Forty interviews done. Now what did they say?
The diagnostic phase of an engagement includes interviews with forty stakeholders across the client: executives, middle managers, frontline staff. Four consultants ran them. Some typed notes, some recorded and transcribed, one wrote by hand. Now the team has a few days to turn it all into themes for the interim readout.
They print everything, highlight, cluster sticky notes on a wall or a Miro board, argue about what counts as a theme, and try to remember who said the striking thing about decision rights. By the end, the synthesis is good but exhausting, and much of it depends on who read what.
Why synthesis is so laborious
- Notes differ in format, depth and style between interviewers.
- Themes emerge while reading, so early interviews are coded differently from late ones.
- Linking a theme back to supporting interviews is manual.
- Anonymising quotes takes care and time.
- The same work is repeated on every engagement with no reusable structure.
Balance is a quieter risk. Senior interviewees are often more articulate and more quotable, and their views can dominate a synthesis simply because they are easier to remember. Without a structured way of counting, frontline views get less weight than they should.
What slow synthesis costs
| Issue | Effect |
|---|---|
| Days spent rereading | Less time for analysis and recommendations |
| Themes shaped by who read what | Findings less balanced than they could be |
| Quotes hard to trace | Weak defence if a client challenges a finding |
| Anonymity slips | Trust with interviewees damaged |
Interviewees speak candidly because they trust the consultancy with what they say. Handling their words carefully is part of the job, not an afterthought.
How we build interview synthesis
- Interview notes and transcripts are uploaded to a secure workspace for the engagement, with each interview's role, function and level recorded.
- The team sets out its initial framework, such as the diagnostic areas in the SOW.
- A language model reads each interview and suggests which passages relate to which area, and proposes emerging themes that do not fit the framework.
- Consultants review, accept or change the coding, and add their own.
- Themes show how many interviews support them, split by role and function, with every supporting passage linked.
- Quotes are pulled with identifying details removed, and a consultant checks each one before it goes in a deck.
- The synthesis exports to your slide templates as a starting point for the readout.
The model is a fast first reader, not an analyst. It suggests; the consultants decide what the interviews mean. Transcripts and notes stay within your own environment, under terms that exclude training on the data.
We usually trial the tool on a past engagement's interviews, with the team that ran them, so they can compare the suggested coding with their own. That builds a sense of where the model helps and where it needs correcting before it is used on live work.
The readout week, afterwards
The team spends its time debating what the themes mean and what to recommend, rather than finding things. Every claim in the deck can be traced to interviews. Anonymised quotes are ready and checked.
The framework and coding approach can be reused on the next diagnostic, so the firm's method gets sharper with each engagement.
Does your diagnostic phase look like this?
- Interview synthesis takes days of rereading.
- Notes come in many formats.
- Themes are hard to trace to supporting interviews.
- Anonymising quotes is slow and nerve-racking.
- Each engagement reinvents its synthesis method.