Before:
weeks of validation, expensive panels, confidentiality blockers.
After:
overnight testing, 92 percent accuracy, zero confidentiality risk
Our pilot unlocked future-oriented cooperation to further build strategic value based on long-term adoption.
Deep dive
Overview
Company
Mattoni is the leading non-alcoholic beverage producer in Central Europe. Their innovation team develops dozens of product concepts each year, but lacks the capacity and budget for traditional market research. Although they use AI for trend scraping, they had no reliable way to validate early-stage concepts with “consumers”, especially under heavy secrecy requirements.
Time
Q3 2025
Timeline
Week 0: NDA, internal data alignment, kickoff
Week 1: Brief, identification of data sources
Week 2: Model setup + dedicated instance creation
Week 3–4: Training & first 10 concept tests
Week 5–8: Fieldwork (testing, iteration, weekly check-ins) Week 8–12: Evaluation, validation, synthesis, recommendations
Main goals
Validate the feasibility of AI-based concept testing
Compare Lakmoos AI outputs to human respondents (benchmarking)
Test speed, stability, and confidentiality for innovation workflows
Identify scalability for Mattoni’s broader product pipeline
People involved
Stakeholders: Pavel, Lukáš, Nikola
Department: Digital, Innovation & Market Research
Key metrics
Total questions asked: 983
Total answers: 235920
Concepts tested: 25
Datapoints infused: Past surveys, trend tracking, innovation concept benchmarks
Who participated
Heroes of the project
What was blocking mattoni before?
Challenges and opportunities
What was blocking Mattoni before?
Validation Bottleneck: Concept ideation outpaces validation; traditional research is too expensive for iterative testing.
Quality Issues: Online research panels are shrinking, leading to inconsistent data quality and untrustworthy results.
Missing "Customer Voice": Existing AI tools scrape general trends but fail to validate specific concepts in early-stage innovation.
Data Security: Confidential product pipelines limit what can be tested publicly without exposing sensitive IP.
Where Lakmoos fits
A private, secure, on-demand research engine that simulates target consumers and validates concepts without exposing sensitive data to the public.
We save money, time and effort
What we delivered
Lukáš sokolák
200k Answers & Precision Use Case Identification
Concept validation pipeline for automated overnight testing.
Agile delivery: 5 surveys completed during one "Mattoni Eagles" sprint.
Proven 92% accuracy compared to Mattoni’s human benchmark data.
Higher precision and stability than legacy market research AI platforms.
Strategic partnership: Weekly training and methodological support.
Custom model for Czech beverage market and consumer categories.
Consumer insights on sustainability, sugar, and functional drinks.
Concept decisions that previously took weeks were made overnight
Innovation team gained independence from external fieldwork
Research bottlenecks disappeared for early-stage concepts
Leadership confidence increased due to benchmarking success
Lakmoos is now positioned as the fastest validation layer before major investment decisions
Timeline
How the pilot went
How we checked accuracy
Validation
Benchmarking process
Verified that the AI panel contains relevant category data
Mattoni ran the survey with a human representative sample
Lakmoos AI answered the same survey with matching socio-demographics
Compared distributions, mean differences (WMAE), and question order
For open-ended answers, measured topic alignment using silhouettes
Result
92% similarity score between Lakmoos AI and human respondents
The pilot demonstrated that AI panels can support fast and confidential concept validation.
Competitor comparison
Lakmoos outperformed other AI panel providers in accuracy
Lakmoos does not hallucinate
Worries we overcame
Is this safe?
Can we trust the results?
Will leadership accept AI-based insight?
Yes, after benchmarking, the innovation board endorsed continuing with Lakmoos.
Can it handle secret products?
Limitations
?
Brand-level precision
Core Lakmoos datasets were not sufficient for brand-specific questions (e.g., brand fit). Additional training on internal Mattoni datasets would be required, representing further time and cost.
?
Category nuance
Beverage sensory cues and positioning attributes would benefit from dedicated category embeddings.
?
Scalability
International scaling appears feasible, but has not yet been verified on local preferences or cultural associations.
?
Data prep
Some internal data preparation effort is needed for higher model fidelity.
AI validation as part of innovation workflow: Mattoni can test every concept immediately after ideation, shifting from selective validation to continuous validation.
Augmentation of trend scraping with consumer simulation: Desk research identifies trends; Lakmoos validates whether the trend resonates with Mattoni’s consumers.
Replacing unreliable panels: Synthetic respondents offer consistent, scalable validation without declining panel availability.
Automated concept scoring: Future integration: automatic scoring and prioritization of all submitted concepts for weekly innovation sprints.
