Contents
Remove the bottleneck The four commitments The principles 01Start at the door 02What reaches a lawyer 03Supervise, don't do 04Keep the system honest 05Off-the-shelf never fits 06Make it land The end state What we're still working outRemove the bottleneck
The Flank Method is about one thing: removing legal as an operational bottleneck in the organisation. That is brutal to say, but it is true. Every deal, every hire, every supplier waits on legal. No one has ever removed that constraint, so no one has ever priced the upside of removing it. At minimum: materially better margins and faster sales cycles across the business.
Here is the tell. Most in-house teams now expect AI to reduce their reliance on outside counsel. Almost none have seen total matter cost fall. The tools are good, but they solve the wrong problem: they make an expensive person slightly faster at inexpensive work. A faster lawyer does not remove a bottleneck. Moving the work does.
So the Method moves it. Routine work runs on a supervised agentic system, and people stay involved exactly enough to take responsibility for the output: sometimes checking every result, sometimes encoding the playbook and auditing a sample. Where that responsibility sits inside the organisation, the work is insourced; where it sits with a provider, it is outsourced. The same system, supervised in a different place. Either way the economics match outsourcing, and the control never leaves your building.
What remains is a different legal department. Lawyers become what they were hired to be: experts running litigation, negotiating the genuinely contested terms in major contracts, setting house positions, carrying strategic projects like M&A. Routine execution leaves their desks entirely, and around it grows a new discipline of supervising the system and owning its output, which does not have to belong to lawyers. None of this resembles the legal intelligence platforms lawyers are using today. It is a different position entirely.
"I don't want my team looking at DPAs and NDAs anymore. I want it done. And I still want control."
A general counsel, on why they stopped shopping for a toolThe Method, in four lines
If you remember nothing else, remember these. Everything after is detail.
Legal is a bottleneck, because the wrong people are doing the work.
Expensive, judgment-trained people spend most of their time on routine, rules-based tasks. Every deal, hire and supplier waits on them. Remove that constraint and the gain is structural, not marginal, which is why it has never been priced.
The fix is not a faster lawyer. It is less work reaching the lawyer.
A tool accelerates the person inside the work. The Method routes the work to a supervised agentic system instead, so most of it never reaches a lawyer at all. The staffing model changes; the backlog changes; the headcount does not have to.
Stay involved only enough to own the output.
Human involvement exists to put responsibility somewhere a board, a regulator or a counterparty can find it. Sometimes that means checking every result; more often it means writing the playbook and auditing a sample. Owned inside the company, that is insourcing. Owned by a provider, outsourcing. The same system, supervised in a different place.
What is left is the work that always needed a lawyer.
Contested terms, house positions, litigation, the deals that decide the company's future. The lawyer moves above the process rather than sitting inside it, and a new supervision discipline grows up around the rest. It does not have to be lawyers who do it.
The beliefs underneath
The commitments are the what and the why. These are the operating beliefs that make them real.
Fix the system, not the individual.
Individual gains compound to ten percent. Redesigning where work enters, how it routes, who executes it and how it is supervised compounds to a different operating model.
Work gets done where work arrives.
The strategic surface is intake, not execution. A system that absorbs the channel work already comes through earns adoption by default. One that asks people to learn a new portal does not.
Supervise exceptions, not everything.
A function that reviews every output cannot scale. One that reviews nothing cannot defend itself. Supervision should fall where the risk is, and it has to scale faster than the volume.
You cannot govern what you cannot see.
The team needs a live, queryable picture of what was done, what was produced, what was overridden and what is drifting. Without it, the function runs on anecdote.
The record outlives any tool or any lawyer.
Documents, decisions and audit trails belong in durable storage below the tools, so they survive every vendor change, every model, and everyone who leaves.
Off-the-shelf never fits.
No two legal functions are the same. The gap between a general platform and a specific enterprise is bridged by engineers, on a platform deep enough to make that fast and safe.
Own your playbook. Tools are a commodity; outcomes are not.
Your templates, terms, fallbacks and escalation rules are your encoded judgment. That is the asset. Anyone can ship software; what a department needs is the work done.
