The matter file lives in six places
Email, the DMS, the practice management system, a local folder, a shared folder, and someone's head. Ask where a document is and you'll get a guess, not an answer.
The short version
4 things that decide this
- 01A matter's real record is scattered across email, your DMS (NetDocuments, iManage or a shared drive), your practice management system, a local folder, a shared folder, and whichever associate or partner happened to touch it last.
- 02Nobody finds anything twice, a new associate re-learns the firm's positions from a blank start, and the knowledge walks out the door with any partner who leaves.
- 03"Just put it all in the DMS" almost never happens, because migrating years of matters is its own project and everyone reverts to habit within a month.
- 04An index over what you already have cites the source with a link, respects who's allowed to see the matter, keeps an audit trail, and never moves a file. Stanford's 2024 research is the reason verification stays a separate, deterministic step, not a model behavior.
Six places, and none of them is the whole story
Ask your paralegal where a specific document from a matter lives, and watch what happens. They'll check email first, because that's where most correspondence with opposing counsel actually happened. Next comes the DMS, whether that's NetDocuments, iManage or a shared drive with a folder structure someone set up years ago. After that, they'll try your practice management system, Clio, MyCase, PracticePanther, Smokeball or whatever you run, since the matter record technically lives there, even though hardly anyone treats it as the real source for documents.
If none of those turn it up, there's a local folder on somebody's laptop. Or a shared folder your practice group set up on its own. And, more often than you'd like to admit, a partner's memory of roughly what happened and roughly where it went. Six places. Ask where a document is, and the honest answer is usually a guess, followed by three separate searches.
Nobody finds anything twice
You'd think finding a document once would make finding it a second time easy. It doesn't work that way. Somebody on your team finds the right precedent, the right clause, the right position your firm took on a similar issue three years ago. That knowledge then lives in their head, or in one email thread, until the next person who needs it starts the search from zero.
A new associate feels this hardest. They join your firm, which has already decided how it handles a given clause, a given fact pattern, a given discovery motion. None of that is written down anywhere searchable. So they re-derive it from scratch, sometimes taking a different position than your firm's established one, purely because nobody could point them to the file where the last version lived. The same gap shows up when an engagement letter references a position the firm took years ago, and nobody can find the memo that backs it up.
Then there's the cost that keeps managing partners up at night. A partner who's been at the firm fifteen years genuinely holds the map: where everything is, why decisions got made, which position you took and why. When they retire or leave for another firm, whatever wasn't written down goes with them. It's rarely written down, because writing it down was never anyone's job.
"Just put it all in the DMS" never actually happens
You've had this conversation before. Someone proposes consolidating everything into the DMS, once and for all, so there's a single place to look. It sounds simple in a partner meeting. Then you scope it, and it's a multi-month migration project: years of email threads to sort, folder structures from three eras of your firm to reconcile, and a practice management system whose document store doesn't map cleanly onto the DMS's.
Even if you push through and finish the migration, old habits reassert themselves within weeks. Someone attaches a redline to an email instead of checking it into the DMS, because it's faster in the moment. Your six-places problem quietly starts rebuilding itself. Consolidation was never really the fix. It just moved the mess and gave it a head start before it came back.
Index what you already have. Don't move it
A migration isn't the fix that actually holds up. What works is an index that reads email, the DMS, your practice management system and your shared folders where they already sit. Ask it a question, and it points you at the real document, with a link. Nothing moves. Your firm doesn't have to change how it works.
It has to respect who's allowed to see what. Your ethical walls exist for a reason, and an index that ignores them turns a knowledge tool into a confidentiality problem. So access is scoped per user, at the moment the query runs, not filtered after the fact. That matters as much for privileged material and work product as it does for an ordinary matter file. And there's an audit trail of who asked what and when, the same way you'd want a record of who pulled a file from a locked cabinet.
Your files stay exactly where they are. Nothing gets copied into a separate system that someone then has to maintain and secure on top of everything else. Your DMS stays your DMS, your email stays your email, and the index sits over the top, reading and citing rather than holding a second copy of your firm's confidential material.
