AI for law firms
AI for law firms
Intake, precedent research, and document prep without the grind. 5x ROI in 30 days, or we work free.
- Litigation boutiques
- Full-service firms
- Insolvency practices
- Immigration firms
- Personal injury
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Assembly bills like law until the client refuses to pay for it.
Standard documents drafted from scratch, every matter, and sophisticated clients now read the bill line by line and push back on hours that look like assembly. The first draft should arrive for review, not get typed.
The firm has answered this before. Nobody can find where.
The closest precedent is in a closed matter from 2021, in a folder named after a partner who left. Finding it takes longer than answering again, so the firm pays twice for the same thinking.
Intake leaks matters.
Slow conflict checks and manual follow-up lose clients you already won. The retainer that takes four days to open goes to the firm that opened it in one.
Fixed fees turned assembly hours into cost.
On a flat-fee matter every hour of drafting, formatting and chasing comes out of profit instead of onto an invoice. The hours did not change. Who pays for them did.
Undertakings run on memory.
The chase for missing documents and outstanding undertakings lives in someone's head and an inbox. When that person is in discoveries all week, the file goes quiet and the client notices.
Your lawyers already use AI. The firm just does not govern it.
Over 90 percent of legal professionals use at least one AI tool while firm deployment sits far below that (Wolters Kluwer, 2026). The gap is privileged material in personal accounts.
Law firms, before and after
The manual path is dashed: Drafts started from scratch, Precedent hunts eat hours, Intake follow-up slips. The system path replaces it, and a person approves before anything ships: First drafts arrive for review, Firm knowledge found in minutes, Intake and conflicts cleared.
Before: by hand
- 01Drafts started from scratchhuman
- 02Precedent hunts eat hourshuman
- 03Intake follow-up slipshuman
After: the system
- 01First drafts arrive for review
- 02Firm knowledge found in minutes
- 03Your approvalhuman
- 04Intake and conflicts cleared
The research
Law-firm AI adoption tripled from 11% to 30% in one year (ABA, 2026).
American Bar Association · 2026
Proof, honestly
We have not published a law firm case study yet, and we say so. Founding clients get priority scheduling, direct founder involvement, and a co-published case study once the numbers are real. The 5x ROI guarantee applies from day one.
We guarantee 5x ROI inside 30 days of deployment, in writing, measured against a baseline you sign before we build. If the system misses the bar, we keep working for free until it clears.
The numbers in law firms
Law-firm AI adoption tripled from 11% to 30% in one year (ABA, 2026).
American Bar Association · 2026
Nearly 60% of in-house counsel see no noticeable saving from their outside counsel's AI. Of the 40% who do, only 13% point to fewer billable hours (ACC and Everlaw, 2025).
Association of Corporate Counsel and Everlaw · 2025
Over 90% of legal professionals use at least one AI tool, while firm-level deployment sits far below that (Wolters Kluwer, 2026).
Wolters Kluwer · 2026
Almost half of employees admit using AI in ways that violate company policy (KPMG, 2025).
KPMG / Melbourne Business School · 2025 · 48,000 people, 47 countries
Built around your rules
| Regime | What it demands here | How the system complies |
|---|---|---|
| Solicitor-client privilege | Privileged material must never persist outside firm control and must never train a model. | Documents live in the firm-controlled store, model calls run under no-training, no-retention terms, and only the draft comes back across the line. |
| Law Society of BC, Code of Professional Conduct | The lawyer remains responsible for every work product, and client confidentiality survives every tool choice. | Every agent output is a draft with a named reviewer. Nothing files, sends or advises. Confidentiality is enforced in the database, not in a policy memo. |
| LSBC guidance on generative AI | Due diligence on cloud and AI providers before client data touches them. | Builds are designed around the guidance, including its cloud due-diligence checklist, with provider terms settled in writing before the first document moves. |
| Court practice directions on AI | Several Canadian courts, the Federal Court among them, require disclosure and human ownership of AI-assisted filings, and the rules differ by forum. | The court document agent produces drafts a lawyer reviews, owns and files, and we help the firm write the disclosure policy so nobody improvises at the registry counter. |
| Law Society trust accounting rules | Trust money moves only under human authority. | Agents never write to the trust ledger. They prepare reconciliation and reminders around it, and signoff stays with people. |
| BC PIPA and PIPEDA | Client personal information handled lawfully, with residency answered before anything is built. | Residency is scoped per engagement and stated in writing, and where Canadian residency is required the architecture is designed for it first. |
The regimes that govern law firms, what each demands, and how the system complies
Regime
- Solicitor-client privilege
What it demands here
Privileged material must never persist outside firm control and must never train a model.
How the system complies
Documents live in the firm-controlled store, model calls run under no-training, no-retention terms, and only the draft comes back across the line.
