How to build your law firm's management reports with AI

TLDR: A firm is governed with one page a month, not a dashboard. Production, economics and pipeline, with the ledger in Xero feeding the money half and Claude drafting the reading. The number that matters is not what you billed, it is what survived the trip from worked to banked, and the two gaps in between have different causes and different fixes.

Why the best month can be the worst month

March was the firm’s biggest billing month in two years. In the third week of April there is a conversation about whether payroll clears, and nobody can explain how both things are true at once. The bank balance says one thing, the billing report says another, and the honest answer sits in a gap neither of them shows.

Between an hour worked and a dollar in the operating account there are two leaks, and a firm that watches only its billings cannot see either. The first sits between what was worked and what was billed. The second sits between what was billed and what was collected. They are caused by different people and fixed in completely different ways, and most firms find out about both a quarter late, from the wrong number.

Follow the hour from worked to banked

Three figures, in order, and the two gaps between them.

Worked. Billable hours recorded in the month. This one is only as good as the time recording habit behind it, and if entries are reconstructed on Fridays the whole page inherits that error.

Billed. What actually went out on invoices. The distance from worked to billed is realization, and it is made of specific decisions: hours written down before the bill went out, work sitting in unbilled WIP because nobody drafted the invoice, time cut because a partner felt the total looked high.

Collected. What arrived. The distance from billed to collected is a different animal entirely, made of clients who pay slowly, clients who dispute a line, and invoices that aged past the point where anyone chases them.

Calculate both gaps from your own two numbers and resist the urge to hunt for an industry benchmark. A published average says nothing about a four-lawyer practice carrying one large contingency matter, and comparing against one tends to end in an argument about whether it applies rather than a decision about what to change. The comparison that works is your firm against your firm last quarter.

Put the firm on one page

Three blocks, each with its comparison against the previous month and the same month last year. A figure with no comparison beside it does not govern anything.

The last two rows are not business metrics and they do not belong in a trend line. A missed deadline and an overdrawn client trust account are both problems of a different kind, and the only acceptable value for each is zero. Any other number ends the reporting conversation and starts a different one, which is why they sit at the bottom of the page where they cannot be averaged into anything.

The test for every other line is the one that kills most dashboards. Which decision changes if this moves? A page that passes that test gets read every month. A forty-metric dashboard gets built once, admired, and never opened again.

Take the numbers from systems already running

A report assembled by hand dies in its second month, so the build rule is that every figure comes from something that already produces it as a side effect of the work.

The money half comes out of the ledger. Xero starts at $25 a month and gives billed, collected and aged receivables directly, provided invoices carry a matter and a practice area tag from the day they are issued. That tagging is the entire build cost of this report, and it happens during the month-end close or it does not happen at all. Retagging two years of invoices is a project. Tagging the next one is a dropdown.

Production comes from time entries. Pipeline comes from the intake log, which needs to record where each enquiry came from or the marketing half of the page is guesswork.

This page deliberately stops at the firm as a practice. Revenue, margin and cash as a company, with the closing date and the audit trail behind each figure, belong to financial reports, and the two are read side by side rather than merged. Production hours and pipeline conversion have no ledger source, which is exactly why they need a page of their own.

A number with no system behind it is telling you which system to build next. Profitability by practice area does not exist without daily time recording, and pipeline conversion does not exist without an intake process. Leave the line off the page and write down what it is waiting for. A report that grows at the pace of the operation is honest. One that is padded with estimates teaches everybody to distrust the whole page.

Turn the numbers into a reading

The page needs a short narrative and no more than that. Three paragraphs: what changed enough to notice, what the data suggests caused it, and what deserves a decision this month.

Claude starts at $17 a month and keeps a permanent free tier, and this is a good use for it because the job is nuance rather than volume. Give it the figures with their comparisons and it returns the draft. What it cannot see is the context that actually explains the month, so a partner adds that: the one large matter that distorts everything, the seasonality of the practice area, the client who always pays in the second week of the following quarter.

Gamma is freemium with paid plans from $10 a month, and it earns its place only when the report has an audience beyond the person who wrote it. A partners’ meeting or an annual review is worth a deck. The ordinary month is not, and building one anyway is how a thirty-minute habit turns into a two-hour production nobody sustains.

