Turn Call Transcripts into Coaching Notes

Tier: Core · Hands-on time: ~30 minutes · Tools: Claude, ChatGPT, Google Docs

Coaching from memory has two failure modes: the manager who only remembers the fumbles, and the manager who says "great call!" and coaches nothing. A transcript pass with forced balance — strengths and growth areas, every claim time-stamped — fixes both. The model finds the moments; you decide what they mean and what to do about them.

When to use — and when not to

• Regular call-coaching cadenceone call per rep per week reviewed properly beats ten skimmed.
• New-hire rampearly calls reviewed with timestamped evidence teach faster than shadowing, because the new hire can re-listen to the exact moment.
• A pattern you suspect but can't namedeals stall after demos, handle times creep up; several transcripts through the same lens can surface what.

And when not to:

• Performance-management documentation. Notes feeding a formal process (PIP, disciplinary) need HR's process and your own direct review of the calls — an AI summary is not evidence of record. Keep this recipe strictly developmental.
• Consent is unclear. The rep should know calls are recorded and reviewed for coaching. Surprise surveillance dressed as coaching destroys the trust coaching runs on.
• You haven't heard the call at all. The notes prepare the conversation; they don't replace knowing what happened. Listen at least to the moments the model flags before you sit down.

The recipe

1. 1.

   Pick one call and one focus (3 minutes). Choose a single representative call — not the week's disaster — and one coaching focus if you have it (discovery questions, objection handling, empathy statements, call control). One call, one focus, one conversation: coaching bandwidth is the scarce resource, not analysis.

2. 2.

   Run the balanced-review prompt. Paste the timestamped transcript:

   Copy

   You are a coach reviewing a [SALES/SUPPORT] call for a developmental
   1:1. The agent is [NAME/ROLE]. My coaching focus this month:
   [FOCUS, or "general"].
   
   Produce coaching notes with EXACTLY this structure:
   
   1. Call snapshot: 2 sentences — purpose of the call and outcome.
   2. What worked (exactly 3 points): each with the timestamp and a
      VERBATIM quote showing the skill in action, plus one line on why
      it worked.
   3. What to change (exactly 2 points): each with timestamp +
      verbatim quote, what a stronger alternative would have been —
      phrased as the actual words the agent could have said — and the
      likely effect.
   4. Pattern check: does anything here look like a habit vs a
      one-off? State the evidence or say "single call — no pattern
      claim possible."
   5. One suggested practice action for the week.
   
   Rules:
   - Balance is mandatory: 3 strengths, 2 changes, in that order.
   - Every point carries a timestamp and exact quote. No quote, no
     point.
   - Critique the behaviour, never the person. No mind-reading about
     attitude or motivation.
   - Judge only what the transcript shows.
   
   TRANSCRIPT: [PASTE]
   

   A good output looks like this:

   Copy

   What worked
   1. [04:12] "Before I answer, can I ask what's driving the
      deadline?" — turned a pricing demand into discovery; the
      customer revealed the real constraint (budget year end).
   ...
   What to change
   1. [11:47] "Yeah, so, our tool basically does that too" — answered
      a competitor mention defensively. Stronger: "What do you like
      about how Flowline handles that?" — keeps discovery going and
      surfaces the actual gap.
   ...
   Pattern check: single call — no pattern claim possible.
   

3. 3.

   Verify the flagged moments (7 minutes). Jump to each timestamp in the recording and listen. You are checking two things: the quote is real, and the tone matches the reading — text famously misses sarcasm, warmth, and a customer laughing. Cut any point the audio doesn't support. This step is what makes the notes yours rather than the model's.

4. 4.

   Rewrite the "what to change" points in your voice. These two points are the coaching conversation. Rework them into questions where you can — "at 11:47 the competitor came up; how did you feel that went?" lands better than a verdict. The model's "stronger alternative" wording is raw material for the conversation, not a script to read at someone.

5. 5.

   Store notes in a running per-person doc. One Google Doc per team member, newest on top. After four or five sessions, paste the accumulated notes back: "Across these coaching notes, what patterns repeat? What has visibly improved?" — that answer, quote-anchored across weeks, is the honest core of a development review, and the improvement half matters as much as the repeat half.

Make it yours

• Claude for full-length calls: hour-long support calls fit in one paste, and it follows the "exactly 3 / exactly 2" balance reliably. A Project holding the prompt plus your team's call-quality rubric makes every review consistent with your standards, not the model's defaults.
• ChatGPT for teams with rubrics: if your QA team scores calls against a form, include the form and ask for the balanced notes plus a provisional rubric score per criterion, each with its timestamp — the coaching notes and the QA sheet stop disagreeing with each other.
• Google Docs as the coaching system: the per-person running doc, shared with the team member after each session, turns coaching from a manager's file into a shared record — and the transparency itself improves how the notes get written.

Quality check

• Every point has a timestamp and a real quoteverified against the recording, tone included.
• The balance heldthree strengths, two changes; if the model drifted negative, you rebalanced before the 1:1.
• Changes come with alternative words, not just verdicts — "say this instead" is coachable; "be more consultative" is not.
• No pattern claims from a single callpatterns need multiple calls on the record.
• Behaviour, never personnothing in the notes speculates about attitude, effort, or motivation.