DEV Community

Precision AI OS
Precision AI OS

Posted on

An AI readiness assessment a small firm can run in one week

Here is a January many small firms will recognize. A return is nearly done and sits in a folder waiting on a K-1 from a client who has stopped answering email. Someone on staff has sent the same request four times. The partner who has to review it is booked solid. By February the document chase has eaten hours nobody billed, and by March the review backlog has pushed filing dates around. In April nobody can say where the month went. An AI readiness assessment is a low-cost way to stop guessing about which parts of that mess a machine could actually help with, and which parts only look like they could.

The costs stack up quietly. Hours spent chasing documents. Filing dates that slip. Time logged at the end of a long day and never put on an invoice, because writing it up feels like one more task. Receivables that age while nobody has the appetite to make the call. None of that shows up as a line item called overhead, but it is the overhead.

This is not a forty-page consultant deck. For a small accounting or professional firm, readiness is a week of looking at your own work. Five steps, done honestly, with the people who touch the work in the room.

Start with the jobs, not the tools

Pick the three jobs that repeat every week and cost the most hours. In most firms the candidates are familiar: chasing documents, triaging the inbox, following up with clients who went quiet, and reviewing time that was logged but never billed. Write each one down in plain language. "Send the missing items request and track who replied" is a job. "Improve client communication" is a wish.

For each job, count the hours. Not a precise time study. A rough weekly number from the people who do it, plus the delay it causes. If a request goes out three days late because the inbox was buried, that delay belongs in the count. So does the write-off when a partner eats the time to keep a client happy.

Then ask the only question that matters at this stage: is the work repetitive enough that a set of instructions could produce a draft? If the answer is "sometimes, depending on the client," write that down too. Ambiguity is a finding, not a failure.

Follow the data and name the approver

For each of the three jobs, write down where the data lives. Often it is in four places at once: the practice management system, someone's email, the client portal, and the accounting file. Note who can access each one, and whether that access is current. Former staff who still have live portal logins are a readiness problem most tools won't fix for you.

Then decide who approves anything that goes to a client or moves money. This is the step firms skip, and it is the one that keeps an unreviewed draft from reaching a client. A draft email to a client is still a client communication. A prepared journal entry that changes a balance is still a money movement. Name a person for each category. If the answer is "whoever is around," you have found the real blocker.

At this point you can write the one-page policy. Keep it short enough that people read it. It should say what client data may be used for, where it may go, who may approve output, and how consent is handled. For US tax preparers this is not optional. Under IRC section 7216, a preparer must get taxpayer consent before using or disclosing return information for purposes other than preparing the return. Feeding return details into an AI tool to draft a client summary, for example, is a use of return information. The policy step is where you find out whether the idea survives contact with the rules.

Record the baseline before you change anything

Before any pilot, write down today's numbers for each of the three jobs: hours per week, and turnaround from request to response. Keep it simple, one line per job. This is the only way to know in thirty days whether anything changed. Firms that skip the baseline end up arguing about impressions. Firms that keep it can point at a number.

Add one more line if collections are a problem: how many invoices are more than thirty days old, and how often someone actually calls. Chasing a document and chasing a payment are different jobs, but both are mostly follow-up, and both get dropped when the week gets busy.

A short list helps here. For each job, you should be able to fill in four lines:

  • The job, in one sentence.
  • Where the data lives and who can access it.
  • Who approves the output before it leaves the firm.
  • The baseline hours per week and turnaround today.

If you can fill in all four, you are ready. If you cannot name the job, the data, the approver, and the baseline, you are not ready, and no purchase will change that. You would be buying a tool to find a problem.

One more thing about the week. Do it with the people who do the work. A partner alone in a room will list the jobs they think happen. The staff who chase documents know which client never opens the portal, which request always comes back incomplete, and which follow-up gets ignored. Their version of the week is the one worth putting on the list.

The jobs that look easiest are often the ones with the messiest data. The jobs that look hardest are sometimes a single missing approval decision away from being safe to draft. A week of looking is cheaper than a year of a tool nobody trusts.

And when the week is over, keep the page. Revisit it in thirty days against the baseline. The point was never the document. The point was knowing what you are doing before you spend money doing it faster.

Precision AI OS builds an Agent OS for CPA and professional service firms. Agents prepare the work, and anything that goes to a client waits for a person's approval by default.

https://getprecision.ai/?utm_source=devto&utm_medium=social&utm_campaign=ai-readiness-assessment

Top comments (0)