The problem
Stakeholder interviews are the lifeblood of diagnostic consulting. Twenty conversations with people across an organisation reveal what the data alone cannot: the unofficial narratives, the hidden blockers, the things everyone knows but nobody says in the board report.
The analysis phase is where the value is created — but it is also where consultants lose the most time. A typical diagnostic engagement might involve 12 to 20 interviews, each producing a 45-minute transcript. Manually reading and coding these transcripts, identifying themes, and cross-referencing perspectives across seniority levels can take three to five days.
The risk with rushing this phase is that you miss important signals. A comment that one manager makes about a colleague's team might not seem significant in isolation, but when five other people make similar observations, it becomes a key finding. Manual analysis, especially when you are tired, misses these patterns.
This workflow uses AI to surface the patterns while keeping you in control of the interpretation.
The system
Step 1: Record and transcribe all interviews (Otter.ai)
Conduct your interviews with Otter.ai recording (with participant consent, as required). Otter.ai transcribes in near-real time and automatically identifies different speakers.
After each interview, spend five minutes in Otter.ai adding brief annotations:
- Any particularly significant quotes (use the highlight feature)
- Your initial interpretation or reactions
- Any claims that need verifying
Export each transcript as a text file at the end of the interview phase.
Step 2: Upload all transcripts to NotebookLM (NotebookLM)
Create a new NotebookLM notebook for the engagement. Upload all interview transcripts as source documents. NotebookLM will index all of them and allow you to query across the full dataset.
Start with broad synthesis questions:
"What are the most common themes across all these interviews? List the top five to seven themes with brief descriptions and note which interviews they appear in."
"What are the key points of disagreement or tension across these interviews? Where do different respondents have contradictory views?"
"What do senior leaders seem to believe that differs significantly from what middle managers believe?"
"Are there any topics that multiple interviewees raise without being directly asked?"
Step 3: Drill into specific themes (NotebookLM)
Once you have the main themes, interrogate each one:
"Tell me more about the theme of [theme name]. Which interviewees mentioned it? What were the specific concerns or observations? Were there any particularly striking quotes I could use in the report?"
NotebookLM will pull from the actual transcripts, so you can trace findings back to source.
Step 4: Build the synthesis narrative (Claude)
Take NotebookLM's theme analysis and paste it into Claude to build the narrative:
"Here is a thematic analysis of 12 stakeholder interviews for a consulting engagement on [topic]. Create a synthesis narrative for the diagnostic report that:
- Opens with the headline finding — the most important thing the data reveals
- Walks through each theme with evidence
- Calls out the key tensions or contradictions
- Closes with the implications for the client
Write this as consultant-quality narrative prose, not bullet points. The audience is the client's executive team."
The results
Before: Three to five days of manual transcript analysis and theme identification for a 15-interview diagnostic.
After: Four to six hours of structured analysis using NotebookLM and Claude, producing a more thorough and evidence-grounded synthesis.
The pattern recognition improvement is significant. Manual analysis tends to weight the most memorable interviews (usually the most senior or most articulate respondents) over the quieter voices. NotebookLM counts mentions across the full dataset, so a theme that appeared in eight transcripts ranks above a vivid anecdote from one powerful stakeholder. That leads to more accurate and defensible findings.