Voice, chat or human? How a consultancy scaled its method with AI interviewers
Vendbridge turned its own method into AI interviewers and tested them with 98 people. Voice beat chat: half the interview time, 58% more distinct needs. And per minute, voice surfaced at least two and a half times as many insights as an experienced human interviewer typically does. Run in parallel, AI interviews scale their method from a handful of interviews to an entire population.
Key takeaways
- ≥2.5×More insights per interview minute from the AI voice interviewer than from an experienced human interviewer
- 7.6 vs 14 minMedian voice conversation against chat, and voice still surfaced 58% more distinct needs
- 5 daysFor all 98 interviews, voice and chat, run in parallel; voice in four languages from one setup
- ~8.3/10Rating in both channels: people rated voice and chat alike
About the study
Qualitative interviews are one of the best ways to learn what people need, but they do not scale. A typical project needs 20 to 30 interviews per group, which takes six to eight weeks and many hours of senior staff. So most projects hear a handful of people and guess the rest.
Vendbridge, a Zurich strategy boutique, wanted to know whether AI interviewers can take over this work without losing depth. Together with moweb, which recruited the respondents from its panel, and Cartorga, Vendbridge ran a controlled test on an everyday decision: how people choose a holiday. Half the respondents spoke with an AI voice interviewer, half typed with an AI chat interviewer, both following Vendbridge’s method.
How it works
- Bring your method.The consultancy brings what makes its work good: the questions, the steps to explore and the rules for what counts as a finding. Here: Vendbridge’s CFI method.
- Build your interviewer.On Cartorga, the consultancy turns its method into its own AI voice or chat interviewer, without a development team.
- Scale it to everyone.The interviewer talks to a whole company or customer base at once, in several languages, at any hour.
- Analyse with the same method.Findings come out in the method’s own terms, each traced to what a person said. The consultant judges what they mean.
In this study, the voice interviewer ran on Cartorga. The chat interviewer was a prototype that Vendbridge built itself.
Does an AI interviewer surface as much as a human?
Per minute, it surfaces more. Each voice interview produced 8.1 insights (Job Metrics) on average, about one for every minute of conversation. From past CFI projects, Vendbridge puts an experienced human interviewer at about 0.3 to 0.4 per minute. Minute for minute, the AI voice interviewer surfaced at least two and a half times as much.
And AI interviews do not have to be long, because they can be many: they run in parallel, so interviewing ten times the people does not take ten times the weeks.
The human figure is an estimate from past projects, not a controlled comparison. A direct test with the same guide and the same analysis is next.
Can it scale to a whole company?
Yes. All 98 interviews, voice and chat, ran in parallel, at whatever hour suited each respondent, and were done in five days, from 22 to 26 September. The voice interviewer ran in German or Swiss German, French, English and Italian from one setup, with no extra work per language.
Do more interviews find more needs?
Scale also brings depth. After 49 interviews, each further voice interview still added about 1.8 new needs, and more than half of all needs came up in just one interview. The usual rule of 20 to 30 interviews was made for hour-long human interviews. With AI interviews, extra interviews cost little, so the point where new interviews stop adding needs can decide when to stop.
“AI did not replace our method. It finally let us use it with everyone, not just the twenty people we had time for.”
Beat Walther, Partner, Vendbridge
Do people like talking to an AI interviewer?
Yes, in both channels alike. Voice and chat were both rated about 8.3 out of 10 on average. Whether people spoke or typed, they rated the AI interviewer about the same.
In voice, 78% gave 8 or higher and 53% a 9 or a 10; the most common rating was 10. Only 4 of 36 rated it below 7, mostly because the AI misheard them or a question felt repetitive.
We said at the start of every interview that it was run by an AI and that people could stop at any time. None of the 98 respondents broke off.
“Nine, because you just have to talk, no need to write.”Voice respondent, rating 9/10, translated from French
Does age matter?
Every age group rated the voice interview between 8.0 and 8.8 on average, around the overall average of 8.3. Younger participants rated slightly higher, but the link between age and rating is weak and not significant.
In chat, too, we found no significant link between age and rating. But the chat averages swing much more from one age group to the next, from 7.6 to 9.4, while voice stays between 8.0 and 8.8.
Voice seems to more reliably offer a predictable experience to people of all ages.
“I would give an eight, because you as an AI responded quite well to my questions, as if you were a real person.”Voice respondent, rating 8/10, translated from German
Voice or chat?
Voice hears more, chat asks more precisely. People say about three times as many words as they type, so voice found more distinct needs, including rare ones: 80 needs came up only in voice, 38 only in chat. Chat answers shrank by half after the first fifth of the interview; voice answers stayed long to the end.
Chat has its own strengths. The question stays on screen, the answer arrives as exact text, and chat gave exact money amounts more often. That makes voice the better tool for discovering needs in depth, and chat strong for screening and collecting figures. A hybrid that shows and speaks each question is the next test.
Why not build your own?
Vendbridge tried. With AI coding tools, it built a working chat interviewer in two weeks, and it ran well as a prototype. Running it in client projects is another matter: data protection, tracing every interview, stable quality as the AI models change, and fixing errors all keep costing time.
“Building our own interviewer showed us where the effort lies. The prototype took two weeks. Making it fit for client projects, with data protection, traceability, stable quality across model upgrades and support, would take far longer and keep costing. Our conclusion as a consultancy: we build the method, not the tool.”Yann Wermuth, Vendbridge
What this means
- More insights per minute.The AI voice interviewer surfaced at least two and a half times as many insights per minute as an experienced human interviewer.
- Scale to everyone.Hear a whole company or customer base instead of a handful of people, in their own language, and let new needs decide when to stop.
- The method stays with the consultant.The consultancy designs, steers and judges. The AI runs the conversations, and people rate it alike whether they speak or type.
About the partners and the method
Vendbridge (Zurich) is a consultancy that measures what customers want, using its Customer-Focused Innovation (CFI) method, built over 20 years and more than 200 growth missions. moweb (Düsseldorf) has run online panels since 2004 and today reaches more than 3 million members in over 30 countries. Cartorga provided the platform on which Vendbridge deployed its method, to scale it to a much larger group of people sourced through moweb.
Respondents were leisure travellers from Germany, Austria and Switzerland, recruited from the moweb panel and split into two groups of 49. Both interviewers followed the same interview logic, and insights were extracted and merged with the same CFI rules for both channels.
← All resultsRead next: Will people talk to an AI interviewer? 99 calls say yes