A law firm's product is judgment, and judgment is accumulated: a clause fought over across forty deals, a regulator's habits learned through a decade of filings, a client's risk appetite mapped through years of close calls. Firms sell the accumulation. What almost no firm does is store the raw material of that accumulation in a form that holds its value.
The default storage media are archived matters nobody revisits, precedent that lives in particular partners' heads, and laterals who arrive with context and leave with it. Each one depreciates. The firm pays to create the knowledge once, then pays again every time it fails to find it.
The leak nobody itemizes
The losses are real, recurring, and absent from every management report.
A partner retires and takes decades of negotiation memory: not the documents, which stay in the DMS, but the knowledge of why the terms landed where they did, what that counterparty accepted last time, and which positions were tried and abandoned along the way. The documents remain. The reasoning leaves in an elevator.
An associate rotates off a matter, or out of the firm, and the matter's working context resets. The next team re-reads, re-derives, and re-asks the client questions the file already answers. Turnover is booked as a recruiting cost. Most of it is a knowledge cost.
Some of the leak is billed, which is worse than invisible. Clients pay for re-derivation until they notice, and sophisticated clients have started noticing: outside counsel guidelines increasingly prohibit charging for work the firm has, in substance, already done. The remainder gets written off, and partners see the write-offs at year end without ever seeing the cause.
And the firm's most valuable question, "have we seen this before," is answered by whoever happens to be in the room. The real answer sits across twenty years of matters. The available answer is the memory of the people at the table, which means the firm's effective experience is a fraction of its actual experience, and the fraction is decided by scheduling. Pitch teams feel this most: the same experience lists get rebuilt from memory every quarter because no system of record sits behind the question "how many of these have we done."
Assets that depreciate and assets that compound
The distinction worth internalizing is between knowledge stored as artifacts and knowledge stored as structure.
A filed document depreciates. Its findability decays as naming conventions drift and the people who remember it move on. Its context evaporates: which draft was final, what the redlines meant, why the indemnity reads the way it does. Ten years on, it is a string of words in a repository that a keyword search may or may not surface. None of this is the DMS's fault. It was built to store and retrieve artifacts, and it does. It was never built to remember what the artifacts meant.
A graph edge appreciates. When the same knowledge is stored as structure (this matter, for this client, against this counterparty, produced this negotiated position on this clause, with this outcome), every new matter added makes the older ones more retrievable and more useful, because there are more paths to them and more comparisons for them to anchor. We wrote a fuller treatment in our field guide to legal knowledge graphs. The short version: artifacts sit, structure connects, and connection is what compounds.
A filed document is a cost the firm already paid. A graph edge is the same knowledge converted into an asset that pays back on every future matter it touches.
The mechanics of compounding
Compounding is not a metaphor here. It has specific machinery, and every piece of it is buildable today.
- Closed matters ingest automatically. Closing a matter triggers ingestion into the firm's Context Graph instead of archival into cold storage. The end of the engagement becomes the start of the knowledge asset rather than the end of it.
- Outcomes and positions get extracted and linked. Executed agreements yield the negotiated result: caps, baskets, survival periods, the fallbacks accepted under deadline pressure. Linked to sector, deal size, counterparty, and year, they become queryable experience instead of buried text.
- Market intelligence emerges from the firm's own deals. What survival periods has the firm actually accepted in sponsor-side carve-outs, by sector and by year? That is not a treatise question. It is the firm's own deal history, structured, and it beats any survey because it is real and it is yours.
- Lateral onboarding is measured in days. A lateral who can query the firm's experience directly does not need two years of hallway apprenticeship to become effective. The graph is the institutional memory they would otherwise borrow one anecdote at a time.
- Pitches and pricing run on matter data. Claims of experience become demonstrable with specifics, and fee proposals rest on what comparable matters actually consumed rather than on optimism.
One constraint governs all of it: compounding cannot come at the price of confidentiality. The graph has to inherit matter-level permissions and ethical walls from the systems it reads, so that what compounds for each user is exactly what that user is entitled to see. Structure should amplify access. It must never widen it.
This is the layer where Reframe concentrates. The Context Graph is built per firm, inside the firm's own boundary, because the compounding asset belongs to the firm and should live nowhere else.
From personal capital to firm capital
The technology is the smaller half of the shift. The larger half is cultural. Partners have historically held knowledge as personal capital because the firm offered no better instrument for it: a book of business is portable, while a contribution to collective memory has traditionally earned nothing at compensation time. The rare exceptions prove the point: the firms that made knowledge contribution a named factor in credit decisions built the strongest precedent collections of their generation.
Firms that want the compounding have to acknowledge the commons in their credit systems, directly (recognizing contribution of matters, forms, and annotations to the graph) or indirectly (measuring the leverage a group achieves on top of it). Knowledge management professionals have argued for this for years, usually without infrastructure that makes contribution visible; structured capture finally makes it measurable, a shift we examine in our piece on knowledge management becoming infrastructure. None of it requires partners to become librarians. Capture happens as a byproduct of closing matters on a platform built for it; the cultural work is making sure contribution is seen and never punished.
The moat, and the clock
Foundation models are commoditizing on a visible curve. Every firm can buy the same intelligence, and within a pricing tier, largely does. Whatever edge exists in having a given model, a competitor can buy next quarter. Clients have noticed; panel questionnaires increasingly ask not whether a firm uses AI but what proprietary knowledge grounds it.
Nobody can buy your firm's twenty years of matters. That corpus is the one input competitors cannot acquire at any price, and structured, it is the durable advantage: the thing that makes a commodity model answer like your firm instead of like the market. The same structure cuts key-person risk in both directions: departures take less with them, and arrivals contribute sooner.
The reason to move now is arithmetic, not urgency theater. A graph started this year compounds across every matter the firm closes from here forward. A graph started in three years misses that context at its richest, because extraction from live engagements, with the team still present to validate it, beats archaeology every time. Compounding rewards time in the market. Reframe's process is built to start the clock quickly: a live, queryable graph inside the firm's boundary, growing with every matter that closes after it. The graph is most valuable to whoever starts earliest. That is not a sales point. It is how compounding works.