Problem. A regional law firm was losing junior associates to burnout. First-pass review of discovery documents was eating 40% of associate time and 0% of margin.

Approach. Built a retrieval-augmented document classifier with a human-in-the-loop review queue. Embedded into the existing matter management system. Eval harness built before the agent.

Outcome. First-pass review time down 78%. Associate satisfaction up. Realised margin per matter improved by 14 points within two quarters.

Problem. AP team of seven coding ~2,400 invoices a month. High error rate, slow close, no path to scale without hiring.

Approach. Deployed LLM-augmented agentic workflow that read invoices, pre-coded line items against the chart of accounts, and routed exceptions. Human review for anything below confidence threshold.

Outcome. 85% of invoices auto-coded with 99.2% accuracy. AP team redeployed to vendor management. Month-end close shortened by three days.

Problem. Operating partner inherited eight portfolio companies, each running its own ad-hoc AI experiments. No shared roadmap, no shared infrastructure, no shared learnings.

Approach. Ran a parallel readiness assessment across five priority portcos. Identified three workflows common to all of them. Built one reusable pilot, deployed across the portfolio.

Outcome. Single agent platform now serving four portcos. Combined annual savings projected at $2.1M in year one. Operating partner can show the LPs a real AI story.

Problem. A specialty practice was drowning in prior-auth requests. Staff were copy-pasting between EHR, payer portals, and an internal tracker.

Approach. Built an agent that read incoming requests, pre-filled payer forms from EHR data, and surfaced exceptions for a clinician review. Hard guardrails on anything touching clinical recommendation.

Outcome. Turn-around time on routine prior-auth requests dropped from 4 days to under 8 hours. Staff time freed for patient-facing work. Zero clinical decisions made by the agent.