A partner at a mid-sized litigation firm asked me last month why their new AI tool kept citing the wrong precedent. They’d spent $40,000 on a legal research platform that promised to cut associate hours in half. Instead, it kept pulling cases from the wrong jurisdiction, missing critical distinctions, and occasionally inventing citations that didn’t exist.

The problem wasn’t the model. It was the context.

A recent VentureBeat study found that 57% of enterprises traced incorrect AI outputs directly to missing business context. Not hallucinations. Not bad training data. Missing context about how the business actually works, what matters to clients, and which details change everything.

For law firms, this gap shows up in billable hours that vanish, intake calls that never convert, and document review that takes three times longer than it should. The average firm in the $1M-$25M range leaks $80,000 to $250,000 annually because their AI doesn’t know what their people know.

The Context Problem in Legal Work

Most legal AI tools train on public case law, statutes, and regulatory filings. That’s table stakes. What they don’t have is your firm’s institutional knowledge.

They don’t know that your real estate practice always flags force majeure clauses in commercial leases because three clients got burned during COVID. They don’t know that Judge Martinez in the Eastern District hates boilerplate motions and will deny them on sight. They don’t know that your employment clients care more about reputational risk than settlement cost.

This missing context creates three expensive problems.

First, billable-hour leakage. An associate spends 90 minutes reviewing a contract, then another hour correcting the AI’s first pass because it missed clauses your firm always negotiates. That second hour rarely makes it onto the invoice. Across a ten-attorney firm, we typically see 4-6 hours per week per attorney of unbilled time tied to rework. At $300 per hour, that’s $60,000 to $90,000 annually per attorney.

Second, intake delays. A potential client calls at 6pm on a Thursday. No one picks up. They leave a voicemail. By the time someone calls back Friday afternoon, they’ve already hired the firm that answered. High-intent calls convert at 30-40% lower rates when the response window stretches past four hours. For firms that generate 40% of new matters from inbound calls, that delay costs six figures.

Third, document review bottlenecks. Discovery in a mid-sized commercial dispute might involve 15,000 pages. A junior associate bills 60 hours at $250 per hour for first-pass review. The client pays $15,000 for work that’s mostly pattern recognition. If the associate misses a key email thread because they don’t know the client’s internal terminology, the partner spends another ten hours fixing it.

All three problems share the same root cause. The AI doesn’t know your practice.

What Structured Context Looks Like

Before you deploy any AI in legal work, you need a knowledge base that captures how your firm thinks. Not a document repository. A structured map of practice areas, precedents, client matter details, and decision rules.

Here’s what that looks like in practice.

Start with practice area taxonomies. Your intake form might list “employment law” as a single option. But your firm handles wrongful termination, wage-and-hour disputes, non-compete enforcement, and EEOC defense. Each has different economics, different timelines, and different partner expertise. Your AI needs to know that a caller asking about “getting fired” might mean wrongful termination (high value, long cycle) or at-will employment advice (quick consult, flat fee). That distinction changes how you route the call and what you quote.

Next, precedent libraries. Not just case citations. Annotated briefs, motion templates, and negotiation playbooks that explain why your firm takes a certain position. When your Document Review Agent scans a commercial lease, it should flag the indemnification clause not because it’s boilerplate, but because your real estate partner always rewrites it to cap liability at 12 months of rent. That’s institutional knowledge, and it lives in the notes your senior associates leave on closed matters.

Then, client matter details. Your corporate clients have internal jargon. One calls their sales agreements “SOWs.” Another calls them “engagement letters.” A third calls them “project authorizations.” They’re the same document, but if your AI doesn’t know the synonym map, it’ll miss half the relevant contracts during discovery. We see this constantly in M&A due diligence, where the target company’s document naming conventions don’t match the acquirer’s.

Finally, decision rules. Your firm probably has a conflict-check protocol, a matter-acceptance threshold, and a fee structure that varies by practice area. Those rules need to live in a format your AI can query. When your Intake Voice Agent answers a call, it should know that personal injury cases under $50,000 go to a referral partner, employment cases require a $5,000 retainer, and any caller who mentions opposing counsel by name triggers an immediate conflict check.

