Contract Review with AI: a 3-Tier Triage Process for SMBs

When a founder of a 90-person services company told me she'd spent $14,000 on outside counsel last quarter reviewing renewal MSAs that hadn't changed a clause in three years, I asked her the obvious question. She didn't know which of those contracts genuinely needed a lawyer and which didn't. Nobody on her team did. So every contract got the same treatment — the most expensive one.

Why does contract review eat SMB legal budgets?

Because the decision "does this need a lawyer?" gets made implicitly, contract by contract, by whoever is panicking that day. Without a written triage rule, the safe path is always "send it to the lawyer." Multiply across 200 contracts a year and the bill is brutal.

Definition: Contract triage — a written rule that classifies an incoming contract into a review tier before anyone reads it, based on contract value, standardness, and counterparty risk.

The companies that escape this aren't using less legal advice. They're using it on a smaller, sharper subset of contracts and handling the rest with AI plus an ops-level review. The lawyer signs off on the triage rubric itself — once — and then ops runs the rest.

What does the 3-tier process actually look like?

Three lanes. Each contract enters exactly one. The classification rule is written down, not improvised.

Tier 1: AI alone (template + low value)

The contract uses your standard template, the counterparty is signing your paper unchanged, and the value sits below a written threshold. AI summarizes deviations, flags anything outside the approved redline playbook, and produces a one-page reviewer note. Ops signs.

Typical examples: standard customer order forms under a value threshold, vendor NDAs on your template, simple statements of work that reference an already-reviewed MSA.

Tier 2: AI plus ops review (template-ish or mid-value)

Either the contract is on a familiar template with non-trivial markup, or it's on the counterparty's paper but the value and risk sit in the middle band. AI generates a redline summary, a deviation list, and risk-flag suggestions. An ops lead — usually a finance, procurement, or operations head — reviews the AI output against your written playbook and decides whether to accept, negotiate, or escalate to tier 3.

Definition: Redline playbook — a written document of pre-approved positions on common contract clauses (limitation of liability, indemnity, IP, termination, payment terms) that ops can accept without lawyer review.

The redline playbook is what makes tier 2 work. Without it, "ops review" collapses into "ops shrug and forward to lawyer." The lawyer writes the playbook once, ops runs the daily decisions against it.

Tier 3: Lawyer (high value or non-standard)

The contract is high-value, on the counterparty's paper, in a regulated domain, contains terms outside the playbook, or involves a counterparty with known dispute history. AI still produces a pre-review summary — a redline draft, a deviation list, a question list — but the lawyer is the decision-maker. The AI output saves billable hours; it does not replace judgment.

Sample categorization rules

This sits on one page, signed off by your outside counsel, and lives in whatever your contract intake tool is.

TIER 1 (AI alone, ops signs):
- Contract type: standard order form, mutual NDA, simple SOW
- On your template: YES
- Value: ≤ $[X] annual
- Counterparty risk: low (no regulated industry, no prior disputes)
- Deviation count from template: ≤ 3 minor

TIER 2 (AI + ops review against playbook):
- Contract type: customer MSA renewal, vendor SaaS agreement, partner DPA
- Template fit: yours OR familiar counterparty paper
- Value: $[X] to $[Y] annual
- Counterparty risk: medium
- Deviations from playbook positions: any, but inside playbook coverage

TIER 3 (lawyer review required):
- Any contract above $[Y] annual
- Any contract in regulated domain (health, finance, defense, EU/UK consumer)
- Any contract with IP assignment, exclusivity, or change-of-control terms
- Any counterparty paper outside the playbook coverage
- Any counterparty with prior dispute history with us
- Any contract where ops is unsure — escalation is always safe

REVIEW PERIOD: this rubric is re-reviewed by counsel every 6 months
or any time we lose a material clause negotiation.

The thresholds are illustrative; your counsel sets them based on your business size and risk profile.

Tool tip (AIAdvisoryBoard.me): The reason 3-tier triage drifts back to "everything is tier 3" inside SMBs is that nobody surfaces the daily Plan → Fact → Gap on contract flow. Plan says "tier 1 handles 60% of contracts in under a day." Fact shows you that last week three tier-1 contracts sat with ops for nine days. Gap surfaces it before the lawyer bill arrives. The 7-day diagnostic at https://aiadvisoryboard.me/?lang=en shows you the gap automatically across every operational workflow, not just contracts.

