Boomerang Ideas develops modern market research methods together with researchers — through AI (simulation, digital twins) and through real people (social sampling). Our science-first approach delivers better results on both sides than conventional methods — and it is primary research: fresh answers from people we ask, never crawled from the web.
Best of Swiss Web 2026Easy, fast and affordable high-quality research — for marketing teams testing campaigns, product teams shaping features, executives validating a strategy, and analysts forecasting votes. The method changes with the question; the standard doesn't.
Respondents recruited ad hoc where they already are — Instagram, Facebook, TikTok, Snapchat, LinkedIn, WhatsApp — micro-targeted for the specific question, reaching groups no panel holds. Cross-quotas over age, gender and region are steered live by the engine; incentives are benchmarked to fair per-minute pay and released after quality checks. Attention checks, speeder and duplicate detection run in every field. Mobile-first questionnaires, 21 countries, fresh for every study.
Persona-conditioned large language models (LLMs), fine-tuned on decades of our own anonymized production responses — German, French and English — answer the identical questionnaire in minutes, one model call per synthetic respondent. Validated on held-out Swiss surveys: overlap (1 − mean TVD) of 82.6% in benchmark v1, with strict integrity rules — contaminated surveys dropped, seen questions excluded. Every run is scored by Confidence Guard™ (0–10), declared as simulation, labelled in every export, never mixed into a human sample.
University of Zurich tests socials-recruited samples against established methods — attention checks, retention, social-desirability bias.
Second round against an established online panel: 81% vs. 37% passing attention checks, lower desirability bias.
Same questionnaire, our field vs. a commercial panel — coverage and quality compared directly.
Our channel becomes the field layer for gfs.bern's vote surveys — their forecast, our field.
Head-to-head against a client's own fieldwork, on their instrument.
0.25% off the presidential result, against an industry average of roughly 3.6% — 100% human sampling; the accuracy came from coverage. One race: a case, not an error rate — internal analysis of our own survey, not a published validation.
Simulated answer distributions land on average 82.6% of the way to the human ones, across five Swiss surveys held out of training.
Confidence-driven augmentation of synthetic survey data, with the University of Basel and Johns Hopkins University.
Method documentation: Institutsprofil & Methodik (PDF). Reports: Swiss Insights 2021 · UZH Pre-Study 2022 · UZH Validation Study 2023/24
Our first benchmark measures the distance between the distribution of answers real people gave and the distribution our production model produces.
82.6%
Mean overlap with the human answer distributions
5
Real Swiss surveys held out of training
<52%
Score of untuned frontier models on the same test
Fig. 1 — Benchmark v1, Aug 2026. Overlap = 1 − mean total variation distance. Topline only, Swiss surveys, production model. Measures faithfulness to the human survey, not correctness about the world — tracked separately. Cohort-level scoring follows. Full method and all models in the technical write-up.
A simulation that always answers is worth little. The useful part is knowing when it shouldn't be trusted — so every result carries a plausibility score, and the platform says when to go and ask real people instead. The score (0–10) is assigned by an independent judge model that rates each simulated result for consistency with the question, target group and answer logic — it measures trust in the simulation, not correctness about the world; validation of the criterion against human re-fieldings runs with the University of Basel and Johns Hopkins. Automatic escalation per cohort is in development with our research partners.
Fig. 2 — Average plausibility across all simulations per month, Dec 25 – Jul 26. Everyday use, not lab conditions: arbitrary topics, started by users themselves. The May dip is real and stays in the chart — every human sample feeds back into the model, which is why the curve climbs. 5.6 → 7.1.
Founded in Zürich. Seed-backed since 2020, Innosuisse research funding, and gfs.bern as strategic investor since 2023. A small team that runs its own field, its own models and its own benchmark — and publishes the misses along with the hits.
Partners & memberships
Myrto ZehnderHead Incentives & Ext. Surveys
David FurrerHead of AI
Jakub MotyčkaFull Stack Developer / CTO
Mark KorondiDev & AI
Tina Olivia SeilerPublic Relations
Hadrien Jean-RichardDigital Marketing
Fritz SeidelCo-Founder
Raphael UeberwasserFounder & CEO
Lukas GolderBoard Member (gfs.bern)
Nina SchachtBoard Member · qualitative market researcher
Jean-Marc HenschBoard MemberRaphael UeberwasserChairman of the BoardDr. Andrea Bublitz (Academic Advisor · Universität Basel) · Dr. Linda Stougaard Nielsen (AI / ML Advisory) · Dr. Kristina Gligorić (Ethical AI Advisor · ex-Stanford, now Johns Hopkins) · Katia Murmann (Product Advisory) · Beat Seeliger (Technology Advisory · Panter) · Martin Steiger (Data Privacy Advisory · Steiger Legal) · Peter Erni (Social Media Advisory) · Trevor W. Goodchild (Facebook Strategist) · Martin Brettenthaler (Sales Strategy) · Jean-Marc Hensch (Business Angel)
For Corporates & Institutes
Fully fledged research for experienced researchers who need it all — research & insight teams, institutes and agencies: social sampling, simulation, trackers and full methodological control.
For SME & Startups
For business professionals who need simple decision support along the way — small budgets, no research background needed: our agent takes you from question to insight.
Research collaborations, methods questions, media: hello@boomerangideas.com. Projects and quotes run through the two doors above.
⠿ verschieben
Als anonym