This isn’t a six-month IT project. It’s a structured interview process. You sit down with each practice group, walk through the last ten matters, and document the decisions that didn’t make it into the case file. What did you look for first? What always gets negotiated? What’s a red flag? What’s a time-waster?

One litigation partner told me his team always checks whether the opposing party is represented before drafting a demand letter. Obvious to him. Not obvious to an AI that’s never seen a bar complaint for improper contact. Now his Matter Triage Agent flags unrepresented parties and routes those matters to a senior associate for review.

Building Agents That Use Your Context

Once you have structured context, you can build agents that work the way your firm works. Not generic legal AI. Agents trained on your precedents, your client base, and your decision rules.

Our Intake Voice Agent is a good example. It answers every call, 24/7. When someone calls asking about a “non-compete issue,” it doesn’t just log the inquiry. It asks whether they signed the agreement, whether they’ve started a new job, and which state they’re in. Then it runs a conflict check against your client list, scores the matter for fit, and books a 30-minute consult with the right partner. The caller gets a calendar invite before they hang up.

That agent knows your firm’s context. It knows you don’t take California non-compete cases because they’re unenforceable. It knows your employment partner charges $400 per hour but offers a flat-fee review for agreements under five pages. It knows that if the caller mentions a current client by name, the call routes to the managing partner for a conflict review.

Compare that to a generic answering service. They take a message. You call back. You ask the same questions. You check conflicts manually. You send a retainer agreement. The caller signs it, or doesn’t. Three days have passed, and you’ve spent an hour on intake work that doesn’t bill.

The Intake Voice Agent collapses that cycle into one call. Firms that deploy it see after-hours conversion rates jump from 15-20% to 50-60%, because the friction disappears. The economics are straightforward. If you close 40 new matters per year and 30% come from after-hours calls, improving that conversion by 30 percentage points adds twelve matters. At an average matter value of $8,000, that’s $96,000 in new revenue.

Our Matter Triage Agent works the same way. It monitors your intake email and web forms, classifies each inquiry by practice area, scores it for fit, and routes it to the right partner with a one-paragraph brief attached. The brief includes the conflict-check result, the estimated matter value, and any red flags from your decision rules.

One family law firm we work with uses the Matter Triage Agent to separate high-net-worth divorce inquiries from routine custody modifications. High-net-worth cases go to the senior partner with a summary of assets mentioned in the intake form. Custody modifications go to an associate with a checklist of required documents. The partner doesn’t waste time on $3,000 matters, and the associate doesn’t get overwhelmed by $200,000 cases they’re not ready to handle.

The Document Review Agent is where context matters most. It performs first-pass review on contracts, discovery batches, and matter files. But it doesn’t just highlight every indemnification clause. It flags the ones that deviate from your firm’s standard position. It doesn’t summarize every email. It pulls the threads that mention your client’s key decision-makers or the opposing party’s litigation history.

A commercial litigation partner described it this way: “I used to spend three hours reviewing an associate’s discovery memo, because I had to check whether they caught the stuff that matters to us. Now the agent produces the memo, and I spend 30 minutes confirming it’s right. The agent knows what I care about.”

That’s the difference between generic AI and an agent built on your context.

The Omni Audit: Where Context Starts

Most firms don’t have a structured knowledge base yet. They have SharePoint folders, practice management software, and a lot of institutional knowledge that lives in partners’ heads. That’s fine. You don’t need to solve that problem before you start.

We run a 60-minute Omni Audit that maps your current state and identifies the highest-value agent to build first. No deck. No sales pitch. Three outputs: a process map of your intake and matter workflow, a leakage estimate tied to your billable rates, and a one-page build plan for the agent that saves you the most money.

Book a 60-min Omni Audit and we’ll walk through your last ten matters. We’ll ask what you looked for, what took longer than it should have, and where your team spent time on work that didn’t bill. Then we’ll show you what an agent handling that work looks like, and what it costs to build.

