A commercial real estate firm we have been working with had a partner who, over eleven years, negotiated hundreds of commercial leases on behalf of landlord clients across a concentrated geographic market. She knew which concessions landlords in that market routinely accepted, which lease structures had been tested in local courts, and where the leverage points were in tenant negotiations at various deal sizes. When she left the firm in early 2025 to take an in-house role, almost all of that knowledge left with her.
The deals she worked on were still in iManage. The final executed leases were there. But there was no mechanism to query that archive in the way you would query a knowledgeable colleague. The new partner who took over her practice area had to reconstruct her institutional context from scratch, through a combination of reading old files, building his own experience, and making mistakes that she would not have made.
This is a story about a single departure, but the dynamic it describes plays out continuously in most transactional practices, at smaller scales and with less drama. Institutional knowledge about how a firm handles specific deal types accumulates in people, not in systems. When those people leave, the knowledge does not transfer cleanly.
The Three Forms Knowledge Takes in a Transactional Practice
It helps to be precise about what kind of knowledge is at stake. In a transactional practice, institutional knowledge about deal-handling takes three forms that are distinct in how they are stored and how they are lost.
The first form is explicit, documented knowledge: the form books, playbooks, and standard position guides that firms maintain to record their standard practices. This knowledge is relatively durable because it is written down. It is also relatively thin, because documenting positions is labor-intensive and the forms tend to lag behind the evolution of actual practice. Many firms have form books that are years behind current practice on at least some clause types.
The second form is tacit positional knowledge: the internalized understanding that experienced attorneys have of what positions work in practice, which concessions are real and which are not, and how to negotiate toward favorable outcomes in specific contexts. This knowledge is primarily in people's heads. It accumulates through deal experience and supervision, and it erodes when experienced attorneys leave or when the practice group loses continuity.
The third form is latent deal knowledge: the knowledge that is technically preserved in closed deal files but is not meaningfully accessible because there is no way to query it efficiently. This is where the largest gap tends to be. A firm with fifteen years of closed transaction files has substantial knowledge about what positions it has taken and how negotiations have resolved. Virtually none of that knowledge is usable for the next deal, because the format it is stored in, individual document files without systematic indexing, does not support the kind of query that would make it useful.
The Closing Moment as a Knowledge Event
The moment a deal closes is also the moment the firm's knowledge about that deal starts degrading. Not the documents, which remain in the DMS. The knowledge: the understanding of which provisions were the contested ones, which concessions were made and why, what the counterparty pushed hardest on and where the final compromise landed.
That knowledge exists in a concentrated form right at close. The attorneys who worked the deal have it fully in mind. The negotiation history shows the progression of positions. The final agreement reflects the resolution. If someone else at the firm needed to handle a very similar deal the next week, the information to brief them well would be fully available.
Six months later, the attorneys have moved on to other matters. A year later, some of them may have left the firm. Three years later, the matter is an archive entry in iManage that no one actively thinks about. The knowledge has not been destroyed, but it has become practically inaccessible. Finding it and extracting it would require reading through many documents across many files, which is only feasible when someone has a specific reason to do so and sufficient time.
The iManage Problem Is Not Search
When we describe the latent deal knowledge problem to transactional attorneys, a common response is: "But we have full-text search. We can find anything in iManage." This is true but misses the nature of the problem.
Full-text search solves the problem of finding specific language, provisions, or document fragments when you know what you are looking for. It does not solve the problem of understanding, across a body of comparable deals, what positions the firm has typically taken and what outcomes have resulted from those positions.
Suppose an associate needs to know: in commercial agreements in the technology sector where the firm represents the vendor, what limitation of liability structures has this firm used, and what caps has it successfully maintained? Full-text search cannot answer that question. It can find documents containing limitation of liability clauses, but it cannot tell you which deal types those clauses appeared in, which ones were accepted versus negotiated, or what patterns emerge across a comparable set. That query requires structured knowledge, not full-text retrieval.
What Structured Deal Knowledge Looks Like
August is built on the premise that the knowledge in closed deal files can be made accessible if it is indexed in the right way. The index we build is not a document index. It is a deal characteristics index: each closed matter is associated with a set of attributes that describe the deal type, party positions, industry sector, economic parameters, and governing terms. Each major clause type in the final agreement is extracted and associated with that deal's attributes.
The result is that when an attorney is working on a new deal and wants to know what the firm has historically done on a specific clause in a specific deal context, the answer can be surfaced quickly from the indexed bank rather than through manual file research. The latent knowledge in the closed files becomes queryable in a way that full-text search alone does not enable.
This does not reconstruct the tacit positional knowledge that walks out the door when experienced attorneys leave. There is no automated substitute for the contextual judgment built through years of deal experience. What it does is preserve and make accessible the explicit record of outcomes, the layer of knowledge that firms have historically allowed to become inaccessible simply because storing it in a queryable form required more investment than it seemed worth at any given point in time.
The Investment Timing Problem
Most of the difficulty in addressing this problem is that the investment in making deal knowledge accessible has to be made while the firm is also busy doing deals. The moment when a newly closed deal's knowledge is most valuable, most complete, and most easily captured is the closing moment, which is also a moment of high activity and low bandwidth for documentation tasks.
The approach that works in practice is to make the capture task as lightweight as possible: a small set of structured metadata fields at matter opening, confirmed and supplemented at close, combined with automated extraction from the final documents. The manual investment is minimal. The result, accumulated across many deals over time, is a bank that makes the firm's history of deal-handling accessible in ways that form books and full-text search alone cannot achieve.
The compounding value of this investment is significant. Each deal that closes and enters the bank well-structured makes the bank better for every future deal in that category. A firm that begins building in this way today has a materially better knowledge resource in three years than a firm that continues to let deal knowledge become inaccessible at close. The loss is gradual and diffuse enough that it rarely triggers urgency. But it accumulates, and the gap between firms that address it and those that do not will be meaningful as the volume of transactional work in AI-adjacent practice areas continues to grow.