Know-how banks and AI: turning past work into usable knowledge
By the ChamberOne team 5 minute read
Every firm already owns a know-how bank. It is just badly organised. It lives in the heads of senior lawyers, in old precedent folders, in email threads nobody can find and in a handful of notes saved under names like 'final v3'. When a partner retires or an associate moves on, a slice of it goes with them. AI does not create knowledge, but it can make a collection of written notes far easier to search. The hard part is still getting good notes written in the first place.
Key points
- Capture lessons at matter close, with a short prompt, while the file is still open.
- Remove client and counterparty detail before an entry enters the shared bank.
- AI search finds notes by meaning, but it can only be as good as the notes behind it.
- Give every entry an author, a date and a last-checked mark, and give each area an owner.
What a know-how bank actually holds
People use the phrase loosely, so it helps to be specific. A useful bank holds four kinds of material, and each needs a different habit to keep it alive.
- Precedents and templates: the documents you start from, with a short note on when to use each one.
- Standard positions: what the firm normally accepts, resists or trades on a given clause, and why.
- Lessons from matters: what went well, what caught you out, what you would do differently next time.
- Research notes: the answer to a question that took a day to work out, together with where the answer came from and when it was checked.
Capture at matter close, not in a quiet week
Firms that try to build a bank as a separate project usually stall. Nobody has a quiet week. The better moment is matter close, when the facts are fresh and the file is open in front of you anyway.
Keep the prompt short enough that a busy associate will actually answer it. Five questions are plenty. What was the matter about, in two sentences? Which document or clause took the most effort? Did any position we normally take not work here, and why? Is there a note, checklist or template worth keeping? Who in the firm knows most about this topic now?
Notice what this avoids. It does not ask people to write an essay, and it does not ask them to file client documents into a public area. It asks for the lesson, not the file.
Strip out client detail before anything is shared
A know-how entry is read by lawyers who may not be on the matter. Before an entry goes into the shared bank, remove names, figures and anything that would identify the client or the counterparty. Describe the situation in general terms: 'a distributor resisted an exclusivity clause because of an existing supply agreement' is more useful and safer than the names.
If your firm uses confidentiality walls or restricted matters, the bank needs the same care. A note drawn from a restricted matter can leak what the restriction was meant to protect. Decide who reviews entries from sensitive matters, and let that person say no.
- 1Matter closes
- 2Answer the five questions
- 3Remove client detail
- 4Sensitive entries get a reviewer
- 5Entry joins the bank
- Then back to step 1.
Where AI search helps
Once notes exist, the usual complaint is that nobody can find them. Keyword search fails when the person searching uses different words from the person who wrote the note. A lawyer looks for 'early termination fee' and the note says 'break costs'. AI-assisted search can match on meaning rather than exact words, and it can summarise several notes into a short answer with links back to each one.
That is a real gain, and it is a modest one. The tool is only as good as the entries behind it. A bank of thin, outdated or contradictory notes will produce a confident summary of thin, outdated and contradictory notes.
Rules to keep the bank trustworthy
Treat an AI answer from the bank the way you would treat a colleague's recollection. It is a starting point, and the next step is to open the source note.
- Every entry carries an author, a date and a practice area, so readers can judge age and relevance.
- Every answer links to the underlying entries. If a tool cannot show where an answer came from, do not rely on it.
- Entries about the law carry a 'last checked' date. Law changes, and a note that was right two years ago may now be wrong.
- A named knowledge lawyer or partner owns each practice area and prunes it twice a year.
- Entries that are superseded are marked, not silently deleted, so that people see why the position changed.
A small firm's version
Suppose a four-lawyer firm in a mid-sized city handles employment, property and general commercial work. It has no knowledge lawyer and no budget for one. A workable version is a shared folder with one document per practice area, a five-line template for matter close, and a monthly half-hour where the partners read the new entries over coffee.
After a year, the firm has maybe sixty short entries. That is enough to be useful, and enough for AI search to be worth trying. A small firm that starts with sixty good notes will do better than a large one that starts with six thousand unreviewed documents.
What AI does not do for a know-how bank
- It does not decide which lessons are worth keeping. A person has to do that.
- It does not check that a note about the law is still correct.
- It does not remove client confidentiality problems from the notes. Review before sharing is a human step.
- It does not replace conversation. Some of the best know-how is a five-minute chat with the partner who did the last one.
Be honest with the team about this. A know-how bank built on the belief that the software will do the thinking tends to fill up with unreviewed text, and people stop trusting it within a few months.
- Finds a note even when the wording differs, such as 'break costs'
- Makes a collection of written notes easier to search
- Decide which lessons are worth keeping
- Check that a note about the law is still correct
- Review entries for client confidentiality before sharing
- Hold the five-minute chats with the partner who did the last one
Getting started this month
Pick one practice area. Write the five-question close-out prompt. Ask each lawyer to complete it for the next three matters they close. Have one partner read every entry for two months and give feedback. Only then decide whether search tooling is worth the effort. By that point you will know whether people are writing useful notes, which is the part that matters.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
How to check AI legal research before you rely on it
A routine for tracing every AI-suggested authority to its source, confirming it is still good law and testing the proposition it is cited for.
5 minute readGeneral chatbots versus legal research tools: what differs and why it matters
How general chatbots and legal research tools differ on sourcing, links, jurisdiction coverage and currency, and when a general model will do.
5 minute readRetrieval-augmented generation explained for lawyers
A plain explanation of how AI answers from a document set, why it reduces but does not remove fabricated content, and what to ask vendors.
5 minute read