Forecasting Law Firm Revenue When Every Matter Is Different

Ask most managing partners what the firm will bill next quarter and you get a number that arrived by feel. It is usually last quarter plus a bit, adjusted upward if the mood in the office is good and downward if a big client just went quiet. Nobody writes down how it was derived, so nobody can check it later, and when it turns out to be wrong there is no way to tell which part of it was wrong. The forecast was one undifferentiated lump, so the error is one undifferentiated lump too.

The reason firms accept this is that legal work genuinely resists forecasting. A manufacturer knows roughly what a unit costs and roughly how many units ship. A firm has a defence matter that might settle in six weeks or grind for three years, a transaction that dies at diligence, a flat fee estate plan that closes on schedule, and a personal injury file whose entire value depends on a resolution date nobody controls. Averaging those together produces a number with no meaning, which is exactly why the feel-based estimate survives. It is not obviously worse than the arithmetic.

But that is only true if you insist on forecasting the firm as a single thing. Break the forecast into layers by how knowable each one is, forecast each layer with its own method, and the picture changes. You end up with a number you can defend line by line, a range instead of a point, and, more usefully, a monthly conversation about which specific assumption moved rather than a vague sense that things feel slower than last year.

A revenue forecast is four questions, not one

The money that will arrive in the next two quarters is sitting in four distinct places right now, and each one behaves differently. Work you have already billed and not been paid for is accounts receivable, and it is the most knowable layer because the amount is fixed and only the timing is uncertain. Work you have performed and not yet billed is work in progress, where the hours are known but the billable value is not, because realisation happens between the timesheet and the invoice. Work you are contractually engaged to do but have not started is committed future work, where the scope is roughly known and the timing depends on the matter calendar. And pipeline is everything that has not been signed, where both existence and value are open questions.

Forecasting all four with one method is the core mistake. Firms typically apply pipeline-style optimism to receivables, which makes cash look closer than it is, or apply receivables-style certainty to pipeline, which makes an unsigned matter feel like money in the bank. The discipline is to treat each layer with the method its uncertainty deserves, then add them up with the ranges preserved rather than collapsed. When somebody asks why the forecast changed, you can point at the layer that moved instead of shrugging at the total.

Start with receivables, because that money already exists

Your receivables ledger is the only part of the forecast where the amount is not in dispute. Someone reviewed the work, an invoice went out, and a number is now owed. The only real question is when it lands and whether all of it does. That makes AR the anchor of the whole exercise and the layer you should model most precisely, because getting it wrong is unforced. Age every open invoice into buckets, then look at what your own firm historically collects out of each bucket rather than importing a benchmark from somewhere else.

The pattern that matters is client-type specific. Corporate and insurance clients paying through an e-billing platform run on a predictable but slow rhythm, with a review step that can bounce lines back and reset the clock, and a LEDES 1998B submission that either passes validation or does not. Individual clients on hourly retainers behave completely differently, and their payment probability drops sharply once a matter concludes and the relationship pressure disappears. Model those cohorts separately. A single blended collection curve across a mixed client base will be wrong for every client in it, and the direction of the error changes as your client mix changes, which makes it worse than useless.

Work in progress needs a haircut before it enters the forecast

Unbilled time is real work and future revenue, but the number in your system is not the number that will be invoiced. Partners write down blocks that will not survive client review, some entries are duplicative, some tasks were staffed at the wrong level, and every firm has a gap between the value of recorded hours and the value of hours actually billed. If you drop raw WIP into a forecast at face value you will overstate the quarter, and you will do it consistently rather than randomly, which means the forecast develops a permanent upward bias that everybody eventually learns to mentally discount. At that point the forecast has stopped doing anything.

Apply your own historical write-down rate, calculated per practice area and ideally per timekeeper level, before WIP enters the model. The second variable is age. WIP that is four months old bills at a lower rate than WIP from last week, because the narrative is harder to defend and the client's memory of the work has faded, and in some matters the delay itself becomes the client's argument. This is where the operational fix and the forecasting fix are the same fix. Casely turns every unbilled hour on a matter into one itemised draft invoice in a single click, and a firm that bills weekly rather than in a monthly scramble is not just collecting faster, it is producing WIP data with far less noise in it.

