Contract review automation: every clause checked against your playbook, with the source shown.
A review tool that cannot show you the paragraph it relied on is a tool nobody in legal will use twice.
The same work, twice
- 1Download the redline and the previous versionLegal ops · 3m
- 2Read it against the playbookCounsel · 40m
- 3Mark every deviation and its severityCounsel · 15m
- 4Look up the agreed fallback positionCounsel · 10m
- 5Draft comments and suggested wordingCounsel · 25m
- 6Summarise the risks for the business ownerCounsel · 12m
- 7Track the version and chase the replyLegal ops · 6m
The shape of the run. Playbook, clause taxonomy and severity ranking are yours, written during the audit week.
Why this queue costs what it costs
Contract review is the same problem as insurance claims wearing a different suit. Evidence is scattered through a long document, a written policy says what is acceptable, the judgement is repeatable, and a qualified person has to remain accountable for the outcome.
That is why the pattern we use here is the one behind Nordwind's claims workbench, where decisions came 42% faster because the file was assembled and cited before the adjuster opened it. Nobody was asked to think faster.
A review run parses the agreement, identifies the clauses, checks each against your playbook, ranks the deviations by severity, attaches your agreed fallback wording, and drafts a business summary in which every statement points to a page and a paragraph. Counsel keeps the pen.
The playbook is the product
Most legal teams have a playbook that is partly a document, partly a set of email precedents and partly one person's memory of what the general counsel accepted in 2023. The automation forces it into one place: preferred position, acceptable range, fallback wording, and what always escalates.
Writing that is genuinely useful work regardless of what you build afterwards, and it is the reason the first week of one of these projects looks more like legal operations than engineering.
Citations, or nobody uses it twice
A flag that says the limitation of liability is unusual is worthless. A flag that says clause 11.2 on page 7 caps liability at fees paid in the preceding three months, against a playbook position of twelve months, with the fallback wording attached, saves a reviewer ten minutes and can be checked in ten seconds.
Everything the run produces carries that link back to source. It is the same principle that took Nordwind's adjuster adoption to 95% in week one: people trust systems that show their work and ignore systems that assert.
Where this fits, and where it does not
It fits high-volume, moderately standard agreements: NDAs, vendor MSAs, order forms, DPAs, supplier terms, and the long tail of renewals nobody has time to read carefully. Those are the documents where a review queue turns into a bottleneck.
It does not fit bespoke negotiated agreements where the value is in strategy rather than in conformity checking. For an M&A document or a first-of-its-kind partnership, the run can still extract and summarise, but the review itself is a lawyer's job and we would not pretend otherwise.
We also do not build systems that approve or sign. The run proposes; a named person disposes. No clause is ever accepted because a model was confident.
Version control and the second reading
Half the real cost of contract review is not the first read. It is the third round, where a counterparty returns a document and someone has to work out what actually changed since the version they last approved.
A run that holds every version can answer that directly: what moved, whether the movement crosses a playbook line, and whether a previously agreed concession has quietly reappeared. That is the part legal ops teams tell us they missed most.
Confidentiality and where the documents live
Contracts are among the most sensitive documents a company holds, and the answer has to be structural rather than reassuring. Documents stay in your tenancy, model calls use zero-retention endpoints, access is scoped to the review service, and the log records the decision and the citation rather than the document body.
If your outside counsel or a client contract prohibits third-party model processing for a particular counterparty, that is a per-matter switch rather than a reason to abandon the system.
Cost, timeline and the first pilot
A pilot on one agreement type starts at $4,000 and takes about three weeks, most of which is playbook work rather than engineering. A broader deployment covering several agreement types, version tracking and a reviewer interface runs $15,000 to $50,000.
The number to check afterwards is reviewer hours per agreement and time in queue. Both are measurable before you start, which is why we ask for them at the scoping call.
How the build runs
Turn the playbook into rules
Preferred position, acceptable range, fallback wording and escalation triggers, per clause type, in one file.
Build against real agreements
Twenty executed contracts with known outcomes become the evaluation set the run is graded against.
Review alongside counsel
Counsel reviews as usual and compares. Disagreements become playbook edits, which is the point of the exercise.
Hand over the playbook file
The rules live where legal can edit them, with documentation, version tracking and 30 days of support.
Before and after, in numbers
The measures we baseline before a contract review pilot. The percentage is Nordwind Insurance, where the same cited-evidence pattern was measured over 90 days.
Where we have built this
Related reading
Questions before a pilot
No, and any vendor who says otherwise is selling you a liability. It removes the first pass of reading and locating, so counsel starts at the six clauses that matter instead of at page one. Nothing is approved, agreed or signed by the system.
High-volume and moderately standard: NDAs, vendor MSAs, order forms, DPAs and renewals. Bespoke negotiated agreements are a poor fit for automated conformity checking, though extraction and summarisation still help.
Documents stay in your tenancy, model calls use zero-retention endpoints, access is scoped to the review service, and logs record citations rather than document bodies. Where a counterparty contract forbids third-party model processing, that matter is switched off individually.
Then the first week of the project is the most valuable part. Rewriting the playbook is a deliverable in its own right, and several clients have told us it was worth the pilot on its own even before the automation ran.
It attaches your pre-approved fallback wording, which covers most deviations. Novel wording is drafted by counsel, because a suggestion that has not been through your own approval process is not something we will put in front of a counterparty.
By grading it against executed agreements with known outcomes before go-live, and by running it alongside human review for the first few weeks. Recall on the clause types in your playbook is reported per type rather than as a single reassuring number.
Want to see this run on your own contract review?
Bring twenty real examples to the scoping call. The audit week counts the hours, writes the rules and comes back with a fixed price, and you keep the process map whatever you decide.