Taking an AI platform from zero to $3M ARR

Situation

A global law firm set out to turn its legal and private equity expertise into an AI product sold to Fortune 500 companies and law firm practice groups. The work had to meet enterprise security requirements — SOC 2 Type II and ISO — and produce answers lawyers could trust.

What I built

• Product strategy and commercialization for the platform, including enterprise pricing that accounted for unstructured-data messiness, inference token spend and volume gating.
• An architecture built for trust: a knowledge-graph representation that connects entities, predicates and states into deterministic citation paths, so every answer traces back to source.
• Evaluation as a first-class product surface: automated suites with human-curated and synthetic ground truth, LLM-as-a-judge scoring and deterministic validation gates tracking accuracy, latency and token consumption.
• A secure multi-tenant platform with isolated namespaces, role-based access control and encryption, plus an MCP and REST layer so forward-deployed engineers could reuse components across clients.
• The organization: a 50-person team across core platform, research, validation and benchmarking, and forward-deployed engineering.

Result

• $0 → $3M ARR in year one, with line of sight to $12M.
• Limited Partnership Agreement and side-letter reviews cut from 15 hours to 30 minutes.
• Court filing reviews accelerated from weeks or months to days.

Takeaway

The AI was rarely the hard part. What made it sell was evaluation buyers could see, pricing that held up as usage grew, and an architecture that let the team ship the same capability to many clients.