Most teams run pre-contract reviews with a fixed set of 10–20 questions. That’s necessary—but it isn’t sufficient. Contracts fail in the gaps: the things you didn’t ask, the assumptions you forgot, and the delivery-phase signals that only show up in the inbox.

Here’s the play: build a two-layer safety net.

1. Layer 1: Known-knowns — your departures template (CSV) that nails the must-haves every time.
2. Layer 2: Unknown-unknowns — a broad AI sweep (“what else could hurt me?”) plus a tuned Email Monitor that keeps listening once the job goes live.

Do both and you’ll stop leaving risk on the table—without bloating the process or turning a review into a research project.

Layer 1: Your departures template (the “known-knowns”)

You still need the backbone. A good departures template:

• Standardises the must-ask items (delay notice period, LD triggers, variation procedure, site access, indemnities, insurances, security, caps).
• Cites the clause and links to the page—every time.
• Autofills your fields (Clause Ref, Issue, Risk, Preferred Position, Proposed Wording, Next Step).
• Exports cleanly to the client format.

This is your quality floor. It ensures you never forget the basics and you always have a clause-linked, defensible first pass.

But checklists are blind to context. That’s why we add the second layer.

Layer 2A: The open-ended AI sweep (find what you didn’t think to ask)

Once you’ve run the departures template, run a broad exploratory pass to hunt for risks you didn’t anticipate. This is a single prompt you save and reuse on every contract:

Starter prompt (contractor lens):

“Acting for the contractor, scan the entire contract (all schedules/annexures). List the top 10 additional commercial/operational risks not covered by a standard departures checklist. For each, include: the clause reference + hyperlink, a one-sentence implication, and a suggested mitigation (preferred position, process step, or KPI to monitor). Consider: hidden time bars, pass-through obligations, vague ‘reasonable endeavours’, non-standard approvals, consumables ownership, onboarding/training duties, site access dependencies, data/reporting burdens, step-in rights, and termination edge cases. Use a stronger search depth.”

Why this works:

• Perspective shift. You force the model to think like a contractor who loses money when things slip.
• Category prompts. You seed “buckets” (time bars, pass-throughs, approvals, consumables, data, step-in) that often hide gotchas.
• Output discipline. Clause link + implication + mitigation keeps the result actionable.

You’ll typically surface 5–10 extra items per contract—things that never appear on the base checklist but cost real money on site.

Layer 2B: Tune the Email Monitor (listen where risk actually shows up)

After award, the action moves from PDFs to emails and daily reports. That’s where delays, access issues, consumables bleed, and KPI misses first appear. Your second-layer safety net has to listen continuously.

How to tune it:

1. Set project-specific thresholds. The same event is “low” on one job and “high” on another. Tell the monitor what you care about (e.g., access delays ≥2 hrs, crusher downtime ≥3 hrs, rework >$5k, permit blockers >1 day).
2. Add operational KPIs to watch. Don’t stop at pure contract triggers. Add the leading indicators that precede claims: daily “Take 5” counts, near-miss reports, man-hours vs plan, subcontractor non-conformance rates, inspection pass rates.
3. Lock the “human-in-the-loop.” Drafts never auto-send. Humans approve notices and RFIs. Speed with control.
4. Promote what matters. If the monitor flags something as medium but you know it’s critical for this project (say, vendor change approvals), elevate it. Likewise, demote noise.

End state: a steady feed of clause-linked, actionable signals—not an alert firehose.

Putting it together: the 60-minute runbook

0–15 min: Run the departures template

• Select the contract folder.
• Apply your CSV review.
• Get clause-linked answers and draft departures.
• Edit the edge cases.

15–35 min: Run the open-ended sweep

• Paste the broad prompt.
• Review the “top 10 additional risks.”
• Convert the real ones into departures or “watch items.”

35–60 min: Tune the Email Monitor

• Add thresholds and KPIs (what you actually track on this job).
• Set the initial routing/approvals (who approves notices).
• Create your first Case from a likely workflow (e.g., EOT/Delay) so deadlines are live from day one.

You’ve now got knowns covered, unknowns hunted, and live listening turned on.