Manager scan (2-minute digest example)

• Plan: tier 1 contracts close in ≤ 2 business days, tier 2 in ≤ 5, tier 3 within counsel SLA
• Fact: last week tier 1 median was 6 days, tier 2 was 11, tier 3 sat untouched in counsel queue for 14
• Gap: ops capacity for tier-1 review is the bottleneck — not the lawyer
• Plan: ≤ 10% of contracts get re-classified mid-review
• Fact: 23% of tier-1 starts ended up bouncing to tier 2 because the template fit was misjudged
• Gap: template-fit detection at intake is too lenient — tighten the rule
• Plan: redline playbook covers ≥ 80% of clauses we see
• Fact: 3 clauses last month were "send to counsel" because the playbook didn't cover them
• Gap: schedule a playbook update with counsel before next quarter
• Plan: zero tier-3 surprises (contract that should have been tier 3 but wasn't flagged)

Micro-case (what changes after 7-14 days)

A 90-person professional services firm with no in-house counsel was sending every contract to outside counsel — roughly 200 a year — averaging around $70 per review and a 4-day turnaround. The founder couldn't tell which contracts genuinely justified the spend. After two weeks of running the 3-tier rubric (written and signed off by counsel in a single 90-minute session): tier 1 handled about 55% of contracts in under a day with AI plus ops sign-off, tier 2 covered about 30% with the playbook, tier 3 saw 15% — the high-value and non-standard ones. Counsel spend dropped by roughly two-thirds, the median turnaround on standard contracts collapsed from 4 days to under 1, and the conversation with counsel shifted from "review this paper" to "update our playbook so ops can handle this category." That last shift is the one that compounds.

Note on this case: This example is illustrative — based on typical patterns we observe with companies of 30-500 employees, not a single named client. Specific numbers are rounded approximations of common ranges, not guarantees.

Tool tip (AIAdvisoryBoard.me): The triage rubric only stays accurate if you actually watch how contracts flow through it. The Plan → Fact → Gap pattern surfaces the cases where tier-1 contracts silently grew into tier-2 work and nobody updated the rule. Without a daily view of routing decisions, the rubric drifts within a quarter and you're back to "everything goes to the lawyer." See the 7-day diagnostic at https://aiadvisoryboard.me/?lang=en.

FAQ

Is this a substitute for hiring in-house counsel? No. It's a substitute for treating every contract as if you already had one. Once you're past roughly 250-300 contracts a year or two regulated jurisdictions, the math on a fractional or in-house GC starts to work. The triage process still applies — it just changes who runs tier 3.

What if a tier-1 contract turns out to have a hidden risk the AI missed? That's why the rubric is signed off by counsel and reviewed every six months, and why ops sign-off is the human checkpoint on tier 1. If you see a real miss, the post-mortem updates either the playbook, the threshold, or the AI's reviewer note template. Nothing is "AI alone with no human eyes."

Can the AI just be ChatGPT or do I need a legal-specific tool? Either can work for the summary and deviation-detection layer. The harder question is whether the tool integrates with your contract intake and stores the reviewer notes for audit. That's an ops decision, not a model-quality decision.

What about confidentiality of contract text going to AI tools? Use a tool with a written no-training, data-residency, and confidentiality clause your counsel has reviewed. Public free-tier consumer chatbots are not appropriate for contracts you've signed an NDA on. This is also a tier-3 sign-off the lawyer should make once, for the tool, not per contract.

How often should the rubric be updated? Every six months as a default, immediately after any contract dispute or material clause loss, and whenever your business enters a new regulated jurisdiction or product line. Treat it as a living document, not a one-time set-up.

Conclusion

The expensive failure mode isn't AI making a contract mistake. It's treating every contract the same and paying lawyer rates for work that didn't need lawyer judgment. Three tiers, one written rubric, one signed-off playbook — and a daily view of how the routing actually behaves.

This is not legal advice — your GC or outside counsel signs off the rubric, the playbook, and the tool choice. The pattern just gives you a defensible structure to bring them.

If you want a system that surfaces the Plan → Fact → Gap automatically — including which contracts are sitting in which tier and where the routing is silently breaking — see how the 7-day diagnostic works at https://aiadvisoryboard.me/?lang=en.