AI built for billable hours, not blog posts.

---

What’s actually broken in your firm right now

The associate who pasted a draft brief into personal ChatGPT last week did not do it to be reckless. She did it because the partner needed a cleaner draft by 5pm and she had not been told whether the firm’s Copilot license covered the work product. She used the tool she had at home because the firm hadn’t built her one at the office.

The partner is still doing $400/hour work that an associate plus a well-built AI agent could do in 40 minutes. Not because the technology isn’t there — Microsoft Copilot has been in your tenant for fourteen months — but because nobody’s built the workflows, written the prompts, trained the associate, and measured the result. The partner reads “AI hype” articles and reasonably concludes the tools aren’t ready. The tools are ready. The workflows are not.

IT was not in the conversation. AI got onto the corporate card through Marketing, or through a partner’s brother-in-law who runs a consultancy, or through the practice management vendor’s latest release notes. Now there are six different AI tools on six different bills and no one is sure which ones have your client data inside them. The compliance officer is asking questions that nobody can answer in writing.

The clients have started to notice. Three of your last fifteen RFPs asked for your AI use policy. Two asked whether AI is involved in producing the deliverables. One had a clause about training on client data. You answered them in ways your insurance carrier would probably ask you to revisit.

Three workflows we set up first

First-draft document review and summary

Before. A second-year associate spends 90 minutes reviewing a 40-page contract, drafting a memo for the partner, and flagging issues. Average: 90 min per document. Bottleneck: time.

After. The associate uploads the contract to a Copilot agent in your Microsoft tenant configured for this workflow. The agent produces a structured first-draft memo — issues list, ambiguous clauses, comparison against your firm’s standard term library — in 6 minutes. The associate reviews, validates, edits. Net: about 25 min per document. The partner sees the same memo format every time. Time saved per associate per week, on average: 4–7 hours.

Tooling. Microsoft Copilot in your tenant, with the firm’s standard clause library loaded into a SharePoint knowledge base. No data leaves your tenant. Output stays inside the same DMS or document store you already use.

Setup cost. Approximately one Custom Agent Sprint ($32,000). Ongoing inside Operate tier; no additional licensing beyond your existing Copilot seats.

Discovery and due diligence response drafting (legal) / Engagement letter and SOW generation (accounting)

Before, legal. Discovery response on a mid-complexity matter takes a senior associate 4 hours: re-read pleadings, locate relevant documents, draft response, internal review.

Before, accounting. Engagement letter and SOW for a new audit client takes 2.5 hours: pull prior years’ templates, customize for industry, run partner review.

After. An agent with read-only access to your matter or client folder (scoped, audited) drafts the response or letter against your firm’s templates. The associate or manager reviews and finalizes. Legal: 40–60 minutes per response. Accounting: 30 minutes per engagement letter. The agent does not file. The agent does not send. A human approves and ships.

Tooling. Custom agent built in ACS Automate or your tenant, reading from your matter management or practice management system via approved API or read-only file access. Permissions reviewed by your firm’s IT and managing partner before go-live.

Setup cost. One Custom Agent Sprint per workflow. Ongoing inside Operate.

Knowledge capture from senior staff

Before. Two partners are within five years of retirement. One of them is the only person who knows how the firm handles a particular line of work for a particular industry. That knowledge lives in his head and in a handful of folders nobody else opens. Succession planning conversations have happened. Knowledge transfer hasn’t.

After. A structured interview agent — text or voice — runs 30 to 60-minute sessions with the partner, pulling out matter taxonomies, decision criteria, client-handling patterns, war stories with the lessons attached. Output: a structured knowledge base, searchable from inside your firm’s existing tools, that mid-career associates can query when the partner is unavailable. The partner stays the expert; the knowledge stops being a single point of failure.

Tooling. ACS Automate-hosted interview agent + retrieval system, surfaced via Teams or your firm’s intranet. Knowledge stays inside your environment.

Setup cost. One Custom Agent Sprint for the first knowledge area, less for subsequent ones. Lives inside Operate or Strategic.

Confidentiality and regulatory

Microsoft Copilot inside your tenant operates on data you already own, under the same Microsoft 365 Data Processing Addendum you’ve already signed. Prompts and outputs are not used to train Microsoft’s foundation models. That covers most of what your associates need to do, most days, for most clients. It does not cover everything.

Personal ChatGPT in a partner’s home browser is different. Free-tier conversations may be used for training. Plus and Team tiers can be configured not to be, but the configuration depends on the account, not the firm. There is no audit trail you control. If a privileged document gets pasted in, you have no way to retrieve it or to prove it was not used. State bar opinions in California, New York, Florida, and a growing list of others now treat unrestricted use of general-purpose AI with client information as a competence-and-confidentiality issue under Model Rules 1.1 and 1.6. AICPA’s interpretation in ET §1.700.001 is consistent. Several professional liability carriers now ask about AI policy at renewal.

The practical position we recommend: AI on, but only inside your tenant. Personal accounts off, by policy, with enforcement at the identity layer. Written AI Acceptable Use Policy. Approved tool registry. Quarterly review. We build all of this during onboarding and maintain it for the life of the contract. The bar, the board, and the carrier get an answer that holds up to a deposition.

Compare the tiers

What could go wrong

Book an AI Readiness Assessment

Three weeks. Fixed fee. A real report and a 90-day roadmap. You keep it whether you hire us or not.

Book an AI Readiness Assessment · Or talk to us for 15 minutes — no pitch.