An NDA playbook is the documented set of acceptable, fallback, and walk-away positions an in-house team applies when reviewing or drafting non-disclosure agreements. Its purpose is that most NDAs never reach an attorney: the positions are decided once, encoded into a review tool, and then applied by a coordinator, a paralegal, or an AI agent. NDAs are the highest-volume agreement most legal teams handle and the lowest-value use of attorney judgment, which is the entire argument for writing the positions down instead of re-deciding them 500 times a year.
Start from a published form, not a blank page
Two free public forms already encode a defensible set of positions, and adopting one is faster than drafting your own:
- oneNDA (v2.1) is a community-drafted mutual NDA whose language is fixed — you populate cover-page variables (parties, purpose, confidentiality period, governing law) and change nothing else. It carries no residuals clause, a 30-day return-or-destroy window on the discloser’s request with archival and record-retention carve-outs, and injunctive relief as the named remedy. If you negotiate the wording, the project asks you to remove the oneNDA branding, which is the enforcement mechanism that makes the form worth adopting at all: two counterparties who both run unamended oneNDA have nothing to negotiate.
- Common Paper’s Mutual NDA (v1.0) separates fixed Standard Terms from a Cover Page you configure. Its defaults are one year of confidentiality from the effective date, indefinite protection for anything qualifying as a trade secret, the four conventional exclusions, and Delaware law and courts.
Notice the gap between them: Common Paper defaults to one year of confidentiality, and most in-house playbooks land at two to three. Your playbook has to have an opinion about that gap. Adopting a public form does not remove the need for a playbook — it moves the playbook’s job from drafting language to choosing configuration values and deciding what you accept on counterparty paper.
Mutual vs unilateral
- Mutual NDA. Both parties disclose. The default for partnerships, vendor evaluations, early-stage M&A conversations, and most B2B engagements where information moves both ways.
- Unilateral NDA. One party discloses, and only the recipient carries obligations. Correct for interview processes, one-way product demos, and inbound vendor pitches.
Default to mutual unless there is a specific reason not to. A unilateral NDA sent for a genuinely two-way engagement usually means the counterparty reached for the nearest template, not that they made a decision.
The clause grid
| Clause | Acceptable | Fallback | Walk-away |
|---|---|---|---|
| Definition of Confidential Information | Marked, or reasonably understood as confidential given the circumstances, tied to the stated purpose | Marked writings plus oral disclosures confirmed in writing within 30 days | ”Any and all information” with no purpose limit and no exclusions |
| Term of confidentiality | 2-3 years from disclosure; indefinite for trade secrets only | 5 years across all information types | Perpetual obligations on everything, or a term shorter than the information’s useful life |
| Permitted use | The evaluation purpose named explicitly on the cover page | A broader but bounded business purpose | Any use, unbounded |
| Exclusions | All four: public domain, prior knowledge, independently developed, rightfully received from a third party | All four, with a reasonable documentation standard | Fewer than four, or a burden of proof you cannot meet from your own records |
| Return / destruction | On written request, with carve-outs for one archival copy, routine backups, and legal holds | Within 30 days of termination, carve-outs intact | Certified destruction of all copies including backups on short notice |
| Remedies | Injunctive relief without bond plus actual damages; each side bears its own fees | Injunctive relief plus actual damages | Liquidated or punitive damages, or one-way fee-shifting |
| Governing law | Your home jurisdiction, or a neutral one with developed trade-secret law | The counterparty’s home jurisdiction if it meets that test | A forum connected to neither party, or one hostile to enforcement |
| Assignment | Not without written consent | Permitted to affiliates and in a change of control | Free assignment to any third party |
Two things the grid will not tell you on its own.
Residuals depend on which side of the table you sit on. A residuals clause lets the receiving party use what its people remember without notes. That is good for a net recipient of information and bad for a net discloser, so a single blanket position is wrong for one half of your deals. If your team spends most of its NDA volume evaluating other companies’ technology, a narrow unaided-memory residuals clause — excluding trade secrets and deliberate memorization — is worth keeping. If your NDAs mostly protect your own roadmap, pricing, and customer data, strike it. Write both variants into the playbook and route on counterparty type rather than picking one and living with it.
Industry overlays sit on top of the grid. Biotech NDAs add IP non-use language, defense-sector work adds export-control terms, and anything touching patient or EU personal data pulls in a HIPAA acknowledgment or the DPA checklist instead of resolving inside the NDA.
Worked example: what the auto-approve rate is actually worth
Take a team signing 600 NDAs a year that routes every one to an attorney for a 35-minute read at a $220/hour loaded internal rate. That is 350 attorney-hours and roughly $77,000 of internal cost, before counting the deal days lost while each agreement waits in a queue.