01Start at the door
All legal work should enter through one door. Today it enters through dozens: email, a CLM intake form, procurement, a privacy portal, messages straight to individual lawyers. Consolidating intake is the first move, because the channel that already carries the most volume is where you earn adoption without asking anyone to change, and it is the only place every request passes through in a uniform way, which makes it the only honest dataset of what the business actually asks of legal.
In practice
Map every channel work arrives through before you design a new one. Most functions discover five to fifteen, weighted toward the informal ones, and the map itself is the diagnostic.
Pick the channel with the most volume, usually email, build the agent there, and let it absorb the rest by being measurably better, not by being mandated. Do not ask people to learn a portal; meet the request where it already lands.
Categorise at the door, not at the destination, and you have a time series of legal demand most teams have never had: volume by type, by requester, by month.
This is what Flank calls the Legal Front Door: an email-native intake layer that receives a request, triages it, and routes it into agentic execution or to a human. No portal, no form, no login, so adoption needs no change management.
02Decide what reaches a lawyer
Decompose the work and sort each piece into one of three buckets: what an agent does on its own inside guardrails, what an agent does while a human reviews the exceptions, and what only a human should ever touch. The purpose of the exercise is to make the third bucket, the work that genuinely needs judgment, the only work that reaches a lawyer. Most routine contracting belongs in the first two.
In practice
A worked example. A counterparty NDA arrives. The agent reads it against your playbook, redlines it to your positions, and either returns a finished draft or, where it meets a deviation it is not authorised to accept, drafts an escalation: the deviation, your position, the risk category, the playbook guidance, and any gaps it could not fill rather than inventing them. A lawyer sees only that exception, with the work already done up to the point of judgment.
Know the difference between an agent and a co-pilot, because the market blurs them on purpose. A co-pilot is a tool a lawyer operates. It makes them faster, but every contract still passes through them, so the bottleneck simply gets a better interface. An agent is a worker a lawyer supervises, and the work leaves the queue.
The honest use of both: co-pilots inside the lawyer's own work, agents for everything that can leave it. Leading with a co-pilot and calling it a strategy is the most common and most expensive mistake.
03Supervise the work, don't do it
Supervision is not a compliance feature you bolt on at the end to satisfy the GC. It is the product. It is the only mechanism by which the system improves, and the only way the team keeps accountability for work it no longer performs by hand. The right amount of involvement is whatever lets a named person stand behind the result, and no more.
In practice
Supervision has to be structural, not a signature. The supervisor can see what the agent did, on what document, with what reasoning, what changed, and can correct or escalate at any of those points. A correction updates the playbook, so the reviewer is teaching the system, not redoing its work.
Permission the system before any execution goes live: who can ask, who can see what was done, who can override, who can audit. The override list is the most important and the shortest. The audit log should not live anywhere the agent can write to.
Supervision comes in three shapes, and choosing between them is the insource-or-outsource decision. Your own lawyers supervise directly. A partner firm supervises on your behalf. Or Flank's supervision team holds the standard for you. In every case the playbooks, the escalation rules and the institutional knowledge stay yours: the work never leaves your building. The economics match outsourcing. The control does not.
04Keep the system honest
You cannot govern what you cannot see, and you cannot improve what you cannot measure. Build observability before you scale, not after you have lost it: a live, queryable picture of what the system did, what it produced, what was overridden and how that is changing over time, available to the legal team without going through the vendor.
In practice
Track distributions, not averages, because the tail is what hurts. The most useful single signal is the override: how often, in which direction, on which clauses, per section of the playbook. That is the actionable view of quality.
Watch for drift, so the system does not discover its own degradation the day a customer escalates, which is the most expensive way to find out.
Keep the record durable and below the tools. Storage can be a well-governed SharePoint; it does not need to be a CLM. Every artefact writes back as a new version with a clean diff and an attribution. When the executor's version and the record diverge, the record wins. The agent is a worker, not a filing cabinet.