Hold the meeting, then check the last one

Thirty minutes, monthly, with the page circulated beforehand and actually read. Fifteen minutes of questions about the report, fifteen minutes of decisions. Three decisions at most, each leaving the room with one name and one date attached.

The decisions this circuit tends to produce are specific. A practice area that loses money on every matter gets repriced or declined. Non-billable hours that keep climbing get investigated, because the answer is usually administration that belongs in document automation or file and deadline management. Receivables past ninety days trigger the collections sequence instead of another month of intending to call. Falling enquiry conversion sends somebody back to how the firm markets itself.

The next meeting opens by reading out the previous three decisions and what happened to them. That five minutes is the difference between managing a firm and holding a meeting about a firm, and it is the part that gets dropped first.

Four ways the page will mislead you

The contingency fee that lands. One matter resolves and the month looks extraordinary. It is not a good month, it is a good year arriving on one day. Note it in the reading and look at the rolling twelve months, or you will make a hiring decision on a number that will not repeat.

Trust money read as revenue. Funds held for a client are not the firm’s money, and a page that shows a healthy balance without separating the trust account from the operating account is describing a firm that may be doing worse than it looks. Keep them apart on the page as strictly as they are kept apart in the bank.

WIP that quietly ages into nothing. Unbilled work sits in a system where nobody is uncomfortable about it. It does not appear in receivables because it was never invoiced, so it never shows up as a problem. Track the age of the oldest unbilled work and it stops being invisible.

The discount nobody recorded. A partner cuts a bill before it goes out and the write-down leaves no trace, so realization looks like a collections problem when it is a pricing decision made privately. Recording write-downs as write-downs is what makes the first gap readable at all.

What to count to know it is working

Three numbers, counted before you change anything and again after two full months.

Count the days between a matter’s work being done and its invoice going out. Count the dollars of unbilled WIP older than sixty days on the last day of the month. Count how many of the previous meeting’s three decisions were actually done by their date.

That third one is the one people skip and the one that predicts everything else. A firm that closes out two of three decisions a month will fix its realization eventually. A firm that closes out none of them can measure whatever it likes and nothing will move.

The approval boundary

Software can pull the figures, compare them to last quarter, draft the reading and build the deck. It should not decide to write off a receivable, change a rate, or drop a practice area.

Those are decisions with money and client relationships attached, and the person who makes them has to be able to explain them a year later. The useful thing automation does here is remove the four hours of assembly that used to stand between a partner and the decision, not make the decision faster.

Keep the month’s page, the underlying figures and the three decisions with their owners. A year of those is the most valuable management document a small firm will ever have, and it costs half an hour a month to accumulate. The rest of the firm’s operation lives in AI for legal and law firms.

Frequently asked questions

Does a solo practice need this?

The half-page version, yes. A sole practitioner has the same two gaps between worked and banked, and usually a worse WIP problem because there is nobody to notice the invoice that never went out. The monthly meeting is with yourself, and it works if you keep the same rule: three decisions, written down, checked next month.

How is this different from what my accountant already sends me?

An accountant’s package tells you what happened to the money. It does not tell you that family law lost money on every matter this quarter, that your best biller has the worst realization in the firm, or that enquiry conversion halved in March. Those come from joining time entries to invoices to collections, which is work nobody does unless the firm asks for it.

Can I put client information into an AI tool to build this report?

Build the report from aggregate figures and the question mostly goes away. Totals by practice area, hours by timekeeper and receivables by age bucket carry no client information at all. If you find yourself wanting to paste matter detail in to get a better narrative, that is the moment to stop and ask whether the better narrative is worth the exposure, and the answer is usually no.

What if the numbers say a whole practice area loses money?

Check the numbers twice before you act, because the usual cause is a tagging error or one distorting matter rather than the practice area. If it survives that check across two quarters, the decision is repricing, changing how the work is delivered, or declining it. Doing none of the three and revisiting it next month is also a decision, just an expensive one.

When is this not worth doing?

In the first year of a firm with a handful of matters, where you already know every client by name and every open invoice by heart, the page tells you nothing you did not know on Tuesday. The threshold is roughly the point where you can no longer list your open matters from memory. Building it before then is a habit worth having, but it is a habit, not a discovery.