Methodology
How ranking works
Rankings should feel like a Google ranking, not a Clutch directory. Here's what's under the hood.
The atomic unit: a claim
Every firm profile is a set of claims. Examples:
- • "Shipped 3 production RAG systems in regulated fintech in 2025."
- • "Team of 12, including 4 ML engineers with ≥5 years experience."
- • "SOC 2 Type II certified, audit closed 2025-11."
Each claim carries a score, a confidence, evidence links, and a freshness timestamp.
How scores are computed
- Source weight. Primary sources (firm site, GitHub, LinkedIn, press) score higher than secondary and tertiary.
- Corroboration count. More independent sources agreeing = higher score.
- Specificity. "3 production RAG systems at named fintech clients" beats "experienced in AI."
- Recency. Evidence scored within 90 days beats older.
- Contradicting evidence. Hard penalty; triggers manual review.
- Firm reputation. Prior claim-accuracy rate feeds back into source weight for that firm.
What doesn't move rankings
- Payment, outside the disclosed Featured slot cap.
- Review volume alone.
- Firm age.
- Any Firmsy employee's opinion.
Re-scoring cadence
- Weekly sweep across all active firms' public surfaces.
- Immediate re-score on profile edit.
- Immediate re-score on disputed flag.
Why you can trust the ranking
Every shortlist and every ranking is explainable. Ask "why is this firm #3?" and Firmsy returns the evidence-weighted reasoning.