The first question came before I was hired. A twenty-person law firm, fully remote, and the owner said: I don’t know if my staff are productive or profitable.
Two more followed. My marketing isn’t working, and I’ve already replaced the agency twice. And: I wonder why I’m not making a profit.
Three questions about three different parts of a business, and the same answer to all of them. Nobody could see anything. Not because the firm was careless. It had case management software, an accounting system, a CRM, and a bookkeeper. It had data. What it didn’t have was any instrument that turned the data into an answer to a question an owner would actually ask on a Monday.
Which is the argument here, and it took me a while to state it this plainly: you cannot improve productivity, marketing or profit until you know whether the numbers describing them can be trusted. Optimization comes second. Most firms attempt it first, because it feels like progress and the alternative feels like bookkeeping.
Time tracking before dashboards
I built the reporting myself, with AI-assisted development, and the part that took longest wasn’t the code.
The firm already owned time-tracking software. What it lacked was the discipline to use it, and no dashboard was going to exist without that. So the first project wasn’t a system at all. It was a mandate, issued in June, with real accountability attached. Daily time entry became a documented requirement of the role, carrying the same weight as any other operating standard.
That is not a popular way to start. The objection came immediately and it came from timekeepers who hadn’t yet logged a single entry: this will take me away from the legal work I’m supposed to be doing.
I want to be fair to that objection, because it sounds reasonable and it’s the one every firm hears. It also can’t be evaluated until the thing being objected to has been measured, and logging time takes a few minutes a day. The objection was an efficiency argument made in the absence of any data about efficiency, which is precisely the condition the whole exercise existed to end.
It was resolved by making transparency a condition of the role rather than a request. Not a threat and not a negotiation. An expectation, written down, applied uniformly. Some people find that harsh. It is considerably less harsh than the alternative, which is a firm where nobody’s contribution is visible and pay is set by impression.
What we compared
Once entries existed, the analysis was simple enough to describe in a sentence.
Every hour got sorted into three categories: billable client work, non-billable client work, and non-client work. Then that was set against billed revenue per person, which the firm already tracked well, and sliced by week.
The three-way split is the part worth stealing. Most firms track billable and treat everything else as one lump, which hides the difference between someone doing real client work that never got billed and someone doing administration. Those are completely different problems with completely different fixes, and a single non-billable bucket makes them look identical.
The categories themselves will differ by business. An agency, a clinic, a consultancy, an accounting practice: each has its own version. What carries across is the shape: separate the work that produces revenue, the work that serves clients without producing revenue, and the work that keeps the organization running. Three buckets rather than one, chosen so that each points at a different fix.
The first version didn’t have the split. It tracked total hours logged, which told me who was busy — something everybody already knew and nobody was arguing about. The categories were what made it diagnostic rather than descriptive.
What it found, which was not what anyone expected
Here is the part I would have preferred to write differently.
Before the mandate, the firm’s billable share looked respectable. After it, with several months of clean data, the billable share was lower.
That reads like a decline and it isn’t one. The old number was wrong in the flattering direction, for two reasons that only became visible once the categories existed. Non-billable client work was being logged as billable, because there was no category for it and it felt like client work. And administrative time was frequently not logged at all, so it vanished from the denominator entirely.
Inflate the numerator, shrink the denominator, and you get a billable share nobody had any reason to question. That was the number the firm had been managing by.
So the honest finding was not that productivity improved. It was that the firm’s picture of its own productivity had been flattering, and the instrument’s first act was to take that away. Total hours captured rose by roughly an eighth, with one fewer person logging. That’s capture improving, not output improving, and conflating the two would be the exact error the system was built to stop.
The productivity gains came later and are measurable elsewhere. Return on production labor — revenue divided by production payroll, so it doesn’t depend on how anybody categorized an hour — went from 4.7X to 6.6X on a quarterly basis. Roughly a forty percent increase.
That doesn’t prove the reporting caused it. Several things changed in the same period. What it shows is that economic production improved while the operating system was being built, which is a weaker claim and the only one the evidence supports.
Two thresholds, not one
The metrics and the compensation structure built on them went live the following January, seven months after the mandate.
Each person’s chart carried their weekly billed dollars against two lines rather than one. A minimum, which is the point at which the role covers its own cost to the firm. And a bonus threshold above it.
Two lines do something one line can’t. Below the minimum there’s a conversation about the role. Between the lines, the person is doing the job they were hired for and nobody needs to say anything. Above the bonus line, the role has already covered what it costs the firm, so what comes after it can be shared more freely.
Set that minimum from the economics of the role, its fully loaded cost and the return it has to produce, rather than from what the team managed last year or what the budget hopes for. A threshold built from cost is harder to argue with than one built from last year’s performance or somebody’s judgment, because the inputs are visible and the person can check them.
What a Monday view needs
If you run a firm between ten and fifty people, this is the order I’d build in.
Make the input reliable before you build anything on top of it. Not because it’s satisfying, but because every hour spent on visualization before the data is trustworthy is an hour spent making a wrong number easier to read.
Split your categories finely enough to distinguish different problems. Two buckets tell you how much. Three or four tell you what to do.
Instrument what someone can act on this week. Per-person production against a threshold changes a conversation on Monday. A quarterly summary changes nothing, because by the time it exists the quarter is spent.
Expect the first reliable number to look worse than the flattering baseline it replaces. This is the part that stops most firms, and it’s the point at which the exercise is working rather than failing.
That last one is the whole of the first step of the method I use, the TAG method — See the Truth. Implement Actions. Achieve your Goals. Establishing what’s true is not a preliminary to the work. It is the work, and it usually costs something to do, because the picture being replaced was more comfortable than the one arriving.
The version you can open
The system I built for that firm is confidential and stays that way.
So I built another one. Meridian Command is a full firm-intelligence dashboard running on a fictional firm and invented data — the same architecture, the same modules, none of anyone’s information. Acquisition, subscriptions and forecasting on one side; production, matters, scorecards, attendance and finance on the other. It exists because the argument in this article is easier to check than to describe.
It isn’t a product and there’s nothing to buy. It’s the thing itself, which is a different kind of evidence than a description of the thing.