If you run a firm of five to fifty people and want to know how to implement AI, here is the short version. Pick one workflow with a clear before and after. Name one person who owns it. Put the data that workflow needs in one place. Build a system that produces a first draft for a person to review, not a system that acts on its own. Measure one number for a month. Then do the next one.
That sequence is boring on purpose. The firms that get value from AI are not the ones with the cleverest tools. They are the ones that got the order right.
Where do you start?
Not with the work that carries your name. Start with the work that eats your week and that nobody will miss.
In every professional services firm we have scored, the hours cluster in the same places: taking new clients in, producing documents, and chasing people for the things the documents need. Producing documents is the biggest number and the worst place to begin, because a first draft that is wrong in a way you did not catch is a real risk. Intake and chasing are smaller numbers, but the before and after is obvious and the downside of a mistake is a follow-up email.
Intake
Every inquiry answered and qualified in seconds, booked onto the right calendar, with a summary for the person taking the call.
Chasing and status
One live record per matter or client, with the system asking for what is missing and telling people where things stand.
Documents
First drafts from your own precedents and templates, every figure traced to its source, for a person to edit.
Ownership
One name, a few hours a month, and a one-page list of what is allowed. This comes first, not last.
Choose whichever of the first two is worse for you. If you lose work because the phone goes to voicemail, intake. If your Fridays disappear into "did they send the bank statement yet", chasing.
Who owns it?
One person. Not a committee, and not the vendor.
The owner does not need to be technical. They need to be the person who already cares about that workflow, who will notice when it misbehaves, and who can say yes or no to a change in under a day. In a law firm that is usually the operations lead or the partner whose practice group goes first. In an accounting practice it is the manager who runs the busy season. In a recruiting firm it is whoever owns the database.
Give them two hours a week during the build and a few hours a month after. Write down, on one page, what the system is allowed to do without asking and what it must hand to a person. That page is the governance. Most firms never need anything longer.
Where does the data live?
Where the workflow needs it, which is almost never where it is today.
A small firm's information is spread across an inbox, a practice management tool, a shared drive, a spreadsheet someone built in 2019 and three people's memories. Every AI pilot that stalls, stalls here. The model is asked to answer a question whose answer is in four places, and it either guesses or gives up.
The fix is not a data warehouse. It is a single live record for the thing the workflow is about: the matter, the client, the candidate, the property. The system reads from the tools you already use and writes that record. Everything else, the drafting, the chasing, the status answers, sits on top of it. The first system you build is mostly this layer. The second and third systems are cheap because the layer is already there.
What should the system actually do?
Produce a first draft for a person, and prove where every piece of it came from.
A draft reply to an inquiry, with the caller's details filled in. A draft chase email, naming the document and the date it was promised. A draft of the report, with every figure linked to the line it came from. The person reads, edits, sends. Nothing leaves the firm without a human pressing the button.
This is not caution for its own sake. It is what makes the system trustworthy enough that people use it. A system that acts on its own gets switched off the first time it is wrong. A system that drafts and shows its work gets corrected, and the corrections make it better.
What does the first 90 days look like?
Two working sessions to agree the workflow, the owner, the rules and the one number. Three to six weeks to build the record and the first draft loop, with the owner reviewing output weekly. Then a month of daily use, measuring the one number and fixing what the owner flags.
At the end of ninety days you have a system in use, a person who owns it, a number that moved, and a data layer the next system can sit on. That is the whole goal. Anything more ambitious in the first ninety days is how pilots die.
How do you know it is working?
One number, chosen before the build, measured the same way before and after.
Minutes from inquiry to first reply. Documents still missing a week before the deadline. Hours a month on status calls. Days from close to the report going out. Pick the one that matches the workflow, measure it for two weeks before anything is built so you have a baseline, and read it every Friday afterwards.
If the number does not move in a month, the system is wrong or the workflow was. Either way you find out in a month, not a year.
What goes wrong?
Three things, in order of how often we see them.
No owner. The vendor demos, the partners nod, nobody is responsible on Tuesday morning, and the system quietly stops being used.
Starting with the most interesting workflow. Drafting advice or opinions is the thing everyone wants to see. It is also the thing with the highest cost of a wrong first draft, the least agreed rules, and the hardest measurement. Do it third.
Leaving the data where it is. A model asked to read from six tools at once will be slower and less accurate than a person, and the firm will conclude AI does not work. It works fine on one record.
What to do this week
Score your firm across the four areas above. Pick intake or chasing. Name the owner. Write the one page. Measure the baseline. Then decide whether to build it yourselves or with someone who has built it before.
If you want the scoring done for you, we built a four-minute roadmap that ranks the opportunities for your firm type and says what each one would take.