Buying a services firm and running it with AI agents is just about pricing the model. You’ll still need the process, and the story changes there.
You can buy the clients. But you can’t treat the firm as a stack of tasks waiting for AI agents.
Walk the work for a week, and you see something else: the same job taking three different paths depending on who opened the email and whether the file arrived complete. Nobody wrote the exceptions down because the exceptions are the job.
The work is not what the workflow says. Sit with twenty live jobs and observe. Write down the trigger, inputs, handoffs, wait time, and rework. Count how many jobs actually follow one path. If most of them fork, you do not have an automation target yet. You have variation. Learn and kill the extra paths first. Make one way the default. Also make the exception visible. Then decide what a machine is allowed to touch.
Speed in the wrong place is a trap. Drafts that arrive in minutes do nothing if the only person who can sign them already has a two-week pile. You will feel busy, and the firm will not get more work out the door. So, just watch where jobs stop. That pile is the business. Put the new capacity there.
People are not a change-management slide. The work only stays standard if someone enforces it on a bad Tuesday. The person who knows every client preference will not become a cheerful reviewer because you bought software. Some will leave. Some will quietly go back to the old way while the agent runs in a tab nobody checks. You will pay the old wages and the new tools at the same time, and a review layer on top. So, plan cash for that overlap.
Clients are not frozen either. The easy slice of the work (clean books and obvious filings) is the same slice they will try to do themselves first. What they still pay for is the ugly remainder: the missing document, the owner who wants a person to say it is fine. If you bought a firm whose revenue is mostly the easy slice, your cost may fall while the invoice gets harder to defend. Underwrite the mix.
A loan does not wait for the process to mature. Trucks break. Listings rot. A manager quits. The improvement work has to fit inside a week that is already full. If your plan only works after the firm is calm, it’s for a different company.
Buy it if you can stand in the work and see the paths. Write the standard and cut the extra motion. Put eyes on the pile that actually limits output. Keep the people who hold the exceptions long enough to put those exceptions on paper. Then use AI agents on the path that finally repeats.
If you cannot do that part, you did not buy a platform. You bought someone else’s Monday.