NDA review at volume for in-house teams
By the ChamberOne team 5 minute read
A sales director wants to share pricing with a prospect this afternoon. A supplier sends over its own NDA. HR asks whether a candidate can see a product roadmap. None of these is hard, but together they can eat a good part of a small legal team's week. The goal of NDA triage is simple: spend your attention where the risk is, and spend almost none where it is not.
Key points
- Write a one or two page playbook before any tool is used; software cannot apply a standard you have not set.
- Use three lanes, green, amber and red, and let AI suggest the lane while a lawyer confirms it.
- Sample green NDAs each week so that occasional errors are found.
- Keep a list of triggers that escalate an NDA whatever the business calls it.
Why NDAs pile up
Non-disclosure agreements are cheap to request and slow to review. The business sees them as a formality, so they arrive late and urgent. The legal team sees them as a minor but real risk, so they hesitate to wave them through. The result is a queue that clears in bursts and annoys everyone.
Volume is also what makes NDAs a good candidate for structured review. The same twenty clauses appear again and again. Once you know what you accept, what you negotiate and what you refuse, most of the work is comparison.
Write the playbook first
Software cannot triage against a standard you have not written down. Before any tool is involved, sit down with colleagues and agree a short playbook. Keep it to one or two pages, and organise it by clause.
- Definition of confidential information: what is acceptable, what is too broad, what must be carved out.
- Term and survival: how long the duty lasts, and whether trade secrets are treated differently.
- Permitted recipients: employees, advisers, group companies, contractors.
- Return or destruction: what is realistic given backups and legal holds.
- Governing law and forum: which are fine, which need escalation.
- Residuals clauses, non-solicitation, exclusivity and standstill terms: usually not routine, so usually escalate.
Three lanes for incoming NDAs
A practical triage uses three lanes. Name them plainly so the business understands them too.
- 1
Green: the NDA matches your playbook or is your own template, signed by someone with authority to sign. Pre-approved, with a quick human glance or a sampled check.
- 2
Amber: the NDA is close, but one or two clauses fall outside the playbook. A lawyer reads those clauses, proposes standard wording, and sends it back.
- 3
Red: unusual structure, one-way obligations that run against you, unclear counterparty, or terms that go beyond confidentiality. Full review by a lawyer.
Where AI fits in the three lanes
An AI model is good at a first read. Given an NDA and your playbook, it can list the clauses, say which ones depart from the playbook, and draft a short note on each departure. That makes the lane decision quicker.
Keep its role narrow. It suggests a lane; a lawyer confirms it. The risk with automatic approval is not that the AI will be wrong most of the time. It is that it will be wrong occasionally, and in a way no one is looking for. Sampling helps: have a lawyer read a small number of green NDAs each week, and track what the reviewer would have changed.
- Lists the clauses in the NDA
- Says which clauses depart from the playbook
- Drafts a short note on each departure
- Suggests a lane
- Confirms the lane
- Reads a small number of green NDAs each week
- Tracks what the reviewer would have changed
Limits on who may approve
Some firms and companies let non-lawyers sign low-risk NDAs from their own template. That can work well, but it needs a few guardrails that belong to your own governance, not to the software.
Decide who has authority to sign and up to what level. Decide which templates are approved. Make sure the business knows that an edited template becomes an amber NDA, even if the edit looks small. And make sure signed copies go to one place, so you can find them when somebody asks what was agreed with a particular supplier two years ago.
What a first-read checklist looks like
Whether a person or a tool does the first read, the questions are the same. Use this as a minimum before anything is marked green.
- Who are the parties, and is the counterparty the entity that will actually receive the information?
- Is the NDA mutual or one-way, and does that match the business reality?
- Does the definition of confidential information cover what we plan to share, without covering everything under the sun?
- How long do the duties last, and is that sensible for the kind of information?
- Does anything in it work as a commercial promise, such as exclusivity or a restriction on hiring?
- Are the governing law and dispute clauses inside the playbook?
Pitfalls to watch for
The first pitfall is the NDA that is really something else. A document called a confidentiality agreement can contain licence terms, standstill language or a restriction on dealing with competitors. A tool that checks only for typical NDA clauses may pass it. A lawyer glancing at the headings can catch it.
The second is the unusual counterparty. If the other side is a government body, a regulated entity or a company in a country where you have little experience, the usual playbook may not fit. Escalate by default.
The third is cultural. If the business learns that NDAs come back quickly when marked 'green', everything will be described as routine. Keep a short list of triggers that put an NDA in amber or red regardless of how it is described.
Measuring whether it works
Keep two simple measures. First, how long an NDA waits before someone looks at it. Second, how often a green NDA is later found to have contained something that should have been escalated. If the first falls and the second stays near zero, the process is working. If the second creeps up, tighten the playbook or the sampling.
Review the playbook itself every few months. Counterparties change their wording, the business starts new kinds of relationship, and a standard that was sensible last year may now be too tight or too loose. A short meeting with the people who send you the most NDAs is usually enough to spot what has drifted.
- 1Time how long NDAs wait
- 2Check green NDAs for missed escalations
- 3Review the playbook
- 4Adjust the playbook with the senders
- Then back to step 1.
Reading incoming contracts in ChamberOne
- Upload a contract and get its type, parties, key dates, clauses and citations, each linked to the source paragraph.
- Risk flags on clauses are suggestions that a partner can override.
What it does not do
- It does not decide whether an NDA can be signed. A lawyer reads the clause and makes that call.
- The analysis comes from an AI model and can miss or misread things, so check it against the document.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
AI contract review: a first-pass checklist for lawyers
What to ask an AI to flag on a first read of a contract, what it tends to miss, and where a lawyer must read every word.
5 minute readWhat AI document analysis extracts well, and what it misses
What AI document analysis does well on parties, dates, clauses and citations, and what it misses, with ways to spot the gaps.
5 minute readAI-assisted engagement letters: what to automate and what stays human
Merge fields and templates handle the routine parts of an engagement letter, while scope, fees and limits of liability need a lawyer's decision.
5 minute read