Research Methodology
    89% internal match rate

    How IdeaProof validates business ideas

    Frontier AI models from Google (Gemini), Anthropic (Claude), OpenAI and Perplexity, selected for each research task — grounded in our proprietary RAG on past validations.

    7-step pipeline · Frontier AI · Cited sources

    Try our methodologySee entity profile

    Maintained by IdeaProof Research Team

    Scientific beaker with floating research documents and methodology checkmarks
    89%
    Internal match rate
    Internal metric, not certified
    60+
    Live Sources + RAG
    Refreshed weekly
    50+
    Validation criteria
    Per report
    4+
    Frontier AI Models
    Gemini · Claude · OpenAI · Perplexity

    How we calculate the 89%

    The 89% is an internal match rate: how often our verdict matched the outcome founders later reported, on a sample of past validations. It is not a certified prediction accuracy.

    Sample
    Past validations with a founder-reported outcome — size being published
    Outcome
    Founder-reported: launched and operating vs. stopped
    Comparison group
    Being published
    Period and confusion matrix
    Being published

    What we score (50+ criteria)

    Weighted distribution across our validation rubric

    50+
    Criteria
    • Market demand & TAM28%
    • Competitive moat22%
    • Unit economics18%
    • GTM feasibility16%
    • Risk & regulation10%
    • Team / execution fit6%

    Frontier AI

    Frontier AI models from Google (Gemini), Anthropic (Claude), OpenAI and Perplexity, selected for each research task

    Autonomous agents

    Each validation spawns specialist agents (market, competitor, financial, risk) that debate before the final verdict.

    Live web search

    Perplexity and OpenAI web search support fresh, cited evidence for relevant research tasks.

    Proprietary RAG

    Grounded in 27k+ validations, 1000+ failure cases and curated sector benchmarks.

    Frontier AI, selected for each task

    We use frontier AI models from Anthropic (Claude), OpenAI, Perplexity and Google (Gemini). Providers are selected by task; not every provider is used in every report. Updated 2026-10-08.

    In production

    • Gemini (Google)Standard validation, research, synthesis and classification
    • Claude (Anthropic)Pro validation synthesis and business plans
    • OpenAIPro validation deep research with web search
    • PerplexityCited live-web lookups for market and news enrichment

    Fallback only

    • Gemini (Google)Alternative validation engines when needed
    • OpenAIValidation fallback and Pro enrichment
    • Grok (xAI)Pro enrichment fallback only
    Our Process

    How We Validate Ideas

    A transparent, seven-step pipeline combining frontier AI models, autonomous agents and a proprietary RAG layer

    STEP 1

    Real-Time Data Aggregation

    We collect data from 60+ authoritative sources including market research databases, industry reports, government statistics, funding trackers and real-time web crawling.

    Market research databases (IBISWorld, Statista, Gartner, CB Insights)
    Funding & financial data (Crunchbase, PitchBook, SEC EDGAR, Bloomberg)
    Patent and regulatory filings (USPTO, WIPO, EUIPO)
    Social & community signals (Reddit, X/Twitter, LinkedIn, Product Hunt)
    Competitor & SEO data (SEMrush, Ahrefs, SimilarWeb, BuiltWith)
    STEP 2

    Frontier AI, selected by task

    We select frontier AI models for research, synthesis and source discovery. Providers vary by task and report tier; not every provider runs on every validation.

    Claude (Anthropic): long-context reasoning and final report synthesis
    OpenAI: deep research with native web search on Pro validations
    Gemini (Google): standard validation research and synthesis
    Perplexity: live web search with cited sources for fresh evidence
    Custom scoring layer: weighted ensemble with confidence intervals and disagreement flags
    STEP 3

    Autonomous Agents + Live Web Search

    When you launch a validation, IdeaProof spawns a swarm of autonomous agents that run in parallel, each browsing the live web with its own specialty and model, then debate the conclusions before the final verdict is composed.

    Market Agent: estimates TAM/SAM/SOM and growth using Perplexity + OpenAI web search
    Competitor Agent: maps direct, indirect and substitute competitors in real time
    Financial Agent: stress-tests unit economics, CAC/LTV and burn against benchmarks
    Risk Agent: scans regulation, IP, platform and macro risks for the specific industry
    Brand & GTM Agent: probes positioning, channels and messaging opportunities
    Consensus step: agents cross-review each other's claims; conflicts are flagged, not hidden
    STEP 4

    Proprietary RAG Knowledge Base

    On top of public sources, every agent is grounded in our proprietary RAG layer: a vector-indexed knowledge base built from years of IdeaProof validations and curated startup intelligence.

