Skip to content

Articles · Dave Stanton · · 5 min read

Your firm's knowledge needs a librarian

The hard part of putting AI to work in a firm isn't the AI. It's keeping the firm's knowledge organized, current and in the right hands, which is what librarians have done for a century and a half.

When people hear what we do, they expect the hard part to be the AI. It isn't. The assistants are good, and they get better every few months without any help from us. The hard part is the knowledge they answer from: what your firm actually does, written down, organized so it can be found, kept current, and seen only by the people who should see it.

That isn't a new problem. Librarians have been solving it since long before computers. Melvil Dewey published his classification system in 1876. The craft behind it, deciding what goes where, what it's called, what's out of date and who may read it, is older still. Almost everything that makes a firm's AI assistant trustworthy is one of those old library jobs, done for a new kind of reader.

Cataloging: knowing what you have

A library you can't search is a warehouse. The first job is a catalog: every item described, so a reader can find it without knowing where it's shelved.

Most firms have no catalog of how they work. The process for onboarding a client lives in one partner's head, a checklist lives in a folder someone made in 2019, and the right way to handle a lapsed deadline lives in an email thread. A new person can't find what they don't know exists.

So the first thing we build with a firm is a catalog of its work: the situations that come up most, the rare ones that hurt most, and for each one, the firm's answer. That's the guide.

Controlled vocabulary: calling things what people call them

Libraries learned early that readers don't search with the catalog's words. Someone looking for "cars" won't find a shelf labeled "automobiles" unless the catalog knows the two mean the same thing. So librarians keep a controlled vocabulary: one preferred term, with all the other ways people say it pointing to it.

Your firm has its own vocabulary, and your people ask questions in it. A good guide answers "the buyer's lender pulled out" and "financing fell through" with the same answer. We collect the ways people actually ask, from discovery sessions and from the questions the guide couldn't answer, and teach the guide to recognize them.

The reference interview: finding the real question

Reference librarians have a name for what happens at the desk: the reference interview. The question a patron asks is often not the question they have. "Do you have books on Italy?" might mean planning a trip, or a school report on the Renaissance. A good librarian asks a question or two before answering.

An assistant can do a version of this. When a question doesn't clearly match anything in the firm's guide, ours doesn't guess. It says there's no exact match, lists the topics it does cover, and lets the person, or their AI, pick the one that fits. If none does, the question is kept so the firm can answer it next time.

Referral: knowing what's not yours to answer

A reference librarian doesn't give legal or medical advice. They know where their job ends and send people to the right specialist.

A firm needs the same line, drawn on purpose. "How does our firm handle this?" is the guide's to answer. "What does the law require here?" belongs to your broker, your attorney or your compliance lead. Every answer in a guide is labeled with what's at stake and who decides, and questions that need a person go to the person your firm has named.

Weeding: removing what's out of date

Libraries remove books on purpose. They call it weeding, and it's routine: an outdated travel guide or a superseded tax manual on the shelf does more harm than a gap, because people trust what they find there.

A firm's knowledge goes stale the same way. Policies change, a form gets replaced, the way you handle something gets better after a bad week. A guide nobody weeds becomes confidently wrong. Keeping it current is monthly work, and it's most of what the service is: the questions people asked that it couldn't answer, the answers that changed, and a short note to the firm on what's different.

Special collections and reader privacy

Libraries have always had material not everyone may handle: rare books, archives, personal papers. Those live in special collections, and access is granted, not assumed.

A firm's version is its records: clients, contacts, financials, case files. They stay in the systems built to protect them, behind the access the firm already controls, and out of the guide entirely. The guide holds how the firm works, never what it holds.

Librarians also have a long tradition of protecting what readers ask. The American Library Association's code of ethics commits librarians to protect each user's privacy. We hold ourselves to something similar. For questions the guide answered, we keep only which answer was used, not what was asked. Questions it couldn't answer are kept in full, because that's how the guide learns, and your owners and managers can see them.

Why a firm doesn't do this itself

None of this is a secret, and none of it is hard to understand. It's just work: steady, unglamorous, never finished. A firm of thirty or eighty people needs it, and has no one whose job it is. The people who know how the work goes are the ones billing, closing and advising. Hiring a knowledge manager is a big step for a firm that size, and it's hard to know what to ask of one.

That's the job we take on. We do the cataloging, the vocabulary, the referrals and the weeding, every month, for a flat fee: our Operating plan is $599 a month. Your people get an assistant that answers the way your firm works, from the Claude or ChatGPT account they already have or from our iPhone app. You get a librarian for your firm's knowledge, without hiring one.

Questions about your own setup?

Tell me how your people use AI today. I'll tell you what I'd change, whether or not we end up working together.