- Law Society of BC, Code of Professional Conduct
What it demands here
The lawyer remains responsible for every work product, and client confidentiality survives every tool choice.
How the system complies
Every agent output is a draft with a named reviewer. Nothing files, sends or advises. Confidentiality is enforced in the database, not in a policy memo.
- LSBC guidance on generative AI
What it demands here
Due diligence on cloud and AI providers before client data touches them.
How the system complies
Builds are designed around the guidance, including its cloud due-diligence checklist, with provider terms settled in writing before the first document moves.
- Court practice directions on AI
What it demands here
Several Canadian courts, the Federal Court among them, require disclosure and human ownership of AI-assisted filings, and the rules differ by forum.
How the system complies
The court document agent produces drafts a lawyer reviews, owns and files, and we help the firm write the disclosure policy so nobody improvises at the registry counter.
- Law Society trust accounting rules
What it demands here
Trust money moves only under human authority.
How the system complies
Agents never write to the trust ledger. They prepare reconciliation and reminders around it, and signoff stays with people.
- BC PIPA and PIPEDA
What it demands here
Client personal information handled lawfully, with residency answered before anything is built.
How the system complies
Residency is scoped per engagement and stated in writing, and where Canadian residency is required the architecture is designed for it first.
Privileged material never leaves firm-controlled systems. No-training, no-retention API terms are an architecture requirement, not a settings checkbox.
Who this is built for
Write-downs stop hiding in the timesheets.
Fixed fees turned assembly hours into cost, and the write-downs land on your desk at month end. When drafting and precedent work shed their assembly hours, the same fee carries less cost inside it. You see it in realization, not in a demo. We publish no law firm number yet because no law firm case exists yet. The founding engagement sets that number, measured against a baseline you sign.
You mark up drafts instead of typing them.
The first draft is on the matter file before you open it, assembled from the firm's own clauses with sources attached. You review it the way you would review junior work. The blank page stops being part of the job.
The precedent bank starts answering back.
You spent years herding precedents into folders nobody searches. Now a question finds the clause by its language or the argument by its substance, and every hit carries the matter and page it came from. Your curation finally compounds.
Conflicts clear before the client cools.
Intake used to sit in a queue between the first call and the engagement letter. Now the case information is gathered, the conflict check is run against the firm's records, and the letter is drafted before a lawyer picks up the file. The matter opens while the client still wants you.
What stays human
- She clears the conflict report
- She does the law
- She signs off, or she does not
The steps the day below leaves to a person, by design.
What we build for law firms
Open the matter
Client intake with conflict checks.
Case information gathered and conflicts flagged against the firm's practice management records before a lawyer touches the file.
Document chase for client files.
Undertakings and missing documents pursued automatically, with the file's quiet weeks eliminated.
Find and draft
Precedent research from your own knowledge base.
The firm's past answers found in minutes, each with matter and page.
First-draft contracts.
Deal parameters in, reviewable draft out. The lawyer does the law.
Court-ready documents.
Formatting and citations handled by deterministic templates, drafts a lawyer reviews and files.
Chronologies from the record.
First-pass review of productions and pleadings into a dated chronology with every entry pointing back to its source document. The junior checks it instead of building it. Litigation boutiques feel this one first.
Keep everyone current
Matter status reporting.
Clients and partners see where things stand without anyone composing an update.
A corporate associate's Tuesday
The referral becomes a matter while she is in a meeting.
A referral lands at 11am while she is in a closing call. By the time she is out, the case information is gathered, the conflict check has run against the firm's Clio records, and a draft engagement letter is parked on the file, flagged for review.
She clears the conflict report.
humanTwo flagged parties, one a former client of the Vancouver office. She walks it to the responsible partner, they clear it, and she approves the matter opening. Nothing opened itself.
The indemnity question is searched, not researched.
She asks how the firm has capped indemnities in share purchases like this one. The system searches closed matters in iManage and returns a ranked list, each hit carrying its matter and page.
She does the law.
humanShe adapts the closest clause to the deal on her desk and drafts the operative sections herself. The precedent gave her the firm's answer. The judgment about this deal is hers.
The ancillary documents assemble themselves.
Resolutions, closing certificates and the officer's certificate are drafted from the firm's own templates, deal parameters filled in, sources attached to every operative clause.
She signs off, or she does not.
humanShe marks the drafts up the way she would mark up a junior's work and approves them for the closing book. No document leaves the matter file without her name on the review.
The chase runs while she bills.
Outstanding undertakings are listed, the missing consent is chased with the other side's clerk copied, and a matter status line is written for the client partner. Nobody composed an update email.
