Project Brainstorm is designed to help you explore a decision over multiple studies, not through a single study. Synthesis is where that accumulated learning comes together.
Where IDIs and Focus Groups help you understand how people think, a Synthetic Survey (Synthesis) helps you understand which patterns hold across perspectives.
A simple way to think about it:
IDI/FG => surface signals
Synthesis => test which signals hold
Decision => act on those signals
This guide explains when to run a Synthesis study, how to set one up well, and how to avoid the most common mistakes. Part 2 will cover how to read the results.
Running example (for this guide)
Throughout this guide, we use a simple example – exploring whether rabbits should be positioned as pets for children – to show how Synthesis builds on earlier IDI and Focus Group work.
What a Synthesis Study Requires
You can only run a Synthesis study after building a lineage.
That means:
- running one or more IDIs and/or Focus Groups
- tied to the same Decision To Support
For example, in one lineage, multiple IDIs and Focus Groups explored how children perceive rabbits as pets. Synthesis was then used to summarize the patterns across those earlier studies.
Important
Synthesis does not stand alone.
It is built on:
- the reasoning surfaced in earlier qualitative work
- the tensions, signals, and contradictions observed across that work
What Gets Locked (and Why It Matters)
When you create a Synthesis study, the following are inherited from the prior studies in the lineage and locked:
- Business Problem
- Decision To Support
- Market / Geography
- Product / Concept
- Target Audience
You cannot edit them at this stage.
This is intentional. It ensures:
- your Synthesis reflects the same decision context
- results are comparable across the lineage
- you are not shifting the question midway
What You’re Actually Creating in Synthesis
A Synthesis study is not a survey in the traditional sense.
It is a structured way to test: Which qualitative signals hold up when applied across multiple simulated persona contexts.
To do this, you define:
- a set of themes to summarize
- and a set of questions per theme
A useful distinction:
- Themes define what you are testing
- Questions define how you test that signal
For example:
- Theme: Cuteness versus ownership intent
- Question: “Rabbits are cute, but I would not want to keep one as a pet.” (Likert)
Step 1: Define Themes (What You Want to Test)
You can define 1 - 12 themes in a Synthesis study.
Each theme should represent a meaningful signal from your earlier studies.
For example, from the rabbit lineage, themes might include:
- Cuteness vs actual ownership intent
- Role of parents in decision-making
- Perceived burden of care
- Impact of social media on interest vs action
These themes should not be invented from scratch. They should come from:
- Executive Summaries
- Themes and Motivations
- Contradictions and tensions
For example:
- participants consistently reacted positively to “cuteness”
- but quickly shift to practical constraints like space and responsibility
That gap between attraction and feasibility becomes a strong candidate theme.
A simple rule
If a theme wouldn’t change how your decision, it shouldn’t be included.
Step 2: Choose What Signal You Want per Theme
For each theme, decide: What are you trying to understand?
Most questions fall into one of three categories:
1. Prevalence / Intensity
- How strongly does this pattern show up?
- How consistently does it appear across contexts?
Example:
How strongly “cuteness ≠ ownership intent” hold?
2. Comparisons / Trade-offs
- How do people choose between options?
Example:
Emotional appeal vs practical feasibility
3. Reasons / Motivations
- Why does a response occur?
Example:
Why do parents or household constraints dominate decisions?
Step 3: Choose the Question Format
Project Brainstorm supports a range of question types, each suited for a different signal:
- Likert (1 - 5 distribution)
- Multiple Choice
- A/B comparison
- A/B with images
- Ranges / scales
- Open-ended (short free-text, summarized conservatively)
How to think about formats
Match the format to the signal:
- Likert => strength / agreement
- Multiple Choice => segmentation of views
- A/B => trade-offs
- Range => thresholds or limits
- Open-ended => short reasoning, but kept short and summarized
Step 4: Write Questions That Reflect the Learning
This is where most mistakes happen.
What not to do
- Don’t write generic survey questions
- Don’t ignore prior qualitative findings
- Don’t ask questions disconnected from themes
A good rule of thumb: If a question could be asked without reading your earlier studies, it is probably a weak Synthesis question.
What to do instead
Every question should:
- map directly to a theme
- be grounded in earlier study patterns
- reflect a real tension or trade-off
Example
Instead of: “Would you consider owning a rabbit?”
A stronger question might be: “How likely is it that you would you consider owning a rabbit if space and parental approval were required?”
Why this works
Because the earlier studies show that decisions quickly shift to those constraints.
You are not asking a generic opinion. You are testing something the lineage already suggests matters.
Step 5: Keep It Focused (Max 12 Questions)
You can configure up to 12 questions in a Synthesis study.
Fewer is usually better.
Why
Synthesis is about signal clarity, not coverage.
Too many questions:
- dilute interpretation
- add noise
- reduce the usefulness of the results
Weak questions degrade output quality more than missing questions.
Practical guideline
5 - 8 strong questions is usually ideal:
- each tied to a distinct theme
- each adding something decision-relevant
Step 6: Review Before Running
Before you execute the study, sanity-check it.
Quick pre-run checklist
- Does every question map to a real theme from earlier studies?
- Does every theme reflect a real tension, not just a generic topic?
