When we first started using AI agents to moderate expert interviews, the reaction usually centered on two questions:
- Do experts actually want to talk to an AI?
- Can an AI agent reach the same level of insight as a human interviewer?
Both are fair questions.
Our team brings experience from Blackstone, BCG, McKinsey, and Bain. We know that a good expert interview involves much more than reading questions from a discussion guide. The interviewer needs to listen, understand what matters, ask the right follow-up questions, and know when an answer is worth exploring further.
The natural assumption is that an AI moderator will feel robotic, struggle with nuance, and produce shallower insights than a human.
We could have argued otherwise. Instead, we asked the experts.
We collected feedback after 320 interviews
The interviews spanned 12 commercial due diligence projects and more than eight languages. At the end of each one, we asked the experts:
- How satisfied were you with the interview?
- How did it compare with a traditional human-led interview?
Of the 320 experts, 298 provided a numeric satisfaction score. For the comparison with human-led interviews, we included responses from 279 experts who had previously participated in traditional expert-network interviews.
The results greatly challenged the conventional assumption.
Satisfaction score
How satisfied were you with this interview, on a scale of 1 to 5?
- 5 — Very satisfied63%
- 428%
- 36%
- 23%
- 1 — Very dissatisfied0%
Against a human-led interview
Was this AI-led interview better, worse, or the same as a traditional human-led interview?
- Better46%
- Same32%
- Mixed5%
- Worse16%
Experts did not just tolerate the experience
Of the experts who provided a satisfaction score, 91% rated the interview a 4 or 5 out of 5. Almost two-thirds, 63%, gave it the highest possible score.
Among experts with prior experience of traditional expert-network calls:
- 46% rated the AI-led interview as better.
- 32% rated it as the same.
- 5% gave a mixed response.
- 16% rated it as worse.
In other words, 78% considered the AI-led experience the same as or better than a traditional human-led interview.
What experts valued most was surprisingly conventional
The most interesting part of the study was not the overall satisfaction score. It was why experts said the interviews worked.
The largest positive theme, mentioned in 36% of the coded feedback, was clear and well-structured questions. Experts valued questions that were precise, logical, and easy to understand.
One expert described the experience this way:
“The questions were insightful. They were very well put together, and the follow-ups were relevant.”
Another 19% highlighted the natural and conversational interaction. Experts said the moderator listened, summarized their responses, and asked relevant follow-up questions.
As one expert put it:
“It did not feel like a robot or robotic voice to me. You were able to listen and also summarize well.”
Scheduling flexibility was the third-largest theme. Because there is no interviewer whose calendar needs to be coordinated, experts can complete the conversation when it suits them.
One expert told us:
“I could do it in the evening, late at night. I could do it on my own terms.”
Others valued the moderator’s comprehension, the speed of the interview, and the reduced pressure of speaking with an AI. A small group said they felt more comfortable being candid because they did not feel judged or put on the spot.
| Theme | Share of mentions |
|---|---|
| Clear and well-structured questionsClear, precise, logical, and easy to understand | 36% |
| Natural and conversational interactionFelt like a real person; good dialogue and follow-ups | 19% |
| Scheduling flexibility and convenienceCould interview at any time, with no schedules to coordinate | 11% |
| Strong comprehension and listeningUnderstood answers, summarized accurately, adapted follow-ups | 8% |
| Efficiency and speedQuick, concise, to the point, with no wasted time | 6% |
| Less pressure, more comfortableLess nervousness, no judgment, room to be honest | 2% |
In conclusion, experts valued DiligenceSquared’s AI agent on the same fundamentals that make any interview good: clear questions, relevant follow-ups, active listening, and respect for their time.
Local language changes the level of detail
One of the most important advantages of AI-moderated interviewing is the ability to conduct research in the expert’s local language.
The DiligenceSquared AI moderator currently supports more than 30 languages.
This matters because there is a significant difference between being able to answer a question in English and being able to explain a complex market in the language you use every day.
When experts speak in their local language, they can be more precise. They use the terminology that is common in their market, explain nuances more naturally, and spend less time translating their thoughts before answering.
This is difficult to replicate with a traditional consulting model.
Consider a European market study involving interviews in France, Germany, Spain, Italy, the Netherlands, and Sweden. Even the largest global consulting firms will rarely staff a team with native or fluent interviewers across every relevant language.
The practical solution is often to conduct most of the interviews in English. That may be sufficient for basic facts, but it can reduce the level of detail and exclude otherwise relevant experts who are less comfortable working in English.
Our study covered more than eight languages. The positive feedback was not limited to English-language interviews.
The real benchmark
A common criticism of voice agents is that they cannot replicate a seasoned consultant who has conducted hundreds of interviews and instinctively knows where to probe.
There are two problems with that critique.
First, it uses a distorted benchmark.
During a typical diligence, expert interviews are not conducted by seasoned partners with deep subject-matter expertise. They are mostly conducted by associates who are new to the topic and often have limited interviewing experience.
Second, even highly experienced people face limitations that AI agents do not.
People take time to learn about a new market. Their execution varies between interviews. They forget to collect data points and cannot retain large context across or even within long conversations.
An AI agent can ramp up on the content instantly, apply the same rigor to every interview, retain massive context, and identify contradictions live.
If one expert’s statement conflicts with an earlier interview, a datapoint, or another piece of research, the agent can recognize that conflict during the conversation and probe it immediately. A person may only discover the contradiction later, while reviewing notes or preparing the final analysis.
Even against the idealized benchmark of a seasoned consultant, an AI agent has structural advantages.
Does that mean every interview should be conducted by AI?
No.
At DiligenceSquared, we believe there is real value in the first few highly exploratory interviews being human-led.
At the beginning of a project, the team is not only looking for answers. It is learning which questions matter.
Clients who come to us have typically already run these interviews. Where they have not, we can run them at DiligenceSquared as well.
Either way, those conversations set the foundation for the broader interview program. Once there is an initial understanding of the market, our consultants translate the emerging hypotheses and research priorities into a clear brief for the AI moderator.
The right division of labor
Our AI moderator runs interviews from beginning to end. It works from the research brief, asks the questions, follows up on the answers, and conducts conversations across more than 30 languages.
That makes it possible to run more interviews, with greater consistency and without the usual scheduling and language constraints.
But scale only matters if the interview still feels worthwhile to the person on the other side.
That is why the most encouraging result from this study is not simply that 91% of experts were satisfied. It is that the things they valued were the things that should matter: thoughtful questions, relevant follow-ups, strong comprehension, and control over their own time.
For systematic information gathering at scale, we believe AI agents are already better suited to the job.