring page. A human can verify and correct every AI score, so the calibration from step 2 works the same way — just on full traffic.