The economics
Before and after economics
Line
- First draft of a standard agreement
Before
Drafted from the blank template, clause by clause, every time
After
The first draft is assembled from the firm's own clauses with sources attached, and the lawyer reviews instead of assembles
- Precedent research
Before
An associate reruns research the firm has already been paid for once
After
The firm's past answer is found by its language or its substance, and every hit carries the matter and page
- Assembly hours in scope
Before
At the page defaults, 8 timekeepers spending 8 hours a week on assembly is 64 hours, $9,600 a week at $150 loaded cost
After
The calculator below prices your own baseline, and the 5x ROI guarantee is measured against a baseline you sign before we build
- Fixed-fee margin
Before
Every assembly hour on a flat-fee matter comes out of profit, not onto an invoice
After
The same fee carries less cost inside it, qualitative by design, your realization report is the measure
No row on this table is a client outcome, because we have not published a law firm case study yet and this page does not pretend otherwise. The single modelled row multiplies this page's calculator defaults, 8 timekeepers at 8 manual hours a week each at a $150 loaded hourly cost, which is 64 hours or $9,600 of assembly per week in scope. Every other row is a qualitative state change. The first founding engagement replaces this table with measured numbers, co-published.
Where the data comes from
Document mgmt
The firm's iManage or NetDocuments workspace holds the asset this whole build runs on, the precedent bank, closed matters, versions and the marked-up history of every clause the firm has ever stood behind. The index points into it so a lawyer finds the firm's own answer first, with public databases as supplement.
Where it stops. Documents stay in the DMS under the firm's own permissions. The index references them, it does not export them, and nothing privileged crosses to a model provider except under no-training, no-retention terms, with only the draft coming back.
Practice mgmt
Clio, Soluno or PCLaw holds the matters themselves, contacts, dockets, limitation dates and the trust ledger. It is where intake becomes a matter, where the conflict check runs against the firm's real client history, and where the engagement letter waits for review.
Where it stops. The trust ledger is read-only territory. Agents prepare reconciliation and reminders around trust accounting and never post to it, because authority over trust money stays with people under Law Society rules.
Court filings
The registries and e-filing systems, Court Services Online in BC and the Federal Court's e-filing among them, hold the pleadings, orders and practice directions that set the formatting a clerk will measure. The court document agent drafts to those requirements with layout enforced by deterministic templates, not model output.
Where it stops. Nothing files itself. Every court document leaves the system as a draft, and a lawyer who owns the professional obligation reviews it, signs it and files it under the forum's own AI disclosure rules.
How the system is built for law firms
See the full capability mapRetrieval
The firm's own precedent, pleadings and closed matters, indexed so a lawyer can find the clause by its language or the argument by its substance. Every passage returned carries the matter and the page it came from.
- pgvector
- Full-text BM25
- Reciprocal rank fusion
Agents and orchestration
Agents do the first pass on document review and drafting. They never file, send or advise. Work arrives as a draft with its sources attached, and a lawyer signs it or does not.
- agent-worker
- Autonomy guard
- Proposal queue
Evaluation
Graded evals built from matters the firm has already closed, where the right answer is known. Retrieval accuracy and citation correctness are measured on that set, so a bad answer is caught by the harness rather than by a client.
- Eval graders
- Citation grading
- quality-worker
Models
Frontier models for long-document reasoning across discovery sets and precedent. Document and layout models to pull structure out of scanned filings and exhibits before any reasoning happens.
- Frontier models, one gateway, routed per task
- Embedding models via the same gateway
- Document layout extraction
Data boundary
Matter-level access control at the database layer, which is how conflict walls actually hold. The build runs inside infrastructure the firm controls, under no-training API terms.
- Supabase row-level security
- Per-matter scoping
- No-training API terms
Document mgmt, Practice mgmt, Court filings feed a hybrid index. The agent runtime works from that index, and every consequential action passes a human approval before it reaches Ask your brain, Knowledge map, Approval queue.
Your systems
- 01Document mgmt
- 02Practice mgmt
- 03Court filings
The system
- 01Hybrid index
- 02Agent runtime
- 03Your approvalhuman
Where your team works
- 01Ask your brain
- 02Knowledge map
- 03Approval queue
Document mgmt, Practice mgmt, Court filings and the system run inside a boundary labeled your accounts. The only path that crosses the boundary is the audited egress to the model API, under no-training terms. The boundary is what answers Solicitor-client privilege and Law Society of BC, Code of Professional Conduct.
Inside your accounts
- 01Document mgmt
- 02Practice mgmt
- 03Court filings
- 04System in your cloud
- 05Audit log
Outside, through the audited port
- 01Model API
The objections
We bill by the hour. Faster drafting means smaller invoices.
Your clients already answered this. Nearly 60 percent of in-house counsel say they see no savings from their outside counsel's AI, and of those who do, only 13 percent point to fewer billable hours (ACC and Everlaw, 2025). Clients are not seeing the savings. And the work moving to fixed fees is where assembly hurts most, because there every hour of it is pure cost.
