← ALL CASE STUDIES
Case Study · Legal

Institutional knowledge. Now an AI agent.

Every firm has two libraries: the one on the server, and the one in the senior partners’ heads. Only the second one answers questions — and it retires, takes vacations, and bills by the hour. This firm built a third: all of the memory, none of the bottleneck.

24/7Expert-level answers, on tap
40sMedian time to cited answer
100%Answers cite firm sources
Mid-sized law firm
IndustryLegal
Corpus20+ years · matters, memos, templates
Built byPromata — Blueprint → AI Build
StatusLive · answering right now

Does this sound like your firm?

The real knowledge-management system is called “go ask Sarah.”
An associate spent six hours yesterday researching a question the firm answered in 2019.
When a senior partner retires, a library burns down.
The 11 PM filing question waits until the 9 AM partner.
Juniors hesitate to ask — every question spends a partner’s patience.
IF YOU CHECKED THREE OF THESE — THIS CASE STUDY IS ABOUT YOU.
The Challenge

The firm knew the answer. Finding who knew it was the job.

Twenty years of matters had produced positions, precedents, playbooks, and hard-won judgment — scattered across a document server nobody could really search and a handful of heads everybody had to. The knowledge existed. Retrieval was the entire problem.

⚠ ASK-A-PARTNER RETRIEVAL
“Have we handled this before?” — asked TuesdayANSWERED THURSDAY
Six associate-hours re-researching a settled positionBILLED TO WRITE-OFF
Partner on vacationKNOWLEDGE ON VACATION
New associate onboardingSIX MONTHS OF OSMOSIS
Retirement partyLIBRARY FIRE
✦ THE FIRM’S MEMORY, ON TAP
Same question — answered with citations40 SECONDS
Settled positions surface with their sourcesNO RE-RESEARCH
Every lawyer queries the full corpusPERMISSIONS-AWARE
New associates ask freely, learn fastWEEKS, NOT MONTHS
Partners retire · memory staysPRESERVED
VS

THE SAME QUESTION — TWO DAYS OF HALLWAY RETRIEVAL VS. FORTY SECONDS WITH SOURCES

  • Re-research was invisible overhead

    Nobody logs the hours spent rediscovering the firm’s own positions — they hide inside matters as written-off time and slower turnarounds.

  • Access was hierarchical by accident

    The best answers lived behind seniority and availability. The question you could ask depended on who owed you a favor.

  • Every departure was an amputation

    Lateral moves and retirements walked decades of context out the door — with no procedure for keeping any of it.

  • The server was write-only

    Documents went in daily. Answers came out never — folder search finds filenames, not positions.

The Solution

An agent grounded in the firm’s own work — and nothing else.

We indexed twenty years of matters, memos, templates, and playbooks into a permissions-aware internal agent. Ask in plain language; receive the firm’s actual position, with citations to the documents it came from. Not general legal knowledge — this firm’s knowledge.

2,400+MATTERSMEMOS& OPINIONSTEMPLATES& CLAUSESPLAYBOOKSNEGOTIATIONFILINGSRESEARCHFIRMMEMORY
TWENTY YEARS, ONE QUERYABLE MEMORY — EVERY ANSWER TRACES BACK TO ITS SOURCE DOCUMENTS
  • Citations, or nothing

    Every answer names the matters, memos, and clauses it drew from — clickable, checkable. If the corpus does not support an answer, the agent says so and points to the nearest partner instead of guessing.

  • Permissions-aware by construction

    Matter walls and confidentiality screens apply to the agent exactly as they apply to people. Lawyers see only what they are entitled to see.

  • The firm’s voice, not the internet’s

    Grounded exclusively in internal documents. It answers what THIS firm holds, argues, and drafts — which is the only answer an associate at 11 PM actually needs.

  • It learns as the firm works

    New matters and memos index automatically. The memory grows the way the firm does — by practicing.

Results & Business Impact

Twenty years of judgment, forty seconds away.

WHAT CHANGEDBEFORENOW
“Have we handled this before?”2 days, if lucky40 seconds, cited
Precedent & position reuseTribal, partner-dependentUniversal, sourced
New-associate ramp-up~6 months of osmosisWeeks, self-serve
After-hours questionsWait for morningAnswered at 11 PM
Partner departuresKnowledge walks outMemory stays
24/7Expert answers, on tap
40sMedian time to cited answer
100%Answers with firm citations
0Uncited answers permitted
  • Associate hours moved up the value chain

    Time once spent rediscovering settled positions now goes to applying them. The write-off line item quietly shrank.

  • The hierarchy of access flattened

    A first-year queries the same memory as a name partner. Better questions, faster growth, less hallway diplomacy.

  • Institutional memory became institutional

    For the first time, the firm’s knowledge belongs to the firm — not to whoever happens to still work there.

  • Partners got their doorways back

    The interruptions that fragmented every senior lawyer’s day dropped to the questions that genuinely deserve them.

Your firm already paid for this knowledge. Once.

Every position in that corpus was expensive the first time — researched, argued, refined on client matters over two decades. Paying associates to rediscover it is buying the same asset twice. The agent does not create knowledge; it ends the re-purchase.

24/7Answers on tap
40sTo a cited answer
100%Firm-sourced

What we didn't solve

The agent does not create knowledge. If the firm's own corpus does not support an answer it says so and points to the nearest partner instead of guessing — so a question the firm has never answered in writing is still a question for a person.

Whose head is YOUR firm’s knowledge in?

Tell us where your positions, precedents, and playbooks live today. We’ll show you what they look like as an agent — before you spend a dollar.

Book a discovery call

Want results like these?

Tell us where your team is losing hours. We'll show you exactly what automation can do about it — and what it's worth, before you spend a dollar.

30 minutes · No pitch deck · An honest first read

Case study: institutional knowledge, now an AI agent