    Past IdeaProof validations with verdict patterns (anonymised)
    1000+ documented startup failures with root-cause analysis
    1,700+ curated startup ideas dataset across 30+ industries
    Internal glossary, frameworks (JTBD, Mom Test, Lean Canvas) and sector benchmarks
    Few-shot grounding: models compare your idea against the closest historical analogues
    STEP 5

    50+ Validation Criteria

    Every business idea is analyzed across 50+ specific criteria tailored to the industry, business model, and target market.

    Market criteria: TAM/SAM/SOM, growth rate, saturation level
    Competitive criteria: number of competitors, market share, differentiation
    Financial criteria: unit economics, CAC, LTV, burn rate feasibility
    Execution criteria: team capabilities, resource requirements, timeline
    Risk criteria: regulatory barriers, technology risks, market timing
    STEP 6

    Historical Validation & Outcome Feedback

    We continuously validate our predictions against real-world outcomes and feed them back into the RAG layer and scoring weights.

    Outcome follow-ups: founders report what happened after validation
    Weekly retraining of ensemble weights based on new outcome data
    Internal match rate (89%) published with its definition and limits
    STEP 7

    Quality Assurance & Bias Prevention

    Multiple safeguards ensure objective, unbiased analysis focused solely on insights.

    No affiliate relationships with tools or services we recommend
    Blind analysis: AI doesn't know user identity or demographics
    Diverse training data: ideas from 40+ countries, 20+ industries
    Human oversight: monthly audits of AI recommendations
    Transparency: the 89% internal match rate is published with its definition and limits
    Data Sources

    60+ Authoritative Sources + Proprietary RAG

    We aggregate live data from the most trusted market intelligence providers and combine it with our proprietary knowledge base

    Market Intelligence

    Statista logoStatista
    IBISWorld logoIBISWorld
    Gartner logoGartner
    Forrester logoForrester
    McKinsey logoMcKinsey
    CB Insights logoCB Insights
    Exploding Topics logoExploding Topics

    Financial & Funding

    Crunchbase logoCrunchbase
    PitchBook logoPitchBook
    SEC EDGAR logoSEC EDGAR
    Bloomberg logoBloomberg
    Yahoo Finance logoYahoo Finance
    Y Combinator logoY Combinator

    Competitive Intelligence

    SEMrush logoSEMrush
    Ahrefs logoAhrefs
    SimilarWeb logoSimilarWeb
    BuiltWith logoBuiltWith
    Wappalyzer logoWappalyzer
    G2 logoG2
    Owler logoOwler

    Consumer & Social Signals

    Reddit logoReddit
    X / Twitter logoX / Twitter
    LinkedIn logoLinkedIn
    Product Hunt logoProduct Hunt
    Google Trends logoGoogle Trends
    Trustpilot logoTrustpilot
    Glassdoor logoGlassdoor

    Industry & Regulatory

    World Bank logoWorld Bank
    OECD logoOECD
    Eurostat logoEurostat
    ISTAT logoISTAT
    WIPO logoWIPO
    USPTO logoUSPTO

    Proprietary RAG (IdeaProof KB)

    IdeaProof Validations (27k+) logoIdeaProof Validations (27k+)
    Failure Library (1000+) logoFailure Library (1000+)
    Ideas Dataset (1,700+) logoIdeas Dataset (1,700+)
    Sector Benchmarks logoSector Benchmarks

    Transparency & Accountability

    We publish how our internal match rate is defined, update our RAG and ensemble weights weekly based on real outcomes, and maintain no affiliate relationships with recommended tools. Our goal is 100% objective, validation you can trust.

    Quarterly Reports

    Internal metric definition and limits, published openly

    Weekly Updates

    RAG layer and ensemble weights retrained with new outcomes

    No Conflicts

    Zero affiliate relationships or paid placements

    Experience Our Methodology

    See our research methodology in action. Validate your business idea with AI-powered analysis in 120 seconds.