Now move 65% of that volume onto an auto-approve path. The 210 escalated agreements still take 35 minutes each (122 hours), and a 10% spot-check sample of the 390 auto-approved ones takes five minutes each (3 hours). Attorney load lands near 126 hours, about $27,700 — and the queue time on two-thirds of the volume drops to same-day.
The whole return sits in one variable: the auto-approve rate. And the auto-approve rate is set by how wide you wrote the “acceptable” column, not by which vendor you bought. Calibrate against these bands:
- Under 40% auto-approved after two quarters of encoding: the acceptable column describes your perfect NDA rather than an acceptable one. Widen it until the escalation queue holds only agreements a lawyer would genuinely have changed.
- 60-75%: the target range for a mutual-NDA-heavy book of work.
- Over 85%: usually a sign the walk-away column is missing a position, not that the playbook is unusually good. Audit a sample of what got approved.
How to operationalize
- Encode the routing at CLM intake. The intake form asks three questions — counterparty type, related deal value if any, and industry or jurisdiction — and picks the template and the approval path from the answers. See legal intake for the form design.
- Put a triage step in front of review. LegalOn shipped a Triage Agent in March 2026 that scores an incoming agreement against your encoded playbook at high/medium/low priority and fast-tracks low-risk agreements while escalating nonstandard terms; it ships 50+ attorney-written playbooks you edit in plain English rather than building from zero. Ivo and Spellbook run the equivalent check inside the Word add-in where the reviewer already works.
- Auto-redline counterparty paper. When the counterparty sends their form, the tool marks it up to your positions before a human opens it. BlackBoiler is the narrow pick for high volumes of third-party paper — you submit the draft, you get back tracked changes in Word. The lawyer then reviews a redline, not an unmarked document.
- Escalate on four triggers. A walk-away position appears; the counterparty is a government entity or in a sanctioned jurisdiction; the related deal exceeds the auto-approve cap; or the counterparty rejects your redline twice. Anything else clears on the automated path.
- Measure two numbers monthly. Auto-approve rate against the bands above, and median cycle time split by path — target same business day for auto-approved, three business days for escalated. A rising escalation rate with flat cycle time means the playbook is drifting out of date, not that volume grew.
- Store counterparty positions. When a repeat counterparty has insisted on the same fallback three times, that becomes a stored default for them. The negotiation you skip is the one you already had.
Common pitfalls
- The playbook names tools that no longer exist. This page previously routed auto-approval through LawGeex, whose enterprise product was dismantled in 2023 — assets to Robin AI, remaining clients to LegalSifter. A playbook that names a dead vendor teaches the business the document is stale, and they stop trusting the rest of it. Guard: date the tool column and re-check it on the same cycle as your renewal calendar.
- Treating NDAs as the most important contract type. They are the most common, not the most consequential. The playbook’s job is to make NDAs nearly invisible to attorney attention so the team can spend that attention on the MSA redlining work that carries real risk. Guard: cap the total attorney hours the NDA path is allowed to consume per quarter and treat a breach of that cap as a playbook defect.
- Perpetual obligations accepted by default. Some counterparty templates say “in perpetuity” and most reviewers let it through because it sounds protective. It is an indefinite obligation your own team has to track and honor. Guard: make perpetual-on-all-information a hard walk-away in the encoded rules so a human has to override it deliberately.
- Trusting the vendor’s accuracy number. Every contract-AI vendor publishes a benchmark measured on its own curated sample; see AI contract review accuracy for what those figures do and do not establish. Guard: pilot on 30-50 of your own signed NDAs, have a senior reviewer grade the output against what they would have written, and set the auto-approve threshold from the miss rate you measured.
When this framework breaks down
The playbook stops paying for itself in three situations. Low volume: under roughly 50 NDAs a year, encoding and maintaining the rules costs more than the reviews it saves — write a one-page position sheet instead. One dominant counterparty: if 80% of your NDAs come from a single customer or partner on their paper, negotiate one master confidentiality agreement and stop reviewing individual forms. Highly asymmetric disclosure: when you are the only meaningful discloser and the information is genuinely core, an NDA is a weak instrument regardless of how well the playbook is written — stage the disclosure, hold back the sensitive material until later in the process, and put the protection in the commercial terms.
Related
- Contract review SOP — the broader review process the NDA playbook fits inside
- MSA redlining rubric — the equivalent grid for higher-stakes commercial agreements
- Clause library design — how to store the fallback language the playbook points at
- LegalOn — playbook-driven review with a triage step in front of it
- BlackBoiler — auto-markup of high volumes of counterparty paper