Map the process before you add AI

Almost every AI conversation with a firm starts with "we want to automate intake," or billing, or client updates. I understand the instinct, and it is usually pointed at the right problem. But it is too big to build, and the firms that get real value from AI do one unglamorous thing first: they write the process down, from the big picture to the individual tasks, before anyone picks a tool.

Why "automate intake" is a wish, not a spec

Intake is not one thing. It is a dozen steps and a few judgment calls, spread across the front desk, a paralegal, and an attorney. AI does not work on "intake." It works on one specific task inside it, like pulling the names out of a new inquiry so they can be checked for conflicts. Until the process is written down to that level, nobody can say what to build, what it will cost, or how you will know it worked.

A useful way to write it down is in five levels, each one more specific than the last.

The five levels, with intake as the example

1. Area. The broad part of the firm: client service.
2. Process group. Related processes within it: client intake, matter opening, and client updates.
3. Process. One process with a measurable outcome: every qualified inquiry has a conflict check, a booked consultation, and an engagement letter sent within two business days.
4. Steps. The sequence, with who does each one and where the decisions are: capture the inquiry, run the conflict check, qualify the matter, book the consultation, send the engagement letter.
5. Tasks. The small units inside each step. For the conflict check: pull the party names from the inquiry, search the conflicts database, list possible matches, and have an attorney review them.

Most firm conversations about AI happen at levels one and two. The useful work happens at levels three through five.

The outcome is what you will measure

Level three is the one firms skip most often, and it is the one that decides whether an AI project can succeed. Without a measurable outcome, an AI tool will run every day and produce plenty of activity, and you will have no way to know whether the process got faster, safer, or just busier. Write the outcome as one sentence with a number and a deadline in it. That is the line you check after the build. We go deeper on that in How to Know Your Firm's AI Is Actually Working.

The decisions are where a person stays in the loop

When you list the steps at level four, circle every decision. Is there a conflict? Is this a matter we take? Is the fee arrangement approved? Does this letter go out? Those are the points where someone with authority makes a call, and in a regulated practice they are exactly where a person stays in the loop when AI joins the process. The AI can prepare the conflict search. A lawyer decides whether there is a conflict. That is the idea behind The Approval Layer.

The tasks tell you what kind of AI fits

Level five is where AI actually works, and the shape of each task points to a different kind of build. Pulling names out of an email is extraction. Drafting a first engagement letter from your own templates is generation. Finding the answer to an intake question in your firm's knowledge base is retrieval. Each is a different tool with different risks, which is why we wrote Which AI Pattern Does Your Firm Actually Need?

And once the tasks are written down, some of them turn out to be pointless: a form field nobody reads, a copy saved in two places, an approval that adds nothing. Delete those before you automate anything. The cheapest automation is the task you stop doing.

I learned the value of this level on enterprise AI work. On a menu-reading pipeline I built at Southern Glazer's Wine & Spirits, matching menu lines against a catalog of more than 600,000 items, the progress came from knowing exactly which task we were improving and measuring it on the items that mattered most, where accuracy rose 12 to 15 percent. A firm process works the same way, at a smaller scale.

One page before any build

Put it together and you have a one-page map: the outcome at the top, the steps in order, the decisions circled, and the tasks under each step, with a note on which ones AI could take on. That page shows what to build first, where a person must approve, and how you will know it worked. It also keeps the project small enough to finish, one task at a time, instead of an "AI transformation" that never ships.

It is where Aday Interactive, Inc. starts a custom AI build. To see where your firm stands before you map anything, take the free AI Executive Readiness Assessment, or read how we approach custom AI.