For most law firms, the first agent is either intake or document review. Intake because it’s the fastest path to new revenue. Document review because it’s the biggest cost center. Both depend on structured context, and both pay for themselves in 90 days.

If you want a head start, we’ve built a worksheet that walks through the context questions your Intake Voice Agent will need answered. Download the AI Client Intake Checklist and use it to document your conflict-check protocol, your practice area routing rules, and your retainer structure. It’s the same checklist we use in the first 20 minutes of an audit.

Why Open-Source Context Beats Proprietary Platforms

The VentureBeat study highlighted another finding. Enterprises that built portable, open-source semantic layers for their business context saw 40% fewer AI errors than those relying on proprietary metadata locked inside a single platform.

For law firms, that means don’t trap your institutional knowledge in a vendor’s system. If you spend six months training a legal AI tool on your precedents and client matter details, and then the vendor raises prices or shuts down, you’ve lost that investment.

We build agents on Omni, which uses open-source semantic code. Your context lives in a format you own. If you decide to switch tools, your knowledge base moves with you. If you want to connect a new AI model, you plug it into the same semantic layer. You’re not locked in.

One estate planning firm told me they’d spent $60,000 customizing a document automation platform, only to discover the vendor didn’t support their state’s new trust rules. They had to rebuild everything from scratch. With Omni, the context layer is separate from the execution layer. When the rules change, you update the semantic map. The agents adapt automatically.

This matters more as your firm grows. A ten-attorney firm might use one practice management system, one research tool, and one intake platform. A 50-attorney firm uses six. If your context is locked in system one, systems two through six can’t use it. You end up with six different AI tools that don’t talk to each other, and partners who don’t trust any of them.

Portable context solves that. Your Intake Voice Agent, your Matter Triage Agent, and your Document Review Agent all query the same knowledge base. When you update a decision rule, all three agents see the change. When you add a new practice area, all three start routing it correctly.

That’s how you scale AI across a firm without creating a maintenance nightmare.

What This Looks Like in Practice

A business litigation firm in the $5M range came to us with a document review problem. Their associates were spending 40 hours per matter on first-pass contract review in discovery. The client paid for it, but the firm couldn’t scale. They had four associates and a pipeline of eight new matters per quarter. The math didn’t work.

We ran the AI audit for law firms and mapped their contract review process. Turns out, 60% of the associate’s time went to finding and flagging three specific clause types: termination rights, liability caps, and payment terms. The other 40% was reading context to understand whether the clause was standard or unusual.

We built a Document Review Agent trained on their last 200 contracts. It knew what “standard” looked like for their client base. It knew that termination-for-convenience clauses were red flags in government contracts but normal in commercial agreements. It knew that net-60 payment terms were a problem for their manufacturing clients but fine for their retail clients.

The agent cut first-pass review time from 40 hours to 12 hours. The associate still did the final review, but they started with a memo that flagged the eight contracts that mattered and summarized the key terms. The firm went from four matters per associate per quarter to seven. Revenue per associate jumped 75%.

That’s the return on structured context. You don’t replace your people. You give them tools that know what they know.

The Next Step

If your firm is leaking billable hours to rework, losing intake calls to slow response times, or bottlenecking on document review, the problem isn’t your people. It’s missing context.

Your AI doesn’t know your practice. It doesn’t know your clients. It doesn’t know the decisions that separate a $10,000 matter from a $100,000 matter.

You can fix that in 60 minutes. Book my Omni Audit and we’ll map your intake and matter workflow, estimate your leakage, and show you what an agent built on your context looks like. No deck. No sales pitch. Three outputs you can use whether you work with us or not.

Most firms start with intake or document review. Both pay for themselves in 90 days. Both depend on structured context. And both scale as your firm grows.

See Omni for law firms and decide if it’s worth an hour.

For more on how AI agents integrate into professional services workflows, explore our guides or read through recent case studies on the EDNA blog. If you’re earlier in your AI journey, our learning resources cover the fundamentals of agent design and semantic context layers.

The firms that win with AI won’t be the ones with the biggest models. They’ll be the ones with the best context.