Committed future work is where the matter stage tracker earns its keep

The third layer is work you are engaged to do and have not done yet. This is the layer most firms cannot see, because it lives in the heads of the people running the matters. A litigation file sitting between pleadings and disclosure has a predictable next chunk of work attached to it. A transaction that just cleared diligence has a drafting and negotiation block coming. If you know where each open matter sits and what your history says the next stage typically costs and how long it typically takes, you can forecast a substantial part of the next two quarters from matters you have already won.

This is why a matter stage tracker is a financial instrument, not a status widget. Casely's stepper is configurable per firm and per practice area, so the stages match how your work actually flows instead of a vendor's idea of it, and once every open matter sits at a known stage you can attach historical stage duration and stage value to each one. Combine that with the deadline diary, which ties deadlines to the matter with next-date auto-tracking, and you get timing anchors that are not guesses. A hearing date or a filing deadline is a real constraint on when work happens, and work is when revenue is created.

Weight the pipeline by stage, not by enthusiasm

Pipeline is the layer that ruins forecasts, because unsigned work is where optimism lives. The fix is to stop asking whether a prospect will convert and start asking what your own conversion rate is at each intake stage. An enquiry that has only submitted a web form converts at one rate. One that has had a consultation converts at a much higher rate. One that has received an engagement letter and not yet signed converts higher again, and one that has signed but not paid the initial retainer is a different thing entirely. Multiply the expected value at each stage by your own observed conversion rate for that stage and sum it.

Two disciplines make this hold up. First, weight by stage only, never by an individual's confidence, because confidence ratings drift upward across a team and nobody ever revises theirs down. Second, forecast the timing separately from the value, because a matter that converts in March and a matter that converts in June are different quarters even at the same value. Track referral sources while you are at it. Casely's contact labels tag roles and referral sources, and knowing that a third of your pipeline traces to one referring firm is a forecasting fact before it is a business development one, because concentration means the whole layer can move at once.

Contingency practices break the model, and pretending otherwise is the mistake

A contingency matter has an expected value and an unknowable resolution date, and those two facts do not combine into a forecast line. You can estimate the recovery range from comparable matters. You genuinely cannot estimate when a defendant's insurer decides to move, when a court reaches your case, or when opposing counsel's schedule clears. Firms that blend contingency expectations into the operating forecast produce a number that looks fine on paper and then leaves them short in a month where nothing resolved, which is how otherwise healthy plaintiff firms end up borrowing against fees they were sure were arriving.

Run contingency as a separate line with different rules. Forecast the operating side of the firm, meaning payroll, rent and everything else that arrives on a fixed schedule, entirely from hourly, flat fee and retainer revenue. Treat contingency recoveries as capital events that fund reserves, partner distributions and case investment rather than the monthly run rate. Track the portfolio by expected value and by disbursement exposure, because the money the firm has advanced on those files is a real cash outflow with no matching inflow yet. Casely handles hourly, flat-fee, contingency and blended billing natively, so the arrangement lives on the matter and does not have to be reconstructed at forecasting time.

One caution worth stating plainly. Contingency and conditional fee arrangements are permitted in some jurisdictions, restricted in others, and prohibited outright for certain matter types such as family and criminal work in many places, with distinct regimes for conditional fee and damages-based agreements in England and Wales and province-by-province variation in Canada. Confirm what your own regulator and court rules allow before building any of this into a business model.

Practice areas have velocities, and mixing them destroys the forecast

A firm-wide forecast built on firm-wide averages hides the only thing that would have been useful. An estate planning practice producing flat fee matters that open and close within weeks has a short, predictable cycle where revenue tracks intake volume closely. A commercial litigation practice has an eighteen-month cycle where this quarter's revenue was determined by intake from over a year ago. Add those together and the aggregate tells you nothing about either, and worse, a slowdown in the long-cycle practice is completely invisible for a year because the short-cycle practice keeps the total looking healthy.

Forecast each practice area on its own clock and then add the results. This also tells you which lever actually works when the number comes up short. If the gap is in a short-cycle practice, intake and marketing can close it inside the quarter. If the gap is in a long-cycle practice, nothing you do this month will change this quarter's revenue, and the honest answer is to manage cost while the intake work you do now pays out four quarters from now. Firms waste enormous energy applying short-cycle fixes to long-cycle problems, and the aggregated forecast is what lets them keep confusing the two.

