Automation & AI

Zoom Call Grading Pipeline

Recorded Calls to Graded Coaching Output

A Python pipeline that processed recorded sales calls end to end, transcribing them, grading the transcript, categorising objections and exporting results into the sheet the coaching team already worked from.

Industry
Sales Enablement
The problem

Sales coaching depended on managers listening back to recorded calls, which does not scale. Coverage was partial and inconsistent, and patterns across a week of calls were invisible.

  • Manual call review covers only a fraction of calls
  • Coaching quality varied with who reviewed and when
  • No consistent grading basis between reviewers
  • Objection patterns across many calls were not visible
  • Any new tool risked not being adopted by a busy team
What we built

An end-to-end pipeline that pulled recordings, transcribed them, graded each transcript against a rubric, categorised objections and wrote the output into the team's existing sheet.

  • Pull recordings automatically rather than waiting for uploads
  • Transcribe, then grade each transcript against a defined rubric
  • Detect and categorise objections so weekly patterns become visible
  • Export into the Google Sheet the coaching team already used
  • Keep grading consistent across calls rather than per reviewer

How it was built

Coaching that only covers a sample

Coaching from recorded calls requires someone to listen to them, so at real volume it covers a sample at best. The calls that most need review are the ones nobody reaches.

Into the tool they already used

Results export straight into the Google Sheet the coaching team already worked from, rather than introducing another interface to check. An automation that requires people to adopt a new tool tends not to survive contact with a busy week.

Objection categories make the week legible

Detecting and categorising objections matters less per call than in aggregate. A week of categorised objections shows a pattern that no individual call review would surface.

Before
  • Only a sample of calls reviewed
  • Grading varied by reviewer
  • Objection patterns invisible across a week
  • Coaching output assembled by hand
After
  • Every call transcribed and graded
  • Consistent grading basis across all calls
  • Objections categorised for weekly pattern review
  • Output delivered into the team's existing sheet
Stack
Pipeline
PythonBatch processing
AI
Speech-to-textTranscript gradingObjection classification
Integrations
Zoom APIGoogle Sheets API