That has to stop now. Every firm is deciding how AI changes the business of law. The executive who understands the economic engine better than most needs to be in that room. What follows is your way in, and your way to stay.
And you are going to need it.
The first wave of legal AI aimed at the work lawyers bill. So the demos went to practice groups, the licenses went to IT, and the pilots went to innovation. The spend decisions were made before ROI was ever calculated.
But every spend decision comes with a question that many are afraid to ask, or have not asked at all: what did it return? Which workflows actually improved, and by how much, measured against what baseline, at what cost per task, per workflow, per matter? In most firms today that question has no owner. Ask who holds it at yours, and notice the pause.
The discipline that answers it, measurement against a baseline, controls on a process, an audit trail behind a number, is not a technology discipline. It is a finance discipline. It already lives in your office.
Here is what this guide hands you, in order.
The capability firms are racing to hire, decoded for the finance office.
A three phase loop that runs AI adoption the way your office already runs internal controls.
Two questions every AI initiative in the firm must answer, and the seat they earn you.
12 to 16 peers, a half day, zero vendors. Where the capability and the conversation start.
One pressure is about your own realm. The other is about everyone else's.
E-billing appeals. Collections escalation. Rate management. Matter budgeting. Reconciliation. Vendor spend. Every one of these workflows is cost, so every hour returned drops straight to profit. None of them touches billable hour politics.
And the golden datasets already exist: closed invoices, resolved appeals, historical write downs. No other territory in the firm starts this ready for the loop you are about to read.
When a client asks why a faster matter costs the same, the technology has already left the conversation. What remains is pricing architecture, realization, leverage. The question will be answered by whoever has modeled it, at your firm or at the firm across the street.
Advantage will belong to the firms whose finance office can model production under compression before clients force the conversation. That modeling cannot be delegated to a vendor.
Two pressures, one answer. There is a name for the person who can read a workflow, test a system, and price the consequence. The market is already paying for that person. You are about to meet the role, and you are closer to it than you think.
Firms have started hiring for this role. The postings say legal engineer or forward deployed legal engineer, the market for them is accelerating, and the vendors keep the best ones on their own side of the table. Here is what the postings miss: the people firms manage to hire arrive without the financial and commercial judgment you have spent a career building. There are not enough unicorns, and the firms that go unicorn hunting will end up filling the role with people who cannot carry it.
We believe firms are better positioned to grow this capability than to buy it. So the question shifts from who to recruit to who to develop: you, someone on your team, or both. That is your call to make. Starting is not optional, because the capability is built, not bought, and building takes time. The investment is in yourself and your team, and it begins now.
So what is the role? Not a technologist. A person who holds two kinds of judgment at once: judgment about the work and judgment about the system that would do the work.
In lawyer facing teams the first axis is legal judgment. In the CFO's realm it is financial judgment. The structure is identical: judgment about the work, judgment about the system, and the bridge between them.
Click any component. The judgment expands.
Knowing what correct, defensible numbers look like. The firm's economic model as a system, not a spreadsheet.
Leverage, realization, rate structure, and matter mix interact to produce profit. Seeing that interaction as architecture is the CFO's native strength, and it is the axis this whole capability is built on. This one is your home turf. It is selected for, not taught.
What is worth building at all. Incentives, adoption, internal politics, and what clients will actually pay for.
Artificial intelligence changes the production equation. Commercial judgment is knowing which changes matter to margin and positioning, and which are demos. It is also the judgment that predicts whether a partnership will adopt what gets built.
Decoding how the work really happens, not how the process diagram says it happens.
Every financial process has a stated workflow and a real one. The report that pulls from six systems and gets hand corrected before anyone sees it. The appeal logic that lives unwritten in one billing coordinator's head. Workflow intelligence is the trained skill of finding the gap: task level mapping, exception hunting, sitting with the people who do the work.
Enough command of models, agents, and integration to know when AI is the answer and when it is overkill.
You do not need to build. You need to be a sophisticated buyer: what data does this actually run on, what is the failure rate on our edge cases, what does a run cost. The frontier model is often overkill. Fluency is knowing which questions expose that.
Evals, golden datasets, failure categorization, cost per run. Measuring what the system actually does.
The demo always works. Evidence discipline asks whether it works on your data, at your scale, on the exceptions your team spends most of its time on. This is a scientific mindset, not an engineering skill, and it is the closest of the six to what your office already practices every day.
Extracting the unwritten from people's heads, and defending the system twice: as an engineer and as an owner of the business case.
The CIO speaks infrastructure. Partners speak client relationships. You speak margin. Translation is carrying a technical capability into the room as an economic outcome, and carrying an economic goal back out as a technical requirement. The executive who can translate shapes the conversation.
No one needs the full hexagon. Score yourself honestly and the shape is spiky. That is expected. Now think about the team you have, or the team you could have, and compose it so the overlay covers the shape. The point is knowing where your spikes already are, and where you are weak.
Click a profile. The dashed line is the benchmark.
Two of six components are already yours at strength. The rest are learnable.
A capability without a method is a title. What separates the forward deployed legal engineer from every innovation role that came before is a working loop, and it happens to be one your office was built to run.
This loop is forward deployed legal engineering. You need to learn it, because everything else in this guide runs through it. Each phase is critical, each pass builds the evidence for the next, and one phase, evals, is about to feel very familiar. Run the loop once on a single finance workflow and it returns hours and a working system. Run it repeatedly and it becomes the firm's way of adopting artificial intelligence with proof instead of faith.