05Off-the-shelf never fits
No two legal functions are the same. The templates, the terms, the systems, the house positions and the escalation paths are specific to one company, which is why generic software bought off the shelf almost never survives the trip from a good demo to live production. The gap between a capable platform and a specific enterprise has to be bridged by people: forward-deployed engineers who embed with the team, learn the work, integrate into the systems you already run, and build the thing that fits. It is the model behind the most demanding enterprise software, and it only works at speed because those engineers build on a deep, interoperable platform rather than from scratch each time. The platform is the vendor's to run and keep hardening. The judgment encoded on top of it, your playbook, stays yours.
In practice
The platform is what turns bespoke work into something fast and safe: the agent runtime, the conversational layer, the supervision and observability cockpits, the retrieval engine, enterprise-grade isolation and audit, and the integrations into your existing systems. The engineers assemble and configure these rather than rebuild them, which is why most teams are live in weeks, not quarters. Each engagement makes the platform more capable, so the next one is faster. This is the line between an agentic service and a consultancy: the consultancy rebuilds every time and stays expensive; the platform compounds.
Forward-deployed engineers do the work no off-the-shelf product can. They integrate into your channels, sit with your lawyers to extract the judgment that today lives only in their heads, encode it into playbooks, train and tune the agents, and keep the workflows running reliably in production. Flank does not sell a login and wish you luck. The last mile is different in every company, and the last mile is the whole job.
Because the platform plugs into what you already run, you insert at intake and do not rip and replace. Storage, matter management and downstream systems stay where they are.
Keep the playbook portable and exportable. Your encoded judgment is the asset and the source of advantage, and it should never be trapped inside a vendor's product. Tools are a commodity. Outcomes are not.
06Make it land
This fails on adoption, not technology. The change asks lawyers to delegate, a muscle many have not used since they made partner, and the relief arrives later while the discomfort arrives now. So frame it as delegation, not adoption. Measure the work the team no longer touched, not logins. And carry the economics with you, because that is the case that travels upward to the CFO.
In practice
The first ninety days are about trust, not throughput. Until a lawyer watches the agent do work they would have done themselves, no time-savings analysis will move them. Champions beat mandates, and change happens at the inbox, not the town hall.
The economics are the point you take to the board. In a large enterprise the legal budget divides roughly into thirds, weighted heavily to the outside: external execution is most of it, internal headcount a fifth or so, and software a small remainder.
This is a vendor swap, not a reorg. The budget already exists, wherever the routine work currently sits. The Method redirects it to a model that keeps the work, and the control, inside the building.
The end state
A request arrives and is handled the moment it lands, drafted to the team's own standard, with a few exceptions flagged overnight for a lawyer to decide in minutes. Routine work disappears from the lawyer's day. The headcount does not change. The backlog does.
Legal is no longer on the critical path. Lawyers spend their time on contested terms, house positions, litigation and the deals that decide the company's future. A supervision discipline has grown up around the rest, and it is not only lawyers who practise it. Capacity is no longer tied to the size of the team, and cost no longer scales with volume. The judgment that used to live in a few people's heads, or get rented from a firm, is encoded and owned.
This is not a smaller legal department. It is a legal department doing the work that was always worth its time, and nothing else. It is the operating model Flank is built to run.
What we're still working out
A method worth following should say what it has not solved. Junior development is the open one: if associates do not grind through the routine work, how do they build judgment? The current guess is that development shifts from doing the work to reviewing it, earlier in a career, but we do not consider it settled.
Nor is the optimal shape of a legal team once throughput is no longer tied to headcount, or the cleanest way to handle privileged data inside an agentic system, or where the supervision standard lands as regulators catch up. These are live arguments. Where the Method does not match your experience, we would rather hear it than smooth over it.
Insource enterprise legal work to supervised agents.
Flank is the operating model in this guide, delivered end to end. Underneath sits the platform: the agent runtime, supervision, observability, the retrieval engine, enterprise-grade security and the integrations into your systems, deep enough that building on it is fast and safe. On top of it, forward-deployed and legal engineers embed with your team, extract the judgment that lives in your lawyers' heads, encode it into playbooks you own, train the agents, and keep them running in production. A supervision team holds the standard over time. And the commercial model matches the Method: you pay for the work delivered, not for a licence, so Flank only earns when the work is done.