Give lawyers back the hours that aren't legal work.
Much of a lawyer's week goes on work around the advice: first-pass review against a playbook, document assembly, deadline tracking, data room indexing and billing guideline checks. We look for the workflows where agents can do that preparation, with controls that protect confidentiality and deadlines. Legal advice and every legal judgment stay with your lawyers.
Six workflows around the advice, not the advice itself.
These are the workflows we would examine first in a law firm or an in-house legal team. They run through the document management system, practice management software, email, data rooms and client billing portals.
| Workflow | Manual today | AI-native version | Stays human | Verification check |
|---|---|---|---|---|
| First-pass contract review | Associates read routine contracts such as NDAs clause by clause against the firm's or client's playbook and mark them up. | An agent compares each clause with the playbook, marks deviations, proposes the standard fallback and cites the playbook position. | Every review decision, every negotiation position and all advice to the client. | Each flag cites a playbook clause. A lawyer signs off every document before it leaves. |
| Matter intake and conflict search | Staff collect party names and matter details by email, then run and read conflict searches by hand. | An agent structures the intake, normalizes party and group names, runs the search and summarizes possible hits. | Every conflict decision and acceptance of the matter. | Searches cover all parties and known affiliates. Possible hits are never cleared automatically. |
| Document assembly | Lawyers copy the last similar document and edit names, dates and clauses, sometimes leaving the old ones in. | An agent drafts from approved precedents and the matter record, and lists every variable it filled. | Drafting choices, non-standard clauses and final review. | Only approved precedent versions are used. Names, dates and amounts must match the matter record. |
| Deadline and docket management | Assistants read court and authority correspondence, work out deadlines and enter them in calendars. | An agent reads incoming correspondence, proposes the deadline with its basis, and creates the entry for confirmation. | Deadline calculation sign-off and any extension decision. | Every proposed deadline is confirmed by a person. Unconfirmed deadlines escalate daily. |
| Due diligence document review | Teams open every data room file, sort it, and fill a review table with key terms by hand. | An agent indexes the data room, extracts key terms such as change of control and termination, and flags missing documents. | Findings, risk assessment and the report to the client. | Every extracted term links to the source page. A lawyer samples extractions before they enter the report. |
| Billing guideline compliance | Partners and billing staff read pre-bills line by line against each client's outside counsel guidelines. | An agent checks every time entry against the client's guidelines and flags block billing, vague narratives and rate issues. | Write-downs, narrative changes and invoice approval. | Each flag cites the guideline it breaks. Rates must match the engagement letter. |
Six numbers that show whether it worked.
The Audit measures these before anything is built. The Pilot is measured against the same numbers.
| Metric | Definition |
|---|---|
| Review time per contract | Lawyer minutes per routine contract, from receipt to marked-up draft. |
| Intake to clearance | Working days from a new matter request to conflict clearance. |
| Deadline exceptions | Deadlines entered late, wrongly or not at all, found by review, per quarter. |
| Precedent drift | Documents sent using an outdated or unapproved precedent, per 100 documents. |
| Invoice rejection rate | Invoices or time entries rejected by clients under their billing guidelines, per 100 invoices. |
| Write-downs | Time recorded but not billed, as a share of recorded time. |
What can go wrong in legal work, and the control for each.
Legal advice has legal consequences for people and companies, so it stays with lawyers. Agents prepare, compare and flag. Every output a client sees is reviewed by a lawyer, and every agent step is written to an audit trail.
| Failure mode | Control | How we'd detect it |
|---|---|---|
| Agent output treated as advice | Agent outputs are marked as drafts and go to a named lawyer. Nothing is sent to a client without lawyer sign-off. | Document log of who approved each outgoing document. |
| Confidential information crosses matters | Agents work inside one matter at a time, with access set by your ethical walls. | Access logs per agent, per matter and per task. |
| Deviation from playbook missed | Clauses the agent can't map to the playbook are flagged, never passed. Lawyers review every contract. | Sampled second review of contracts, compared with the agent's flags. |
| Deadline missed or miscalculated | A person confirms every proposed deadline. Unconfirmed entries escalate. | Daily report of correspondence received with no confirmed deadline. |
| Invented source or citation | Every extracted term or reference links to a page in your documents. Links that don't resolve are rejected. | Automated link check on every output, plus lawyer sampling. |
| Conflict missed | Possible hits always go to a person. Name normalization covers known affiliates and former names. | Periodic re-run of searches on open matters. |
First-pass NDA review, from diagnostic to pilot.
