Pricing
"If we move the subscription from CHF 39 to CHF 45, who leaves?"
Simulated · 4 price points · plausibility 8.1
AI market research · simulated respondents
Ask 500 customers. Before lunch.
Simulated respondents answer your market question in minutes — Confidence Guard™ tells you when to ask real people.
The agent scopes it in a few questions, then runs the study. No account needed to try.
No panel contract. No minimum order. No sales call. Results arrive live as they come in.
Yes, I'd pay it0%
Only with a loyalty discount0%
No, too expensive0%
Plausibility 8.0/10
Confidence Guard™: High plausibility. What that means →
Awarded 2026
Best of Swiss Web — Marketing (Silver), Productivity (Bronze), Innovation (Bronze)
Best of Swiss Web — Marketing (Silver), Productivity (Bronze), Innovation (Bronze)How it works
Three moves. The third is the one nobody else offers.
FIRST
Ask like a human, not a researcher
Type the question the way you'd ask a customer. We turn it into a proper survey instrument — answer options, order effects, the boring craft that decides whether a number means anything.
THEN
Simulate on survey data, not web text
Respondents are generated from real survey answers we collected ourselves — not from a model's idea of what your customers sound like. Stratified by age, gender and region, weighted against official population statistics. The same ground truth as a human sample. Results arrive live over 10 to 30 minutes.
WHEN IT MATTERS
Anchor it with real people
One click sends the identical questionnaire to real respondents, recruited on social media with our patent-pending stratified sampling. The simulation becomes the hypothesis; the humans settle it. Usually back within a few days.
Why it holds up
Any model will answer you. Ours will also tell you when to stop trusting it.
Asking ChatGPT what your customers think is free and instant. It is also unfalsifiable: nothing sits behind the number, and no one can tell you the answer was wrong until the market does. That's the whole difference here. It starts where a chatbot stops — and closes the gap to professional market research.
Asking a general chatbot
- ×Trained on the internet, weighted toward the loudest markets — rarely the one you sell in
- ×Same confident tone whether it knows or is guessing
- ×No sub-groups you can defend to a board
- ×No route to a real answer when the stakes rise
Asking boomerang.ai
- ✓Calibrated on human samples we run ourselves, continuously, and measured against them
- ✓Every result carries a confidence score and the reason behind it
- ✓Region, age and gender splits, with base sizes shown
- ✓Optional: escalate to real respondents in one click — same questionnaire, no re-setup
What people actually ask
The decisions that were too small for a research budget.
Most companies don't skip research because they don't care. They skip it because a CHF 25,000 study can't be justified for a CHF 25,000 decision.
Six real questions and what happened to each — all of them on boomerang.ai: simulated where the plausibility held, anchored with real respondents where it didn't.
Naming
"Which of these three names sounds trustworthy to a Swiss German ear?"
Simulated · then anchored, n = 400
Positioning
"Does 'sustainably sourced' still move anyone, or is it noise now?"
Simulated · split by age · plausibility 7.4
Creative
"Which of these two ad concepts gets remembered tomorrow?"
Simulated first · then anchored with real people — recall needs humans
Product
"Would customers actually use a self-checkout here, or say they would?"
Plausibility 5.2 · anchored with real people
Expansion
"Does this work the same in Romandie as in Zürich?"
Simulated · regional split · plausibility 6.9
Confidence Guard™
A number you can't check isn't research. It's a horoscope.
Every answer arrives with a plausibility score from 0 to 10, assigned by an independent checking model that rates the result for consistency with the question, the audience and the answer logic — it measures how far to trust the simulation, not whether the world agrees. Each score comes with a plain-language verdict on what it means for your decision. It's the part we'd want if we were buying this.
8.0 – 10
Decide on it. Dense human calibration behind this question type. Simulation has tracked reality closely here.
6.0 – 7.9
Direction, not decimals. Trust the ranking, don't quote the percentage in a board deck. Anchor before you commit budget.
below 6.0
Go to humans. We'll say so plainly, and put the escalation button right under the result. Selling you a weak number would cost us more than the credit.
Pricing sensitivity, retail CH8.1
Brand associations, insurance7.6
Ad recall, 48h unaided4.8
plausibility score · last human anchor
Recalibrated with every human sample we field.
How good is good?
We publish our own error rate.
Every vendor claims their AI is accurate. It only becomes checkable when someone shows the number — including the months it went down. The plausibility score itself (0–10) comes from an independent judge model that rates how credible each simulated result is for your question and audience — it measures confidence in the simulation, not facts about the world.
6.94
Average plausibility in August 2026, against 5.6 in December 2025.
82.6%
Overlap between our simulated answer distributions and the real human ones — five Swiss surveys held out of training, scored on the model in production.
Why the curve climbs: every human sample feeds back into the model. Tomorrow's accuracy depends on today's fieldwork — which is why we run both.
Pricing
Start free. Pay when a decision is worth it.
Simulations are cheap because they're computed. Human samples cost what fieldwork costs — we don't hide either number.
All prices excl. VAT.
Starter
For finding out whether the answers are any good.
CHF 0
2 studies per month
- up to 300 respondents per study
- Plausibility score on every result
- Overall results (topline)
- 1 study in pre-test at a time
- Help centre
Startups & SMEs
Plus
For teams putting real money behind the answer.
CHF 79 / month
or CHF 790 a year — full allowance up front
- 5 studies per month
- up to 1,000 respondents per study
- Region, age and gender splits
- A/B tests, long surveys, raw data and PDF export
- Up to 3 studies in pre-test at once
- Email support
- Add real people any time: 5–10 credits per respondent (1 credit = CHF 1)
Worked example
Simulated answers come with your plan. Real people are bought in credits — one credit is CHF 1, and one completed interview costs 5–10 of them. A typical study, at the sample size we recommend, with the arithmetic done:
The study
- Audience
- Swiss general population
- Sample
- 300 respondents
- Questions
- 3
- Field time
- 3–7 days
Worked example
CHF 1'500–3'000
5–10 credits per completed interview, for 300 respondents · excl. VAT
A worked example, not a quote. Your price is firm for the stated assumptions: audience, n, length, incidence. If length or incidence move by more than ±25 %, we bill the actual effort — and tell you first. The free pre-test shows both before any cost arises. Narrower audiences and longer questionnaires sit at the top of the range; other markets can cost less. Prices in CHF, 1:1 in EUR and USD.
An institute, agency or insights team? At Boomerang Ideas you self-serve too, with onboarding help and a phone hotline on top of email support.
One last thing
Ask the question you've been arguing about.
First one's free, and the answers start arriving while you read this page.
⠿ verschieben
Als anonym