- Are you testing signals, not asking generic opinions?
- Would the answers actually change how you approach the Decision To Support?
If any of these fail, revise before running.
Also note:
- Precheck will flag if required elements are incomplete
- Synthesis cannot run until all themes are fully configured
What Synthesis Is (and Isn’t)
Before running, it’s important to calibrate expectations.
A Synthesis study:
- summarizes patterns across earlier qualitative work
- shows directional distributions across simulated contexts
- helps you compare signals at the lineage level
It does not:
- measure real-world prevalence
- replace validation research
- produce statistically representative results
A Final Reminder Before You Run
Synthesis cannot create new insight from weak inputs.
It reflects:
- the quality of your earlier IDIs and Focus Groups
- the clarity of your themes
- the precision of your questions
If those are weak, Synthesis will be weak.
If those are sharp, Synthesis becomes extremely powerful.
Coming Next (Part 2)
In Part 2, we’ll cover:
- how to read Synthesis outputs
- how to interpret distributions, confidence bands, and decision signals
- how to connect Synthesis back to real decisions
------
Good vs Bad Question Examples (Synthesis)
Synthesis questions work best when they test signals already surfaced in the lineage and help reduce uncertainty for the Decision To Support.
Avoid writing “generic surveys”.
You are not measuring real-world prevalence.
You are testing whether patterns hold across simulated contexts.
Reminder of the example
The following examples continue the same lineage used throughout this guide: exploring how children (and their households) perceive rabbits as pets, especially the gap between initial attraction and practical ownership constraints.
Each “good” question below ties directly to signals surfaced in earlier IDIs and Focus Groups.
1. Theme: “Cuteness vs willingness to keep” (Likert)
Bad: “Do you like rabbits?”
Why it fails: Too broad, not decision-relevant, and it doesn’t test the core tension.
Good: “Rabbits are cute, but I would not want to keep one as a pet.” (1-5 Likert)
Why it works: It directly tests the gap between attraction and commitment, which is a recurring pattern in the lineage.
2. Theme: “Practical barriers dominate feasibility” (Multiple Choice)
Bad: “What is the biggest reason people don’t buy rabbits?”
Why it fails: It implies real-world prevalence and invites overconfident generalization.
Good: “Which constraint most limits feasibility for you?”
Options:
- Space at home
- Time to care
- Cleaning and hygiene
- Parent approval
- Cost
- Not interested in pets
Why it works: It forces structured trade-offs aligned with the constraints surfaced in prior qualitative work.
3. Theme: “Parent approval is a gating factor” (Range/Scale)
Bad: “Would your parents approve?” (Yes/No)
Why it fails: Too binary, loses nuance.
Good: “How required is parent approval for a pet decision in your household?” (0-10 scale)
Why it works: It captures intensity and household gating without pretending to be a population statistic.
4. Theme: “Promotion can increase attention but not conversion” (A/B)
Bad: “Which marketing idea will work best?”
Why it fails:
It is outcome-predictive framing and not appropriate for discovery-stage synthesis.
Good: “Which would make you more likely to engage?”
A: “Mall pop-up and photo spot”
B: “Care education and feasibility guidance”
Why it works: It tests a decision-relevant trade-off that emerged in the earlier qualitative work.
5. Theme: “Feasibility thresholds” (Range)
Bad: “How much would you pay to keep a rabbit?”
Why it fails: Pricing is later-stage and invites false precision.
Good: “What is the maximum daily time you could realistically spend on pet care?” (Range buckets)
Options:
- 0 - 5 min
- 5 - 15 min
- 15 - 30 min
- 30 - 60 min
- 60+ min
Why it works: It tests feasibility constraints directly, which are central to the lineage’s reasoning.
6. Theme: “Responsibility framing vs toy framing” (Likert)
Bad: “Is the ‘fun pet’ idea good?”
Why it fails: Too vague, too subjective, and not diagnostic value.
Good: “Calling a rabbit a ‘fun pet’ makes the responsibility feel understated.” (1-5 Likert)
Why it works: It targets the specific tension the earlier studies raised about commercialization and responsibility.
7. Theme: “When open-ended is appropriate” (Open-ended, short)
Bad: “Explain everything you think about rabbits as pets.”
Why it fails: Too broad and likely to produce low-signal summaries.
Good: “In one sentence, what is the biggest thing that makes rabbit keeping feel unrealistic?”
Why it works: It is short, bounded, and directly tied to a known barrier theme.
Quick guardrails for strong Synthesis questions
- Bad questions ask for outcomes. Good questions test signals.
- Bad questions are generic. Good questions mirror a specific tension surfaced in IDIs/FGs.
- Bad questions imply real-world prevalence. Good questions describe simulated distributions across contexts.
- Keep the total count low. The max is 12, but fewer is often better if questions are not pulling their weight.
------
We’ll continue publishing guides and FAQs on interpretation, validation, and how to get the most value out of each study type.
Remember: Project Brainstorm is an experimental beta. It is designed to help you see the market more clearly, not to replace judgement or downstream validation.
If you’d like to learn more about Project Brainstorm or share feedback on the Beta, you can reach us at contact@projectbrainstorm.xyz