What Do Consumers Really Want From Their Health Insurer? Lessons from 20 Churn Interviews—and What AI Changed.
When we began this project, the strategic question was simple but fundamental:
What do consumers really need and want when choosing a health insurance provider in Switzerland?
So we conducted 20 in-depth interviews with churned customers, each lasting around 30 minutes. Interviews were done online, in Swiss German and French. I personally conducted a large share of them, giving me the intuitive “face validity” needed to compare human vs AI findings later on.
Once all recordings were transcribed, we manually anonymised and edited them to preserve the original voice of the consumer—unbiased and unfiltered. This step, although time-consuming, proved essential. As I often say : LLMs perform far better when fed with carefully transcribed, high-quality data.
🔍 AI vs Manual Analysis: What Really Happened
Before using Insight-lab, my team and I had already produced a fully manual report on the same interviews. That gave us the perfect benchmark.
When we compared the two, here’s what surprised us most:
1. AI delivered 100% topic coverage
When analysing needs, barriers, and expectations, nothing was missed.
The automated report surfaced every theme the human team had identified—plus a few nuances we had recorded but deprioritised.
The completeness was absolute.
2. Seamless multilingual integration
The AI processed French and Swiss-German transcripts and produced a complete English (or German) report.
For anyone familiar with qual: this solves the long-standing bottleneck of bilingual manual analysis.
3. Weighted, meaningful summaries—beyond topic lists
The AI didn’t just enumerate themes—it prioritised them.
It correctly differentiated between core needs, secondary frustrations, and occasional comments—something we previously assumed required senior judgement.
4. Verbatim quotation extraction: 90% accuracy
The system retrieved authentic quotes extremely well.
Why not 100%?
Because large language models sometimes “polish” the wording, we improved the system with a new mechanism that works very differently: instead of letting the AI rewrite the quote, it now searches the original transcript and copy-pastes the exact sentence said by the respondent.
⚡ 60–70% Time Saved: The Real Business Impact
The biggest surprise wasn’t the quality—it was the efficiency.
AI allowed us to save 60 to 70% of the time we would normally invest in analysing long transcripts.
Hours of page-by-page review were replaced with minutes.
That time can now be reinvested into higher-value work:
refining recommendations
connecting insights to business strategy
synthesising the “big picture” for stakeholders
Because—and this is essential— LLMs do not see the big picture.
They don’t understand brand positioning, market complexity, or business context.
This is where the senior researcher remains irreplaceable.
🎯 What Consumers Actually Expect from Their Health Insurer
Once processed, a clear hierarchy of needs emerged:
Foundational needs:
transparent pricing
clear communication
trust and predictability
Practical barriers:
complicated processes
inconsistent service between channels
reactive rather than proactive support
Emotional triggers:
the feeling of being “a number”
lack of empathy in health-related moments
repeated micro-frustrations leading to churn
These insights gave our client clarity they could act on immediately. Not in weeks—in days.
The Bigger Question
If we can analyse 20 multilingual interviews in minutes…
If we can achieve 100% coverage with weighted priorities…
If we can save 60–70% of analysis time…
Would faster qual analysis change how often you use it in your strategy cycles?