Sit with the work until the real workflow shows itself. Not the process on the intranet: the one with six systems, hand corrections, and rules that live in one person's head.
The output is an operating map precise enough to show exactly where a system can compress, eliminate, or transform the work. Most firms skip this and buy tools. That is why most implementations disappoint.
You have run this play before. In your world it is called internal controls. Golden datasets built from your own closed matters. Pass rates. Escalation thresholds. An audit trail behind every output.
An eval is a control on a process that happens to be intelligent. SOX shaped thinking is the native language here, which is why the scariest concept in AI turns out to be the most familiar thing in your office.
Integration with the systems you already run: billing, GL, reporting. Nothing ripped out, nothing replaced. Sandboxed first, on anonymized data, in parallel with the current process.
Autonomy graduates on evidence. The system starts as a first draft a person reviews, earns first pass status where a person reviews exceptions, and runs alone only on narrow tasks the evals have validated.
The first pass through the loop returns hours. Every pass after that returns proof. Proof is scarce right now, and the questions heading toward your firm are economic ones almost nobody can answer with evidence yet.
Artificial intelligence compresses production time. The billable hour prices time. When time compresses, the current model breaks or transforms, and either way the modeling lands in the finance office. The hour does not disappear tomorrow. But the firm that has modeled the scenarios negotiates the transition, and the firm that has not reacts to it.
Illustrative economics, not benchmarks. Your firm's numbers are the ones that matter, and the loop is how you get them.
| Metric | Illustrative value | Note |
|---|---|---|
| Blended rate | $685 per hour | Working assumption |
| Realization | 89% | After write downs |
| Associate hours | 1,820 per year | Billable target |
| Leverage | 4.2 to 1 | Associates to equity partners |
| Revenue per lawyer | $1.05M | Annualized |
A firm shaped like this is the starting point for both scenarios that follow.
| Metric | Illustrative value | Delta |
|---|---|---|
| Production speed | 1.4x baseline | +40% |
| Hours billed, same book of work | 1,300 per year | −29% |
| Revenue per lawyer, pure hourly | $745K | −29% |
| Revenue per lawyer, blended pricing | $1.02M | most of the gap recovered |
Under moderate adoption, pure hourly billing compresses revenue by roughly a third. Blended structures with alternative fee components recover most of it, but only if the finance office architects the mix before clients dictate it.
| Metric | Illustrative value | Delta |
|---|---|---|
| Production speed | 2.5x baseline | +150% |
| Hours billed, same book of work | 730 per year | −60% |
| Revenue per lawyer, pure hourly | $360K | −66% |
| Revenue per lawyer, value based architecture | $1.15M | +10% |
Under aggressive adoption the hourly model stops working. But a firm that prices value while artificial intelligence cuts input cost can grow revenue per lawyer through the transition. That architecture is designed in the finance office or not at all.
Honesty is part of the method. The compression is real; where it settles is not yet measurable. Four questions remain open across the market, and they are the questions a room of peers can examine more honestly than any single firm can alone.
Do standard rates hold, tier, or split into machine assisted and artisanal work?
Nobody knows yetDoes realization recover as pricing adapts, or keep eroding under client pressure?
Nobody knows yetWhat does the pyramid look like when associates produce at multiples of today?
Nobody knows yetDoes capability keep clients loyal, or does it make switching easier than ever?
Nobody knows yetThe thinking is what earns you the standing to say, to everyone running AI anywhere in the firm:
Those two demands, made by the executive who owns the firm's controls, change the conversation. Consistency replaces the demo. The wedge replaces the enthusiasm. Some will end up calling this position the firm's evidence officer; the title matters less than the standard, and the standard is yours to set.
Start in your own realm, where the datasets are ready and the politics are quiet. One workflow, one embed, one eval. The learning compounds fast, and the first result becomes the exhibit that resets how the whole firm buys.
A room can open the door for you, and ours will. But the seat is earned by the capability and kept by the evidence. A finance office that runs its own loop is self dependent: it adopts artificial intelligence on its own proof, at its own pace.
A half day for Am Law 200 CFOs, held in a hosted boardroom under Chatham House rules. No vendors, no demos to sit through, no slideware. A working session that ends with the conversation no CFO can have inside their own firm.
The role, the two judgments, the six components. Where you already stand.
A finance workflow decoded in the room: the e-billing appeal, stated process versus real.
A live build, one concept at a time: model, prompt, tools, memory, audit trail. You see exactly what you have been buying.
Evals, pass rates, escalation thresholds. The scariest concept in AI, translated into your native language.
The leadership position, and the two questions your office puts to every AI initiative in the firm.
The four unknowns, peer to peer, under Chatham House rules.
You leave with the vocabulary to lead the AI conversation instead of auditing it afterward, a controls frame for every AI decision that crosses your desk, the map of your first workflow, and a bench of peers staring at the same questions you are. The room convenes once. What it starts compounds all year.
Two decades advising global law firm leadership on strategy and business model design. Professor at Indiana University Maurer School of Law, where he is training the first law student cohort of forward deployed legal engineers. Built the Legal Business Design Hub at Richmond Law, a Fast Company Innovation by Design winner. Writes The Brainyacts, an AI and law briefing read by roughly 7,000 legal professionals.
Technical builder with two decades inside legal technology, including a company he founded and sold to Thomson Reuters. Runs NexLaw Labs, where he designs and deploys AI systems inside legal businesses. Long C suite relationships across Am Law firms and a working command of firm financials. He runs the live agent build in the room.
Legal Transformation Institute. Attendance by invitation.