Denis Kharchenko — Lead Product Designer
Questy · Case Study
Questy is an AI discovery tool for professionals who need to understand a client before starting the actual work.
I started working on it because I was already using several tools to run research, then moving the answers into AI to analyse them and turn the result into a brief.
Collecting answers was the easy part. The problem was everything that happened after.
My research workflow was spread across a few different tools.
I could use Dovetail, Typeform or Maze to collect responses. Then I would move the answers into AI, look for patterns and contradictions, identify what was missing, work through the findings myself, and eventually turn everything into a brief.
It looked roughly like this: Research tool → responses → AI analysis → manual synthesis → brief
Each tool handled part of the process. But the research didn't end when the questionnaire was completed.
That made me question what the actual product should be.
I wasn't looking for another questionnaire builder. I wanted to bring the research, the analysis and the follow-up work into one process.
A traditional questionnaire assumes that the questions are known before the research starts.
But sometimes an answer creates a new question.
A client might give a vague answer, leave out an important detail, introduce a contradiction, or use a word that means something different to different people.
For example: Q: What do you want your new brand to communicate? A: Something more premium.
The answer sounds reasonable, but it doesn't tell us what "premium" means to this particular client.
In a traditional questionnaire, the next step is usually manual. I have to notice the gap and decide what to ask next.
That was the part I wanted Questy to handle.
The first decision was to make the product project-first, rather than questionnaire-first.
A Project is the context and history of the work. A Discovery is a method used to understand the client within that project. One Project can contain multiple discovery stages.
That changes the mental model.
Instead of: Create questionnaire → collect answers → finish
the process becomes: Project → Discovery → Answers → Analysis → Clarification → Brief
A project can start with an initial discovery, continue with a second round of questions, and use the accumulated context as the work develops.
The questionnaire became one part of the process, rather than the product itself.
I designed three entry points around the different ways research starts.
Create: Start with a goal, context, methodology and questions. Questy can generate an initial questionnaire, but the user reviews and edits it before it becomes the actual discovery.
Import: Bring an existing questionnaire into Questy. This was important because the user's methodology should remain intact. Questy should add intelligence around the questionnaire, not replace it. The user can import, review and edit the questions, send them to the client, use Smart Follow-ups, analyse the answers and generate the Client Brief.
Analyze: Research doesn't always start in Questy. A user might already have a completed questionnaire and its answers. In that case, they can bring the research into Questy and start with analysis. Questy looks for missing information, ambiguity and contradictions. From there, the user can ask the client manually, create a short clarification questionnaire, or mark something as understood.
This meant the product could support the whole workflow without requiring the initial questionnaire to be created there.
The core interaction became simple:
Questy asks the questions you prepared, listens to the answers, and knows when it needs to ask one more question.
Using the earlier example: Q: What do you want your new brand to communicate? A: Something more premium.
Questy can recognise that the answer is still ambiguous: Smart Follow-up: What would make the brand feel more premium to you?
The additional question becomes part of the discovery and can later contribute to the Client Brief.
The important part is what Questy doesn't do.
It doesn't turn the discovery into an open-ended AI chat. It doesn't silently rewrite the questionnaire. It doesn't decide that some of the user's questions are unnecessary.
The user's questionnaire remains the user's questionnaire. Questy adds intelligence on top of it.
That keeps the person running the research in control of the methodology.
Once AI can ask follow-up questions, there is an obvious problem.
It can keep asking.
A 15-question questionnaire could easily turn into a much longer interview if every ambiguous answer generates another question.
So Smart Follow-ups need a hard question budget.
I explored three modes:
The exact limits are still a product hypothesis. The principle is more important: adaptive questioning needs a hard upper bound.
The goal is to know enough to create a useful Client Brief, not to collect every possible detail.
That also gives Questy a practical stop condition: if another question isn't likely to materially improve our understanding, it shouldn't be asked.
Smart Follow-ups happen during the discovery.
Analysis happens afterwards.
That's an important distinction.
During the discovery, Questy can react to an individual answer: Is there enough information to continue? If not, it can ask a follow-up.
After the discovery, Questy can look across the complete set of answers and identify things that are still unresolved:
If something remains unclear, Questy can turn those gaps into a small clarification questionnaire instead of sending the client through the whole discovery again.
So there are two different mechanisms:
The final output is a Client Brief built from what the research tells us, not another version of the questionnaire.
Questy can bring together the client's priorities, needs, preferences and constraints, while keeping track of unresolved questions and other gaps that matter to the project.
This brings the original workflow together: Discovery → Answers → Analysis → Clarification → Brief
Instead of taking responses from one tool, analysing them somewhere else, and manually turning the result into a brief, the whole process becomes part of the same project.
Most of the difficult design decisions were about how the system should behave, not the questionnaire UI.
This also affected the technical model.
The product should remain the system of record for the Project, Discovery, questions, answers, follow-up budget, analysis and current Brief. AI handles interpretation: understanding the objective, identifying gaps and contradictions, proposing questions and synthesising the research.
That separation lets Questy react to the conversation without turning the whole research process into an AI chat.
More case studies and notes: denisxko.com · full text of everything: /llms-full.txt
© Denis Kharchenko. All rights reserved.