How the Simulation Works

You don't have to trust a black box. Here is the whole method, in three steps.

Step 1 · The Baseline
We build your firm, then let it run.

We don't start from a formula. We stand up a working model of your firm — the partners, the associates, the AI tools already in the building — and let them work real matters, quarter after quarter, for several years.

Every quarter, drafts get written and reviewed, hand-offs succeed or fail, clients pay early or late, people stay or leave. The profit that falls out of all that motion is your baseline. Nobody typed it in. It's what your firm produces when it runs exactly as it does today.

That's why the baseline should look familiar. If it doesn't, the model is wrong, and the report tells you where to check.

Produces → the baseline
Step 2 · The Recommendation
We test every change — alone, then together.

There are five moves on the table: how you bill, how work is handed off, how you pay for AI adoption, how fast you act on results, and how flat the pyramid is.

We try each one on its own, then in combinations, rerunning the entire firm each time. A change that only looks good on paper falls apart here, because it has to survive the same rework and write-offs as everything else.

We keep the combination that lifts profit the most — and we note the order, because some moves only pay off after another one clears the way.

Produces → the recommendation
Step 3 · The Stress Test
We run the winning plan again and again to see if it holds.

One good result can be luck. So we take the recommended plan and run it across dozens of fresh scenarios — the same firm, but different rolls of the dice on which matters land, which hand-offs break, and who walks out the door.

If the plan comes out ahead in nearly all of them, it's real, and we say so. If it only wins on average, we tell you to trust the direction but treat the dollar figure as provisional.

That's the difference between a forecast and a stress test. This is the stress test.

Produces → the confirmed result
What to trust
Direction and order are solid. Check the dollars against your ledger.
The direction and the order are solid. They come from running your firm thousands of times, not from an opinion. The dollar amounts are calibrated to a firm like yours, not pulled from your ledger — so check them against your own numbers before you quote them. Every figure in the report traces back to the record at the end of it.
This is a comparison engine, not a prediction of next year's P&L. It doesn't tell you what will happen. It tells you which road ends up ahead, and why.

Under the hood
How each part is actually simulated

The summary above is the shape. Here is the machine. Nothing below is a metaphor — it's what the engine does, quarter by quarter. Read it if you want to know exactly where a number comes from before you stake a decision on it.

The unit of work
Every matter is a chain of hand-offs. The seams are where value leaks.

Each matter runs through a fixed litigation workflow — intake, conflicts, engagement letter, staffing, research, document review, drafting, the senior-associate review, the partner review, filing, billing, collection, closure. Every step is a hand-off from one role to the next, and every hand-off carries two numbers set by your firm's structure:

Seam risk
How likely meaning breaks at that hand-off. The partner review is the highest in the firm (0.8); filing and collection are near zero — they run on standardized rails.
Seam tacitness
How much of the judgment there is irreducibly human. 0 = a checklist a tool can codify (legal research, engagement letters). 1 = pure partner judgment (settlement authority, the final read). AI is strong where tacitness is low and fails where it's high.

This is why the firm leaks value at the top of the chain, not the bottom. The cheap steps are safe. The expensive judgment is where a confident-looking AI draft is most likely to be wrong.

The three quality signals
How a broken hand-off is modeled, and how it's counted.

Hand-off failure

Modeled
When AI processes a step it can handle, it either succeeds or loses meaning at the hand-off. Whether it fails is fixed by the state, not a coin flip — the same matter, step, and quarter always resolve the same way — checked against that step's error rate. Volume spikes push failures up; a hand-off whose context has been codified fails about 30% less often.
Counted
Hand-off failure rate = broken hand-offs ÷ all hand-offs that quarter. Measured against every transfer, not "matters with any problem" — over a 15-step matter that would saturate at 100% and tell you nothing.

Redline rework

Modeled
Measured only at the partner review — the seam where the firm's real value lives. Counts the drafts the partner had to genuinely rewrite (the hand-off garbled the meaning, or the partner overrode the AI), not the normal routing of sending work back.
Counted
Redline rework rate = rewritten drafts ÷ partner reviews. This is the law-specific tell: AI automates the 80% of drafting that was never the value; the redline is the 20% that is. Fast drafting with a flat redline rate is the signature of automating the wrong thing.

Exception rate

Modeled
Any decision that needed a partner to step in beyond the designed process — the tail becoming supervision and rescue.
Counted
Exception rate = escalated decisions ÷ all decisions. A firm with a genuine, designed escalation path resolves some of these before they cost money.
The five moves
What each lever actually changes in the engine.

Each move flips a specific setting, which changes specific behavior, which lands on the money through three channels: collections, profitability, and billable hours.

1 · Move to flat fees

flips pricing from hourly → fixed-fee

Under fixed fees, every hour AI saves is margin you keep — profitability rises with AI use. Under the hour, the same speed bills less, so AI adoption quietly drags margin down. And staying hourly while your clients push for alternative fees leaks collections outright. Transactional work, already effectively fixed-fee, softens the hourly penalty.

2 · Pay for AI adoption

raises the comp incentive 0.3 → 0.9

Sets the ceiling on how far adoption can climb — from wherever your partners already are, up toward ~95%. How hard the money bites depends on your comp model (lockstep firms move together and respond harder; eat-what-you-kill firms resist) and on whether a few rainmakers own the book and can simply ignore a firm-wide push.

3 · Act on results faster

shortens the observe-to-act lag 5 → 1 quarters

Doesn't change how high adoption can go — changes how fast you get there. A short lag means the firm reaches its adoption ceiling early, so the benefit compounds over more of the run instead of arriving too late to matter.

4 · Codify the hand-offs

turns on seam codification

Every few quarters the firm permanently fixes its single worst remaining hand-off — builds the precedent library, standardizes the transfer — and that seam stops leaking (its failure rate drops for the rest of the run). Fewer broken hand-offs means less redline and fewer exceptions. It costs a little margin each quarter to fund the knowledge-management work.

5 · Flatten the pyramid

lowers associates-per-partner

AI compresses billable hours, and a steep pyramid has more junior hours to compress — so the steeper the pyramid, the harder AI cuts utilization and the more billable time is lost per partner. Flattening reduces that exposure. This is the one lever that reaches profit through hours rather than rates.

Where the numbers come from. Each of these effects has an explicit strength — how much a codified seam cuts rework, how much of a saved AI hour flat-fee pricing lets you keep, how far comp moves adoption. Every one is a named, sourced number with a plausible range, not a hidden constant. You can reset any of them from your own experience, and the range is exactly what produces the "give or take" band in the report. Nothing is a black box on purpose.
The human loop
People aren't a constant. They respond, and it feeds back.

Between quarters, the engine runs the human dynamics. Associates who lose faith in how AI is being rolled out leave faster — and one departure raises the odds of the next in the same group, so churn cascades. Meanwhile trust in AI spreads through the firm's actual influence network, where a skeptical (or convinced) knowledge-management partner moves everyone around them.

This is why the same plan can hold in one scenario and wobble in another: the people react, and their reaction changes next quarter's adoption and quality. It's also why the stress test in Step 3 matters — it's the only way to see whether a plan survives the human response, not just the spreadsheet.