Giving a law firm a scored AI roadmap and a governance starting point
A law firm's AI adoption had been happening ad hoc, and attorneys carry real ethical duties around competence, confidentiality, and supervision that ad hoc doesn't satisfy. Replacing it meant scoring fifteen use cases against risk and the state bar's own ethics opinion, and being willing to hold some of them back, including one that looked ready until a transcription test came back unreliable.
Category
AI governance and strategy.
What I was trying to accomplish
Replace ad hoc AI adoption with a deliberate, risk-aware plan, given attorneys' ethical duties around competence, confidentiality, and supervision.
The data challenges
Comparing platforms against the firm's own IT provider's input, mapping a state bar ethics opinion point by point to professional conduct rules, and setting aside an earlier AI-generated roadmap entirely so the new one could be judged on its own evidence.
The result
Fifteen use cases scored on payoff, risk, and effort, tied to the ethics opinion and court rules; a committee response covering licensing, cost control, training, and a warning to vet third-party AI skill files for hidden instructions; and one set of AI instructions built to work consistently across two different AI platforms. Higher-risk uses were deliberately held for later phases, and an unreliable transcription result was held for retesting with mandatory human verification rather than shipped.
Technical details
Tools: Claude (claude.ai with Microsoft 365 connector), Microsoft Copilot, Excel
Interest in AI was growing at the firm, free consumer tools with unclear data retention were a known risk across the industry, and attorneys carry specific ethical duties around competence, confidentiality, and supervision. The firm needed a deliberate plan instead of ad hoc adoption. As the firm's Claude account owner, I led that work.
The inputs were the firm's AI research folder, a condensed summary of a published managing partner's guide to AI adoption, and the state bar's AI ethics advisory opinion, which Claude retrieved from the source and mapped point by point to the professional conduct rules. From those, I rebuilt the roadmap from scratch as a scored backlog: fifteen use cases grouped into administrative, attorney, and client-facing work, each rated on payoff, risk, and effort. When an earlier roadmap produced by another AI tool turned up in the same folder, I had it excluded entirely so the new backlog could be judged on its own.
The challenges were about judgment more than technology. Higher-risk attorney uses were held for later phases. An internal test of AI transcription on a long hearing recording showed speaker identification was unreliable, so that use case was held for retesting with mandatory attorney verification. Scope changes forced full rebuilds of the backlog rather than patches.
Pull up a chair.
Let's talk through where things stand, and where they could go.