Forecast a range, and publish the assumptions next to it

A single number invites a single argument, and it is always the wrong argument. Produce three scenarios instead, built by varying the assumptions you are least sure about rather than by scaling the total up and down by a percentage. A conservative case might assume slower collection in your oldest AR bucket and lower pipeline conversion. A base case uses your observed historicals unchanged. An upside case assumes a specific named matter converts and a specific large invoice clears. Each scenario should be traceable to a named assumption, so that when reality lands somewhere in the range you know which assumption was closest.

Writing the assumptions down is what makes the forecast reviewable. When the quarter closes short, the question is not whether the forecast was bad, it is whether collection slipped, conversion dropped, WIP failed to convert to invoices, or a matter stalled at a stage. Those are four different problems with four different responses, and a bare number cannot tell you which one you had. A firm that keeps its forecasts and reviews them gets a compounding asset, because after four quarters you know your own biases, and a known bias is easy to correct for.

Update when matters move, not when the month ends

The monthly forecast refresh is a habit inherited from businesses whose reality changes monthly. A firm's reality changes when a matter changes state. A case settles, a transaction signs, a client pays a large invoice, a prospect goes quiet, a court adjourns a hearing by four months. Each of those is a forecast event, and waiting three weeks to reflect it means the firm spends three weeks acting on a number it already knows is wrong. The stage tracker makes this practical, because a stage change is a discrete event you can hang a forecast update on rather than a subjective judgement someone has to remember to report.

Keep the monthly cadence for the review meeting, but let the number itself move continuously underneath it. The meeting should not be about producing the forecast, which is a data exercise. It should be about the deltas: what changed since last month, why, and what it implies. Firms that get this right stop having the forecast meeting where everyone stares at a spreadsheet somebody built the night before, and start having the shorter, more useful one about the three matters that moved and the one client whose payment behaviour just changed.

Measure your forecast error, or you are guessing with extra steps

Almost nobody scores their own forecast, which is why most firm forecasts never improve. Keep every forecast you publish, and when the period closes, compare it to actuals layer by layer. You are not looking for a perfect hit. You are looking for a consistent direction of error, because bias is fixable and noise mostly is not. If your AR forecast is reliably five percent optimistic, your collection curve needs adjusting. If your pipeline layer comes in high every single quarter, your stage conversion rates are stale or your intake team is scoring stages generously.

Two or three quarters of this changes the character of the exercise. The forecast stops being an opinion the finance-minded partner defends and becomes a model with a known error band, which is what lets you actually make decisions on it. Hiring an associate, signing a lease, funding a case, taking a distribution: these are all bets on a forecast, and knowing that your model runs seven percent optimistic on pipeline is worth more than a false decimal point. Reporting that pulls from the same matter and billing records that generate the work, rather than from a separately maintained spreadsheet, is what keeps the comparison honest over time.

Getting a forecast your firm can actually run on

Start smaller than you think. Take your current open invoices, bucket them by age and by client type, and apply what your firm has historically collected from each bucket. That alone usually produces a more accurate ninety-day picture than whatever number is currently being used, and it takes an afternoon. Add the WIP layer next with a realistic write-down applied, then committed future work once your open matters are all sitting at an accurate stage, then weighted pipeline. Each layer you add makes the forecast better, and none of them require the previous one to be perfect.

The infrastructure question is whether your systems can answer the layer questions without someone rebuilding a spreadsheet every quarter. If matter stage, unbilled time, invoice age and referral source all live in the same place as the work, the forecast is a report rather than a project, and it can update whenever a matter moves. If they live in four places, the forecast will be assembled quarterly by whoever is least busy, and it will be stale before the meeting starts. That is the practical case for matter management software that holds the operational and the financial picture together instead of splitting them.

The point of all this is not precision for its own sake. It is that a firm which knows where next quarter's revenue is coming from makes different and better decisions than one which does not, and it makes them earlier. It hires before the gap opens rather than after, it chases the invoice that is about to age out of collectability, and it notices a long-cycle practice slowing while there is still time to feed it. None of that requires forecasting genius. It requires refusing to answer four different questions with one number.