By: Eric Sedwick
Artificial intelligence is everywhere in legal right now: drafting, summarizing, research, reviewing. But when it comes to broad knowledge management processes and outcomes, many organizations are operating under the assumption:
“If we just feed all of our documents to AI, it will figure out the rest.”
That assumption overlooks something important. AI is powerful, but it is not clairvoyant. AI can analyze patterns and generate responses with remarkable speed. But responsibility for context and priority still rests with people. Even advanced systems may elevate the wrong output if guidance and labeling are unclear. For that reason, human involvement is required at both ends of the loop: at the beginning to structure and signal what matters, and at the end to review and validate the results.
Putting a human at the beginning of the loop means asking knowledge owners to ensure the information AI can access is worth seeing. With the Records and Knowledge Management teams I've overseen in my career, driving change, including appreciation of value and willingness to invest, has always been a struggle. Asking colleagues to update their storage practices to fit new corporate policies with legal or regulatory, but questionable individual benefit, can be a tough sell.
Before AI, There Is Knowledge
Attorneys rely on institutional knowledge to do their best work. Long before AI entered the conversation, legal practice was built on experience, appreciating what has been done before, what worked, what failed, and why.
That institutional knowledge takes many forms:
- Prior work product
- Previously negotiated positions
- Research memos
- Briefs and arguments that worked (or failed)
- Templates refined through experience
Knowledge is the accumulated judgment of the organization. It’s what allows lawyers to move faster, avoid repeating mistakes, and deliver consistent results.
The promise of AI is compelling because it suggests that institutional knowledge can be surfaced instantly. No more digging through folders, email chains, or shared drives. Ask a question, and the system retrieves the right precedent, the best clause, the strongest argument.
The reality is that it can only surface what it can recognize as knowledge. If valuable work product sits buried among outdated drafts, conflicting documents, or irrelevant files, the system may struggle to understand which materials represent knowledge and which are just noise. AI needs structure and guardrails.
What Is Legal Knowledge?
Before we can talk about managing knowledge, we have to define it.
Legal knowledge, at least the knowledge worth continuing to discuss, is prior work product or research that can support current or future legal work. It is information with practical, reusable value.
If it helps a lawyer move faster, argue more effectively, draft with greater confidence, or avoid risk the next time a similar issue arises, it’s knowledge. Not everything stored and available to AI meets that standard.
The Garage Problem: Why AI Can’t Fix a Mess without Direction
Imagine your garage.
Over the years, you’ve thrown everything in there: old furniture, power tools, holiday decorations, tax records, rare collectibles, boxes of clothes to donate.
Now imagine asking someone to find “the most valuable things” in that garage without labels, organization, or context.
That’s what many have been asking AI to do when they point it at every document, email, shared drive, matter folder, and archived record in their various repositories. Without direction, AI can only guess which materials represent reusable work product and which are unfinished drafts. It may mistake outdated analysis from current law, or proven strategy from an abandoned approach.
Unless you help it. With structure, curation, and context, AI can better distinguish knowledge from “garbage.”
Knowledge vs. Information vs. Documents vs. Records
Legal leaders often don’t distinguish between knowledge management, document management, and records management. To the business, it can all feel like “information.” What they care about is whether, when the moment comes, it can be found, trusted, and used effectively.
Key Components of Document, Records, Knowledge Management
Information type | Primary Purpose | Core Focus |
|---|
Document management | Store and organize | Version control, access, collaboration |
Records management | Compliance and defensible retention | Regulatory requirements, retention schedules, deletion |
Knowledge Management | Enable resuse of valuable work | Curation, classification, promotion of high-value work |
You do not need to boil the ocean to structure your AI environment. The effort is real, but it can be focused. Start with your core knowledge assets: find them, label them, and make them accessible. When that foundation is in place, the system can surface trusted knowledge when it is needed, and additional refinement can follow in phases.
General vs. Customized Knowledge Needs
Knowledge management often requires segmentation by type and audience. AI cannot infer organizational nuance unless you design for it. At a high level, every organization shares certain universal knowledge needs.
General Knowledge Needs (Across All Users)
These are foundational capabilities such as search, availability, and secure access. They include shared materials that provide consistent direction across the department, for example: standard contracting positions, approved clause libraries, company risk tolerance guidelines, and internal process playbooks. Everyone benefits from being able to quickly find what exists, understand whether it is current, and apply it.
Customized Knowledge Needs (By Function and Role)
Legal knowledge is highly contextual. What is valuable to one team may be irrelevant to another. Consider the differences:
- Litigation teams need prior briefs, motions, deposition outlines, and judicial tendencies.
- Commercial teams need negotiated clauses, fallback language, and playbooks tied to risk tolerance.
- Regulatory teams need legislative histories, agency correspondence, and position memos.
- Employment teams need investigation templates, policy language, and prior case outcomes.
If everything is thrown into one searchable mass, AI may surface something technically similar but strategically wrong. AI does not inherently understand practice nuance or business context unless you design your knowledge systems to reflect those realities.
Why Segmentation Matters for AI
When knowledge is intentionally segmented, AI performs differently. It can prioritize the right category of material, draw from curated and approved content, and surface insights that are specific to a practice area or matter type. The result is less noise and greater confidence in what’s returned.
Without segmentation, the opposite happens. AI may struggle to distinguish between precedent and draft, current guidance and outdated analysis, without proper file naming conventions or document headers. It may surface volume instead of value. Users spend more time validating what they see than acting on it, and institutional knowledge becomes diluted.
How to Manage Knowledge (So AI Is Set Up for Success)
If AI is going to unlock real value, knowledge must first be intentionally managed.
1. Identify
Not everything in your document management system is knowledge. Some documents are outdated. Some were never finalized. Some reflect positions you would not take again.
Legal teams do not need to manually review every document ever created. But they do need to define what qualifies as reusable work product and focus their effort there.
Start with high-impact categories: standard commercial clauses, frequently used motions, regulatory interpretation memos, high-value templates. Curate those intentionally. Confirm which versions are approved. Add clear labels. Tag them by jurisdiction, practice area, risk profile, or business unit. Summarize them so they can be understood quickly.
AI can assist in this process by surfacing likely candidates for reuse, clustering similar documents, identifying outdated language, or drafting summaries. But human judgment must set the standard for what is authoritative and current.
2. Promote
If people do not know knowledge exists, they can’t use it. High-value content should be embedded where work happens. Promotion can take many forms:
- Surfacing curated materials within workflows
- Sharing successful strategies after matters conclude
- Embedding precedent into drafting tools
- Using AI to highlight approved resources
Without regular maintenance, curated knowledge gradually loses reliability. Approved templates should be kept current. High-value work product should be reviewed periodically to confirm it still reflects current law and risk tolerance. When regulations change, updates should flow through the systems and tools people actually use.
Summary
AI can’t fix your knowledge management problems unless you help it. Like any human or technology partner learning to consistently deliver to your team, providing a playbook in advance and correction when mistakes are made should be an assumed investment.
For legal ops, the work starts with deciding what should be reused, organizing it clearly, and maintaining it over time, with AI assisting but not replacing human judgment. That effort shapes how confidently lawyers can rely on what AI surfaces, how quickly they can move, and how consistently they apply proper judgment.
About the author
Eric Sedwick
Senior Director, Advisory Services
Onit
Eric Sedwick is a seasoned legal operations leader with more than 20 years of experience driving team and process value. He serves as Senior Director of Advisory Services at Onit and previously held senior leadership roles at TIAA, including Head of Legal Operations.