Illustrative example — not a client result. The numbers below are assumptions chosen to show the method and the arithmetic. Your Audit replaces them with measured values.
| 01 | Diagnostic | Suppose a commercial law firm picks "first-pass review of incoming NDAs" on the Other / General Business track. The Snapshot might show a Medium leverage signal with Medium confidence, friction at the Processing and Verification stages, and Level 2 (workflow augmentation) as the indicated AI depth, because a lawyer still reviews every document. |
| 02 | Audit | Suppose the Audit measures 300 NDAs a month at 30 minutes of associate time each. That is 300 × 30 = 9,000 minutes, or 150 hours a month, about 0.9 FTE at 160 hours. It also finds that about 60% of NDAs are close to the playbook and 40% have material deviations. |
| 03 | Pilot | Suppose the agent marks up every NDA against the playbook. A lawyer reviews the 60% close-to-standard ones in 10 minutes each and the 40% with deviations in 25 minutes each. A lawyer reviews every NDA. The Pilot measures whether those times hold and whether anything the lawyer catches was missed by the agent. |
| Line | Assumption | Hours per month |
|---|---|---|
| Baseline | 300 NDAs × 30 min | 150.0 |
| Close to standard, after | 180 NDAs × 10 min lawyer review | 30.0 |
| Deviations, after | 120 NDAs × 25 min lawyer review | 50.0 |
| After, total | 30.0 + 50.0 | 80.0 |
| Difference | 150.0 − 80.0 (about 0.4 FTE) | 70.0 |
At an assumed loaded cost of €90 an hour for associate time, 70 hours is about €6,300 a month. Under fixed-fee or capped arrangements, that time is margin. Under hourly billing, the Audit would look at the effect on revenue as well as cost, because that changes the business case. It would also count new costs: compute, document management integration and playbook upkeep. If the numbers don't hold, the recommendation is "don't proceed".
Where our proof comes from.
We'll be plain about this. Our production proof is in AdTech, not legal services. We have not yet run these legal workflows for a client.
What carries over is the architecture: typed intake, deterministic checks on every agent output, every claim linked to its source, bounded retries, then a person. It runs in AdTech production, where our case studies show the work, and five AI plugins built on it are live at reviveadservermod.com. Our governance working paper describes how runtime controls, human escalation and an audit trail fit together.
That is why legal work starts with the free diagnostic and then the Audit. The Audit measures your own baseline and records what must never be automated, before anything is designed.
Questions lawyers ask first.
Will an agent give legal advice?
No. Legal advice and legal judgments stay with your lawyers. Agents prepare drafts, comparisons and summaries that a named lawyer reviews before anything reaches a client.
How is client confidentiality protected?
Agents work inside one matter at a time with access set by your ethical walls, and every access is logged. Self-hosted or EU-hosted deployment is possible, so documents can stay in your environment. Your own professional obligations still apply, and we recommend involving your risk or compliance partner early.
Isn't this what legal AI products already do?
Many legal AI tools help with a single task such as review or research. We look at the whole workflow, from intake to invoice, and put checks on the handoffs. Where a tool you already use does a step well, the design can include it.
Find out which legal workflow is worth redesigning.
Start with the free diagnostic on the Other / General Business track. Move to the Audit only if the Snapshot points to something worth measuring.
| 01 | Free AI Operational Diagnostic | 10 steps, 8–12 minutes, General track. Snapshot on screen and as a PDF. |
| 02 | Operational Leverage Audit, from $1,500 | Fixed fee, one workflow, 10–15 working days after data access. Measured baseline, verification architecture, benefit model with stated assumptions, go / no-go. |
| 03 | Pilot implementation, from $6,000 | Fixed scope. The workflow built and run in your environment, with an audit trail, measured against the Audit baseline. Typically 4–8 weeks. |