Examples of “unknown-unknowns” the sweep catches

These show up again and again, outside the standard 10–20 items:

• Hidden procedural time bars buried in schedules (e.g., ancillary permits, incident reporting windows, commissioning approvals).
• Pass-through third-party approvals (client’s client, landholder, regulator) that stall progress but sit outside your direct control.
• Consumables and spares ownership/return conditions that quietly shift cost.
• Data/reporting obligations with penalties for missing format/frequency (sneaky admin burn).
• Vague “reasonable endeavours” clauses that mask one-way obligations.
• KPI formulas that double-count downtime or exclude legitimate relief causes.
• Training/induction burdens that create unfunded overhead for ramp-up and turnover.
• Step-in/suspension rights with weak cure periods that spike cashflow risk.

Each one gets a clause link, a short implication, and a mitigation (wording change, process control, KPI watch).

Delivery: connect unknowns to action (so they don’t die in a spreadsheet)

Finding an extra risk is useless unless it drives behaviour. Convert the keepers into:

• Departures (pre-award): original text → strikeout → proposed wording.
• Workflows (post-award): add or edit steps (e.g., “Regulator approval → request within 2 biz days → clause link → owner”).
• Cases (live): when the trigger fires, a deadline and owner exist.
• Key Dates (soft obligations): KPI reviews, gate meetings, audit windows, option notices.

Your site team should never see an essay. They should see the clause and the next step, with a due date.

How to measure if your safety net is working

Leaders don’t need fluff; they need numbers that move:

• Extra risks found per contract (from the open-ended sweep) and how many became departures or watch items.
• Time-bar misses pre/post Email Monitor tuning.
• Notice lead time (first draft vs deadline).
• Noise ratio in the monitor (alerts approved vs dismissed) after the first week of tuning.
• Cycle time for first-pass reviews (template + sweep) vs baseline.

If the unknown-unknown layer is doing its job, you’ll see more relevant risks caught earlier with fewer noisy alerts.

Common objections (and fast answers)

“Won’t a broad sweep create noise?”
Not if you ask for top 10 with clause links + mitigations. You’re curating, not boiling the ocean.

“Our checklist already works.”
Great—it’s layer 1. This adds layer 2 so you catch the blind spots and the live signals.

“My PMs don’t have time for more process.”
They won’t. They’ll see fewer surprises and more “do this next” with a clause and a due date.

“Isn’t this what Copilot/ChatGPT does?”
Generic tools draft paragraphs. This system links every finding to a clause, structures it into departures/cases, and keeps a human approval step. That’s defensible.

The one-pager you can give to every new PM

Two-Layer Safety Net

1. Run the template. Get the standard departures done (clause-linked).
2. Run the sweep. Ask, “What else can hurt us?” Save the top 10 with mitigations.
3. Tune the monitor. Elevate what matters, demote noise. Add the KPIs we care about.
4. Turn signals into cases. If the monitor flags it, we either dismiss or open a case with a deadline.
5. Never ship without a clause link. If it can’t cite, it doesn’t go out.

Tape that above the screen. It’s the habit stack that prevents “we didn’t know.”

Quick start: copy-paste prompts

Open-ended sweep (contractor):

“Acting for the contractor, scan the entire contract (all schedules/annexures). List the top 10 additional commercial/operational risks not covered by a standard departures checklist. For each: clause reference + hyperlink, one-sentence implication, suggested mitigation (preferred wording, process control, or KPI to monitor). Consider time bars beyond delays, pass-through approvals, consumables/spares, data/reporting burdens, vague ‘reasonable endeavours,’ KPI definitions, training/induction burdens, step-in/suspension rights, termination edge cases. Use deeper search.”

Email Monitor tuning (project):

“For this project, elevate to High when: access delay >2 hrs, critical plant downtime >3 hrs, unapproved scope changes or permit blockers >1 day. Track KPIs: Take-5s per person per day (target 3), near-misses (weekly trend), man-hours vs plan (±10%), NCRs (rolling 7-day). Flag any breach with the clause link and draft the next step; do not auto-send—route to approver.”

Bottom line

Checklists protect what you already know. The job of AI is to surface what you didn’t. Pair them. A two-layer safety net—departures template + open-ended sweep + tuned Email Monitor—catches the unknown-unknowns before they turn into cost, delay, or dispute.

Clause. Link. Action. And a system that keeps listening after award. That’s how you stop losing to the stuff you didn’t ask.

‍