- 01Cites the source with a link, not a paraphrase.
- 02Scoped to who may see the matter, enforced when the query runs.
- 03Read-only. Files stay where they are; nothing gets moved or duplicated.
What the index reads, without moving any of it
Six sources, one place to ask a question about any of them.
| The source | What it holds | What the index does with it |
|---|---|---|
| Correspondence with opposing counsel and the client, often the real record of what happened | Read and cited, scoped to the matter and the mailbox owner's permissions | |
| The DMS | NetDocuments, iManage or a shared drive with a folder structure | Indexed in place, inheriting the DMS's own ethical-wall rules |
| The practice management system | The matter record, notes and sometimes documents | Read for matter context and linked back to the source record |
| Local and shared folders | Working drafts, one-off practice-group setups, whatever escaped the DMS | Indexed where accessible; a real answer to what's actually there instead of what should be |
A citation still has to be checked, every time
Stanford's RegLab and HAI ran a benchmark in May 2024 against purpose-built legal research tools, not general chatbots. Their results are why verification can't be left to the model. Lexis+ AI and Ask Practical Law AI hallucinated on more than 17% of queries. Westlaw AI-Assisted Research hallucinated on more than 34%. A general-purpose model with no legal grounding hallucinated on 58% to 82% of legal queries in the same study.
Those are tools built specifically for legal research, tested against real queries. Their error rate still lands well above what your firm would accept from an associate's memo. That's not an argument against indexing your matter file. It's the argument for treating citation checking as its own step, separate from whatever generated the answer, rather than trusting a confident answer just because it cited something.
An answer from the index is never the last word. It points your lawyer at the source, and your lawyer decides what that source actually means for the matter in front of them. That's not a limitation to apologize for. It's the same split as an intake line that screens a call but never decides whether to take the case: the system finds and cites, a lawyer judges.
- QueryScoped to who's allowed to see the matter
- RetrievalAcross email, DMS, PMS and shared folders, in place
- CitationEvery answer links back to the real document
- VerificationA separate, deterministic check against the source, not a model guess
- JudgmentA lawyer decides what the answer means for this matter
Retrieval and citation are mechanical. Verification is checked, not assumed. What the answer means is a lawyer's call, every time.
Where the matter data actually travels
An index that reads across your systems is only as trustworthy as its architecture. You should be able to get plain answers before anyone touches a live matter. Access is scoped to the matter, not your firm as a whole, so a query can't surface something the requesting lawyer isn't entitled to see. ABA Opinion 512 makes vetting a vendor's security practices part of your own duty, and United States v. Heppner is the reminder that a tool's terms decide whether privilege survives contact with it, more than its encryption ever will. No-training terms with the model provider are in writing and retrievable if a client or a court ever asks. Every vendor that touches your data, down to the transcription or logging layer, gets named as a recipient rather than left as an assumption. Private deployment is available where a client's own obligations require it.
- 01Access scoped to the matter, enforced at query time, not filtered after retrieval.
- 02Written no-training terms with the model provider, retrievable on request.
- 03Every vendor named as a recipient; private deployment where required.
Some of the systems we have shipped
Related questions
01Do we have to migrate everything into one system for this to work?+
No. Our index reads email, the DMS, your practice management system and your shared folders where they already sit. That's the point: your firm doesn't have to finish a consolidation project first, and most never do.
02How does it handle ethical walls between practice groups?+
Access is scoped per user at the moment a query runs, not filtered out of results after the fact. If a lawyer isn't entitled to see a matter, the index doesn't surface it, the same way your DMS already restricts who can open a given folder.
03Can we trust an answer without checking it ourselves?+
No, and the system isn't built to ask you to. Every answer cites its source with a link, and citation verification runs as a separate, deterministic check rather than something the model asserts about itself. A lawyer still reads the source and decides what it means.
04What if our documents live in a messy mix of folders with no consistent structure?+
That's the common starting point, not an exception. Indexing works with what's actually there rather than requiring a clean structure first, and we confirm exactly what your systems expose during the audit.
