The short answer.

A firm in a regulated field plans AI the same way as any company: team by team, deciding which work AI should prepare, who owns each workflow and what gets built first. It adds three things. It starts with the business side of the firm, such as intake, billing, scheduling and reporting, where the work repeats and the rules are the firm’s own. It keeps every professional judgment with the licensed person, so a workflow prepares and a named professional reviews, decides and signs. And it settles which assistant is approved for client information before any real work goes in.

This page is not legal or professional advice. Your firm’s own obligations, under the rules of your profession and the law where you work, are for the firm and its advisers to read. This page describes how the plan is organized, not what any rule requires.

The rest of the method is in How organizations plan their use of AI: plan it, develop the workflows, deploy them.

Start with the business side of the firm.

A regulated firm starts its AI plan with the business side because that work repeats every week and the rules for it are the firm’s own. Intake, billing, scheduling and reporting run the same way in a law firm, an engineering practice, a clinic or an advisory firm. Nobody needs a license to prepare them well.

  • Intake.Turning each new inquiry into one packet: who the person is, what they asked for, what is missing and who should see it next.
  • Billing.Preparing a prebill with the unusual entries already marked, or drafting the follow-up on an overdue invoice for the billing lead to approve.
  • Scheduling.Gathering what each appointment, inspection or meeting needs beforehand, so the professional walks in prepared.
  • Reporting.Pulling the month’s numbers from the firm’s own systems into the partners’ or directors’ report.

These first workflows teach the firm how a workflow is owned, tested and kept current, on work where a miss is caught by the person who reviews it. The AI workflow examples for law firms show what this looks like in one profession, and the same pattern holds in the others.

The licensed person decides and signs.

In a regulated firm, every professional judgment stays with the licensed person who is accountable for it. A workflow prepares the material. A named professional reviews it, decides and signs. That line is written into each workflow, not left to habit.

  • The workflow prepares.It gathers the records, puts them in order, checks for what is missing and drafts a first version for review.
  • The licensed person decides.Advice, a diagnosis, a design approval, a filing, a recommendation to a client: each is made by the professional, by name, after review.
  • Nothing goes out unsigned.Anything that reaches a client, a court, a regulator or a patient waits for that person’s sign-off.

The same principle runs through every workflow we build, in any company. AI prepares. People decide. sets it out in full. In a regulated firm the person who decides is simply named more precisely: the professional licensed to make that call.

Approve one assistant before client work goes in.

A regulated firm decides which AI assistant is approved for client information before any client work goes into one. The decision names the assistant, the account it runs on, what information may go into it and who approved it. Staff using personal accounts for client work is the gap the decision closes.

FullyPowered workflows run in the firm’s own ChatGPT, Claude, Copilot or Gemini, on the firm’s own account, run by the firm’s own people. The firm’s information stays in the firm’s account. Is it safe to put company data into ChatGPT, Claude, Copilot or Gemini? covers what to check in each, and We work in your systems sets out how we handle a firm’s information when we build with it.

Write the decision down in two paragraphs and put it on the plan. Does a small company need an AI policy? shows how short that document is.

A one-page sheet to copy.

Fill in one sheet for each workflow on the plan before anyone builds it. Copy these six headings.

  1. The work.What the workflow prepares, what it starts from and what it produces.
  2. Who is licensed to decide.The role, and the person by name, who makes the professional call on this work. Write “none” if the work is purely business side.
  3. What information it touches.Client records, personal details, financial records, the firm’s own figures, or none of these.
  4. Which assistant is approved for it.The assistant and the account, and the date the firm approved it for this kind of information.
  5. Who reviews.The person who checks each output before it is used, by name.
  6. What is never sent without sign-off.Every output that reaches a client, a court, a regulator or a patient, and the person whose sign-off it needs.

One firm planning its first workflows.

Illustrative example.

Take a 45-person structural engineering firm. The operations director owns the plan. The firm lists its teams: engineering, project delivery, finance and client service. Finance starts with a month-end billing workflow that prepares invoices from the project records; the finance lead reviews each one, and the sheet says “none” under who is licensed to decide. Client service starts with intake: each new inquiry becomes a packet with the site, the scope asked for and the drawings received. The engineering team’s first workflow is a checklist that gathers the inputs for a design review. The sheet names the engineer of record as the person who decides, and says that no calculation, drawing or letter leaves the firm without that engineer’s signature. Before any project file goes in, the managing principal approves one assistant on the firm’s account for project records and writes it on the plan.

Every line of that plan is a decision a person made, and every professional call stays with the engineer who signs.

In a firm of under 200 people.

Most published advice on AI in regulated fields is written for large institutions with compliance departments and governance committees. A firm of 20 or 150 people plans differently.

  • The licensed people are in the room.The partner, principal or physician who signs is the same person who agrees the plan. Ask them which work they want prepared.
  • The firm’s advisers are already known.The firm has an outside counsel, an insurer, an accountant or a professional body it already asks. The firm asks them about its own obligations; the plan records the answer.
  • One assistant is enough.No platform program comes first. The firm approves the assistant it already pays for and plans what it is for.
  • One owner, by name.Usually the firm administrator, the operations director or a managing partner. Who should own AI in a company your size? covers the choice.

Small is an advantage here. The people who decide, the people who sign and the people who do the work sit close enough to settle the plan together, and that is where 10X the ROI of your existing AI comes from.

Where to go next.

The full method, with a one-page plan for the whole firm, is in the central guide to planning AI, linked in the short answer. When the plan is ready, How to roll out ChatGPT, Claude, Copilot or Gemini across a company covers putting the approved assistant in everyone’s hands.

FullyPowered Planner is where the plan lives. You set out your teams, their roles and the work between them, and choose what to build first; colleagues add their own teams. When you want help building the first workflows, a 3-Day Workflow Sprint or a 2-Week Workflow Sprint takes one piece of work to a tested workflow your people run.