Services
AI that earns its place in the practice.
AI solutions, in our practice, mean a small number of narrowly-scoped AI deployments that measurably reduce time, increase quality, or open a new capability, plus the governance to run them without embarrassing the firm. The outcome is one or two AI systems that partners, leaders, and clients trust, backed by a written policy that says what is and is not allowed with client or matter data.
How we run it
Four phases, applied to this practice
Phase 1
Discovery & Audit
We start where AI is most likely to earn its keep: drafting, intake triage, call summarization, document review, internal search. We evaluate current volume, current quality, and the honest baseline cost.
Phase 2
Architecture & Design
We design the deployment: model, data boundary, human-in-the-loop pattern, evaluation criteria, and governance policy. For law firms this includes matter-data handling and client disclosure, both written before the first prompt is issued.
Phase 3
Build, Integrate & Automate
We build a controlled pilot with one team, one workflow, and defined success metrics, before any firm-wide rollout. Prompts and evaluation are version-controlled; usage is logged for later review.
Phase 4
Train, Measure & Refine
We train the pilot team, gather structured feedback, and only expand once the numbers hold up. We refresh the policy and evaluations quarterly as the underlying models change beneath us.
Managed AI
Powerful tools for some, managed AI for all.
A durable AI program gives every team member a company-run assistant they'd actually choose, and reserves the sharpest, most specialized tools for the people whose work demands them. That combination is what keeps proprietary and customer data inside the business while the work still moves forward.
Prevent data leakage by giving the team a better default
When the company runs AI and points it at the right internal data, team members reach for the company assistant instead of pasting sensitive material into a personal account. That single change removes the most common path proprietary and customer data takes out of the business.
Device-level control for sensitive verticals
For firms in regulated or otherwise sensitive verticals, non-approved AI models can be rendered inaccessible on company-enrolled devices through device management. The approved tools are frictionless; the unapproved ones don't open. It's a quieter policy than a memo and considerably more effective.
Per-team and per-user context scoping
Context windows are scoped explicitly. A marketing team sees marketing data, a finance team sees finance data, and named individuals get access to sensitive material only when their role requires it. The same platform serves everyone; what it can see changes with who is asking.
BYOK economics that stay business-effective
Bringing your own API keys keeps AI usage cost-efficient at scale and portable across models as the market shifts. Combined with per-team scoping, BYOK lets a business pay only for what it actually uses and swap the model behind the assistant without changing the assistant.
Governance that fits on a page
The managed environment carries the firm's AI ethos with it: what to automate, what stays human, transparency with clients, and the data red lines. One decision-maker, one review cadence, one place team members turn to when a new tool appears.
What this looks like in practice
Recent work and field notes
AI deployments and governance work in the field: which models, how they were bounded, and what the results looked like.
Case Study
Migrating 60 phone numbers to Aircall — and lifting answer rates past 90%
Consolidating roughly sixty numbers onto Aircall gave a firm visibility into missed calls, a unified call/SMS log in HubSpot, AI summaries, and the runway for AI voice agents.
ReadArticle
Pick a company ethos on AI before the tools pick one for you
An AI ethos is a short, plain-language document that tells your team, your clients, and every future tool where you stand.
ReadArticle
Data governance is the precondition for useful AI
Most stalled AI projects aren't AI problems. They're data governance problems wearing a model's costume.
Read
Engagement models
How this work is scoped
Model
Project Delivery
The default fit. A defined pilot for a single high-value workflow, plus the governance to run it safely.
Model
Audit & Roadmap
For firms with several AI pilots already running and no consistent policy. We write one.
Model
Fractional Leadership
Best for firms treating AI as a strategic capability across multiple practice groups over time.
FAQ
Questions we answer often
Do you build custom models, or use OpenAI / Anthropic / Google?
We use the leading frontier and hosted models by default (OpenAI, Anthropic, Google, and specialized legal-tech models). Custom fine-tuning is rare and only proposed when a benchmark shows it is worth the ongoing cost.
How do you handle client-confidential or privileged data?
Every deployment starts with a written data-handling boundary: which data can enter which model, under which contract, with which retention. Client-privileged data typically goes only to zero-retention endpoints under a signed DPA, and disclosure to clients is handled explicitly.
What is an AI governance policy, and do we need one?
Yes. Even a two-page policy answers the questions clients, insurers, and courts are starting to ask. It covers approved tools, prohibited data, human review requirements, and who owns exceptions.
How do you measure whether an AI pilot is working?
Before rollout we agree on 2–3 numbers, usually time per matter, error rate, and reviewer acceptance. We measure a control group in parallel where feasible, and stop the pilot if the numbers don't hold up.
Where does AI help a law firm today, and where doesn't it?
It reliably shortens drafting, intake triage, discovery review, and internal search. It is not a replacement for legal judgment, cite-checking without a verifier, or client communication where privilege matters, and we say so plainly.
Considering this work for your firm?
Drew replies personally within three business days.