Dare to pivot: bring legal work in-house

A law firm is built to solve matters, not to become the client organization’s memory. Hourly billing prices work performed, not capability retained. The firm is rewarded for effort. The organization gains when the next matter needs less of it. Anyone who has bought legal work at scale knows the signs: the memo nobody asked for, the second round nobody needed, the team sized to the staffing model, not to the question.

The second problem is quieter. Every matter sent outside must first be explained: the business, its history, what was tried before, what nobody wrote down. Context leaves in briefings and comes back in memos, and something is lost each way. The firm never knows the business as the inside does, and the inside pays, matter after matter, to teach it. This is the briefing bottleneck, and most of what it costs appears on no invoice.

The third problem is the one nobody puts in writing. Instructing outside counsel moves more than the work. It can move the responsibility with it: an external opinion is a defensible basis for a decision, which is sometimes exactly what a matter needs and sometimes a way of not deciding it. Work kept inside offers no such cover, and that is its discipline. It requires a named person who answers for the judgment.

Under all three sits the older constraint: the reading. Before anyone can judge, someone must go through the record, the contracts, the law. Reading costs time and money, so organizations buy it outside by the hour, and with the reading go the pace, the price and the knowledge the work produces.

The firms know it. Their argument now is how to price work the hour no longer measures, which is what an intermediary argues about when part of its work can be done without it.

AI systems can now do much of the first-pass reading, extraction and comparison. They are fast, tireless and cheap, and they can be made to show the ground for what they say. They still answer for nothing. The cost of reading falls, and with it one of the main reasons the work had to leave the house. What is scarce now is judgment: deciding what the reading means and standing behind it, before a court, a regulator or a board.

Inside means that the record, the analysis and the memory stay with the organization, on systems it already controls. What accumulates there is more than cheaper advice. An organization that holds its own legal memory can see what recurs: which of its terms cost it, which disputes it keeps having, which questions it answers too slowly. Knowing that is what lets it put its attention where the risk actually sits.

The first pass no longer has to be sent to a service the organization does not control: models good enough for that work run on hardware it controls. What counts as good enough is set by the checking that follows the reading, not by the best model available anywhere. Where a matter is covered by professional secrecy, or where a contract or a supervisory requirement limits the systems the data may sit on, an external service is not excluded outright; it becomes a question of what a supplier may see, what it keeps and what it could be compelled to produce. Running the model inside does not answer that question. It removes it. A file need not leave the house to be read.

Anyone who tried these systems two years ago and put them aside was judging them at their weakest. They have improved quickly since, and what is built now will run on models that do not yet exist. A new model is tested on the work before it is trusted with it, exactly as the last one was. So the model is the part that gets replaced, and the standard, the record and the checks are the part that lasts.

Every incoming draft is measured against the organization’s own standard: the terms it accepts, the departures it will trade, the ones it will not. Each deviation is marked with what it would cost before anything is conceded, so what reaches the person who approves is short. A departure with no discernible purpose is marked too, and saying so ends the argument early.

Each negotiation adds to a knowledge base the organization keeps. Much of that accumulation is work for agents that run on their own: reading each incoming draft against the standard, recording what was conceded and what it cost, and reporting what has moved since anyone last looked. Read across a year, it shows how the industry’s terms are moving, where practice is tightening, and what to prepare for rather than react to. It can also call the standard itself into question, where a clause has cost the organization more than it protected. When a regulation lands or a counterparty shifts its terms, the position is rebuilt that week, inside, rather than reordered from outside at the next crisis.

A dispute is decided in part by who understands the record first, and by who identifies the point it turns on and then keeps everything pointed at it. That coherence, held in correspondence, in negotiation and in court, often decides the outcome.

The record is gathered in days rather than months: the contract file, the mailboxes, the archives, read in full, with whatever cannot be read logged as such. Out of it come a sourced chronology, the weak points, and the two or three findings the matter actually turns on, while the other side is still reading.

A party that knows where its case is strong can settle from strength and litigate only what is worth litigating, holding the same account of the facts throughout. A party still discovering its own record gives ground in the wrong places.

Each element then carries its standing: established, contested, open, computed, withheld. Exposure appears as a span with the findings that move it, and options follow in the order they should be taken, each naming what it costs, what it protects and what would change it. The same knowledge base grows with every matter.

Some of the work runs on a schedule rather than on request: new drafts checked against the standard overnight, a position re-tested when a deadline nears or the other side files, the questions the organization has decided matter kept under watch. Once it is set up, the schedule is the organization’s to run. It is the difference between a legal team that can be called and one that has already looked.

A dispute or a large transaction is a system of moving parts: a notice period, a finding that may or may not be established, an amount that changes with it, a date that limits them all. Written out in prose it runs to pages, and a board still cannot hold it.

The method makes the whole situation visible on a single page. A transaction is shown as the departures from the standard and what each one costs. A dispute is shown as the findings that decide it and the amounts that follow from each, with the dates that constrain them. And because the variables are live, the picture is not fixed: when a finding is established, a term is conceded or a deadline passes, the exposure changes on the same page and the consequence can be seen at once. The reasoning and the sources sit a step underneath, so any part of the picture can be opened and checked. Legal advice stops being a document written on one day and read on another. It becomes something a board can look at, question and use while the matter is still moving.

Every fact that carries weight cites its source, and every statement of law is tied to the provision or judgment relied on, checked against the source rather than recalled. Each document ends with what it could not ground and what it left outside its scope, and that list is read first. Each reissue records what changed and where each open question stands. The attorney signs. A document that names its own gaps can be checked. One that does not, cannot.

An organization that keeps its own record can brief outside counsel in an afternoon. The briefing bottleneck runs in reverse: the sourced chronology and the two or three findings go out with the instruction, so the firm arrives to a record it can check rather than a business it has to learn, and the weeks that used to come first are not billed at all. What is bought then is judgment, and a sparring partner for the people who have to decide. Courtroom advocacy can still be bought. The case knowledge should not be.

What is built is a capability, not an answer to one question. The standard is written down and made checkable, the record is made searchable and traceable to its sources, the checks are built and set to run, and the people who will own the work are shown how it is done, on the organization’s own systems and under its own licences.

A capability built inside a live matter is built badly, because the matter takes the attention that building requires. It starts instead from the contract standard, from the files of past matters, or from the questions the organization already knows it answers too slowly, so that the capability is in place before the next matter needs it.

One step needs a single letter and no mandate. A client may ask counsel to account for a mandate and to hand over what counsel received in the course of it, subject to local rules on counsel’s own working papers. Asked across the last five years of the organization’s own matters, that request shows where its legal memory currently sits. The same applies to this engagement as to any other.

The attorney designs the work, verifies the law and signs what is legal advice. The fee is fixed in advance. The mandate is built to end, and what stays behind is a capability the organization owns rather than a dependency it renews. The first engagement should make the second one smaller. Dare to pivot.