Lawyers keep getting sanctioned for invented citations. Why would we volunteer for that?
Those cases share one design flaw, a model answering from its general training instead of from a record. This system answers from the firm's own closed matters, every passage carries its matter and page, and citation correctness is graded in the eval harness before any lawyer sees output. CanLII and Westlaw Canada stay what they are, research tools a lawyer drives. Nothing here files, and nothing answers from thin air.
Our precedents are a mess. Half live in email, the rest in folders nobody agrees on.
That is the first phase of the engagement, not a blocker to it. We index one practice group's closed matters and measure retrieval against questions the firm has already answered, before any agent is built. If the archive cannot answer its own history, we tell you that instead of building on top of it.
An AI that reads the whole DMS drives a truck through our ethical walls.
It does not read the whole DMS. Access is scoped per matter at the database layer, the same place your conflict walls should live, and the walls are tested with real conflict scenarios before a second practice group is added. A prompt cannot talk a database out of row-level security.
Associates learn the law by doing this work. You are automating the apprenticeship.
Assembly is not apprenticeship. Copying a closing book teaches formatting, not judgment. Associates move to the work clients will actually pay associate rates for, and review remains the teaching moment it always was. A junior who marks up a machine draft is doing the same exercise as marking up her own, minus the typing.
Your lawyers will just keep using ChatGPT on their phones anyway.
They already are, which is the problem. The same surveys that show most professionals already use these tools individually also show firm deployment sitting far below that, and almost half of employees admit using AI in ways that break policy (KPMG, 2025). A governed system inside the matter file is how the firm wins that traffic back from personal accounts.
How we build it for law firms
Start with one practice group.
We index that group's closed matters and precedent and measure citation accuracy against work the firm has already stood behind. One group, one measurable answer, before the firm commits.
Automate the first pass, never the filing.
Document review and drafting produce a draft with its sources attached. A lawyer signs it. Nothing leaves the building without a person who carries the professional obligation.
Extend across groups only after the walls hold.
Matter-level access control is tested with real conflict scenarios before a second group is added. Conflict walls are the reason firms fail at this, not model quality.
What we will not automate
Legal advice, filings and anything that reaches a client or a court unreviewed. We also will not build a system that answers from general legal knowledge rather than the firm's own file, because that is how citations get invented.
What that means in practice
Retrieval and RAG
Connecting AI to your actual documents so it answers from your knowledge, accurately and with citations, instead of making things up.
Where we stop. A dedicated vector database is justified by scale, not by default. Most of RAG quality is won or lost in chunking and indexing strategy, not in the model choice, and we have walked clients back from RAG to plain search when that was the honest answer.
Vision, documents and speech
AI for eyes and ears: reading documents, watching camera feeds, transcribing calls.
Where we stop. Bespoke computer vision is justified by volume and latency, not by novelty. Below that bar, a vision-language model on demand is cheaper to run and easier to maintain, and we will tell you which side of the bar you are on before anything is built.
Security, privacy and governance
Keeping your data yours, and your AI safe to put in front of customers.
Where we stop. We do not claim certifications we do not hold, and we will not ship a customer-facing agent without a human gate and an eval suite. If a vendor cannot offer no-training terms, it does not get into the stack.
Where your team works
Tour the platformAsk your brain
Ask a question. Get the answer and where it came from.
Every answer cites the document behind it, so you can check the source yourself. When the brain does not have the answer, it says so instead of guessing.
Knowledge map
See everything the brain knows on one map.
Every document, person, and decision the brain has learned, drawn as one connected map. Thin spots are visible too, so you know exactly what to feed it next.
Approval queue
The AI proposes. You approve. Nothing sends itself.
Every consequential action arrives as a proposal with the risk, reversibility, and expiry spelled out before you say yes. Control stays in the room.
Ask your team
Chat with a team that already knows your business.
Every agent is briefed on your documents, your data, and your preferences. Ask for the number, the draft, or the plan and cite where it came from.
By team
The same system, seen from the desk that runs it.
Run your numbers.
Your operations
Savings use the low end of our hours-reclaimed range. The math is conservative on purpose.
The math
Calculated at the low end of every range.
Every first build is covered in writing: 5x ROI in 30 days. Or we work for free.
The hard questions
The systems behind this
Custom agents→
For the jobs only your business has.
Client onboarding→
Documents chased until they arrive.
Content engine→
Publishing in your voice, every week.
Knowledge brain→
Cited answers from your own documents.
Front desk→
Every channel caught, nothing dropped.
Voice agent→
Calls answered, booked and filed.
Free · 3-5 days
Know your number in five days.
We map your operations, find the highest-ROI automations, and hand you a ranked plan with the payback math attached. Yours to keep, whoever builds it.
Prefer to talk first? Book 15 minutes with James. No pitch deck.