Score sales calls on evidence

Use this after a sales call when the team needs defensible coaching or deal evidence. Produce a rubric-based scorecard tied to transcript evidence; do not grade confidence, charisma, or transcript summaries.

Establish the evidence

Confirm the participants, account, opportunity, call type, date, and transcript quality. Separate buyer speech from seller speech. Mark inaudible sections, uncertain speakers, transcription errors, and missing call segments before scoring.

Link the call to a deal only when account, participants, timing, and context agree. If several opportunities could match, preserve the ambiguity. Do not write insights to the nearest deal merely because its name resembles the account.

Extract evidence as short verbatim spans with speaker and timestamp. Distinguish buyer-stated facts, seller claims, hypotheses, commitments, and unanswered questions. A seller describing the buyer’s pain is not buyer confirmation.

Score against observable behaviors

Choose the rubric for the call type before reading the outcome. Score discovery, validation, stakeholder coverage, business impact, decision process, next steps, and seller execution only where they apply. Use references/call-rubric.md for a four-level scale.

For each dimension provide:

• score and definition;
• supporting evidence;
• strongest contradictory or missing evidence;
• confidence in the score;
• one coaching action or deal question.

Award credit for observable behavior, not vocabulary. A scripted question earns little if the seller ignores the answer. A concise call can score highly when the buyer supplies clear evidence and next steps.

Separate coaching from deal truth

Create two outputs. The coaching view evaluates seller behavior and the next skill to practice. The deal view records buyer-confirmed problem, impact, stakeholders, process, risks, commitments, and open questions.

Do not convert inference into CRM fact. Proposed field updates should quote the supporting buyer evidence and remain reviewable. Human decisions include opportunity stage, forecast category, amount, close date, and any interpretation that materially affects reporting.

Read references/worked-scorecard.md for an example where polished seller execution does not compensate for missing buyer confirmation.

Calibrate reviewers

Have two reviewers independently score a sample of at least ten calls. Compare dimension-level differences and discuss evidence, not total scores. Clarify rubric anchors when reviewers differ by more than one level on the same dimension.

Maintain benchmark calls for each call type and score level. Re-score a stable benchmark set when the rubric changes. Report model or reviewer consistency separately from seller performance.

Produce the review

Lead with the three most important findings, then the scorecard, evidence, risks, next actions, and proposed record updates. Preserve timestamps so a manager can verify conclusions quickly. Name what cannot be concluded from the call.

Prefer one behaviorally specific coaching priority over a long list. Tie it to the next call: the question to ask, evidence to obtain, or commitment to secure.

What good looks like

• Every material claim traces to a speaker and timestamp.
• Buyer statements are distinguished from seller narration and reviewer inference.
• Scores use behavioral anchors and expose contradictory evidence.
• Coaching and deal inspection remain separate but consistent.
• Ambiguous opportunity links and low-quality transcripts reduce confidence visibly.
• Reviewers can reproduce scores within one level after calibration.

The mediocre version summarizes the call, rewards talk-track compliance, infers qualification from seller statements, and writes optimistic fields without review.

Rules

• MUST preserve speaker, timestamp, transcript-quality, and call-context evidence.
• MUST separate buyer-confirmed facts, seller claims, and reviewer inference.
• MUST require human approval for material CRM or forecast changes.
• NEVER invent quotes, timestamps, participants, or commitments.
• NEVER force an ambiguous call-to-deal link.
• NEVER score a dimension that the call type or available evidence cannot support.