Recruiting
Interview the users who dropped off.
- Pick users by what they did
- Users know it is AI and agree before it starts
- Behavior can trigger invites, with limits you set
✦ AskWho closed the paywall last week without upgrading?
Set your own questions and triggers. The agent runs the interviews, then sums up the findings and what to do next.
The agent turns your question into an interview guide and tests it with simulated users before launch.
Pick users by what they did, or let their behavior trigger the invite.
Choose the languages for your study. Participants can answer by text or voice, and the agent asks follow-up questions.
Each finding shows how many users said it, their quotes and who disagreed, so you know what to fix.
A sample study: 64 of 120 interviews done, three draft findings with their quotes, and answers from Japan, the US, Brazil and Germany.
Choose where users dropped off and invite that group. The agent asks follow-up questions, and you can read the quotes behind each finding.
Recruiting
✦ AskWho closed the paywall last week without upgrading?
Follow-up questions
✦ AskWhy did users rate the new checkout 3.1 out of 5?
Findings
✦ AskWhat do users say about the new onboarding?
Users answer by text or voice in their own language.
Share of answers by language
United StatesEnglish
The export button was hard to find
JapanJapanese
セットアップが長すぎました
South KoreaKorean
보상이 너무 적었어요
SpainSpanish
Me encantó, pero es un poco caro
BrazilPortuguese
Adorei o visual do app
GermanyGerman
Die Einrichtung war zu lang
FranceFrench
J'ai arrêté au bout d'une semaine
TürkiyeTurkish
Uygulamayı çok sevdim
長すぎましたun poco caro너무 적었어요Users answer in their own languageEvery market in one report
Report · Why new users stop at setup
✦ Agent · translated for the report
Quotes keep the original language.
One report holds answers from 16 markets in 12 languages, including the United States, Brazil, Germany, Türkiye, India, Japan and Indonesia, each quote in its own language.
Users join from a link, and the agent writes the transcripts.
Traditional interviews
One at a time
User Interview Agent
Side by side
Traditional interviewsOne researcher runs one interview at a time
User Interview AgentSide by side across time zones · 00:00 UTC
Tokyo09:00
Sydney11:00
São Paulo21:00
Los Angeles16:00
16 interviews running at onceIllustrative comparison of a 30-interview study.
Recruiting and scheduling
Before: Finding a time with each person
With the agent: Users join from a link
Running interviews
Before: One scheduled conversation at a time
With the agent: Participants can answer in parallel
Transcripts and notes
Before: Typed up by hand
With the agent: Automatic, organized by theme
Analysis and report
Before: Written up by the researcher
With the agent: Drafted by AI, confirmed by a researcher
The User Interview Agent answers from past findings first and adds each approved finding to the knowledge base.
An example: asked why new users quit before finishing setup, the agent finds three quotes in past studies that answer two causes, marks the setup checklist as still open, and drafts a new study for it.
See where users drop off before you ask them why.
Explore Product AnalyticsTest the fix your interviews point to.
Explore A/B TestingHear what users say in public, too.
Explore Social ListeningAfter a change ships, Agentic Engine behavior data shows whether each finding held up.
What the user said
“I still love the app. I’ve just been busy lately.”
Said vs. did · Returning users study
What the behavior data showsDaily logins
Day 1Day 28
Logged in daily until the challenge ended, then dropped to once a week. Never opened the new feature.
New featureVisits0
Setting up the new feature put returning users off
ConfirmedChallenge rewards felt weaker than last year's
UnprovenThe premium price is the main reason for leaving
OverturnedCompare public feedback with interview responses to see where people describe the same problem.
Ask users who stopped at the same step what got in their way. Compare their answers with the behavior in your funnel.
Interview findings sit next to product data, so you can check what users said against what they did.
Use the interviews to choose a product change or a message, then test whether it helps.