Contract review AI uses language models to extract clauses such as limitation of liability, indemnities, termination rights, notice periods, payment terms and key dates, compare them with your standard positions and flag risks for a lawyer or contract manager to review. Vendors such as LegalOn and Sirion report clause extraction accuracy in the high 90s, but accuracy on your own contracts must be tested before you rely on it.
Key takeaways
- AI contract review is strongest at finding and extracting clauses across large portfolios; people still interpret effect and enforceability.
- Vendor-reported clause extraction accuracy (LegalOn has cited around 98 per cent when properly configured) depends on clause type, document quality and configuration.
- The four highest-value clause groups are liability, termination, payment and compliance.
- Australian configuration matters: unfair contract terms, Privacy Act clauses, Payment Times Reporting and modern slavery provisions.
- Expect lower accuracy in the first weeks, then improvement once standard positions and risk thresholds are tuned.
The contract review problem
A scenario common in Australian procurement teams: a manager has 40 to 50 supplier contracts to check for limitation of liability before an insurance renewal. External legal review would take weeks. Contract review AI exists for exactly this job.
Missed terms are the real risk. Sirion, a contract lifecycle vendor, has reported that legal teams miss around 9 per cent of key terms in manual review (vendor research). When those include uncapped liability or automatic renewals, the cost can be serious.
- Day 1Receive Contracts40-50 documents waiting for review
- Days 2-5Manual ReviewHours per contract
- Days 6-14Legal ReviewExternal counsel queue and cost
- Days 15-21Decision and SignRisk of missed terms
With AI-assisted review, a batch like this can be extracted and flagged in hours, leaving people to focus on the exceptions.
What does contract review AI actually do?
Modern AI contract review uses natural language processing and large language models to identify, extract and compare contractual terms.
- Upload ContractPDF, Word or scanned documents
- Extract ClausesIdentify and categorise clause types
- Risk Score and FlagCompare against your standard positions
- Human ReviewFocus on flagged exceptions
Clause extraction and categorisation
Platforms detect many clause types, including:
- Limitation of liability
- Indemnities
- Termination rights and notice periods
- Payment terms and penalties
- Confidentiality
- Insurance requirements
- Intellectual property
- Dispute resolution
For example, a construction contractor with 200 subcontract agreements can identify every termination clause across the portfolio in a fraction of the time manual review takes.
Risk scoring and deviation detection
Extracted clauses are compared with your approved positions. If your standard caps liability at contract value and a supplier agreement is uncapped, it is flagged with a risk rating and suggested fallback language.
Obligation tracking
AI also tracks obligations: payment milestones, renewal dates, compliance requirements and notice periods. Automated review commonly surfaces auto-renewals due within weeks that the business meant to renegotiate.
How accurate is contract review AI? LegalOn and Sirion clause extraction accuracy
Accuracy is the most searched question about AI contract review, and published figures need context.
- LegalOn has publicly stated that its system can reach around 98 per cent accuracy on clause extraction when properly configured. This is a vendor-reported figure.
- Sirion has also published high clause extraction accuracy figures for its platform. Again, vendor-reported.
These numbers are plausible for common, well-drafted clause types in clean documents. They tell you less about your own contracts. Accuracy usually falls with:
- Scanned or poorly formatted documents
- Bespoke drafting and unusual structures
- Definitions buried in schedules or incorporated by reference
- Heavily redlined negotiation drafts
- Clause types the system was not configured for
How to test accuracy on your own contracts
| What to measure | Why it matters | Impact | |
|---|---|---|---|
| Recall per clause type | Share of real clauses the AI found | A missed indemnity is worse than a false flag | Most important |
| Precision per clause type | Share of AI flags that were correct | Too many false flags waste reviewer time | Efficiency |
| Value extraction accuracy | Correct dates, amounts, notice periods | Key dates drive renewals and notices | Obligations |
| Risk rating agreement | AI rating versus your lawyer's rating | Shows whether thresholds are tuned | Calibration |
Take 30 to 50 contracts your team has already reviewed, run them through the tool and compare results clause by clause. Ask vendors for accuracy by clause type, not one overall number.
Which AI tool extracts key dates, termination rights and indemnity terms from a commercial lease?
Most contract review platforms and general-purpose AI models can extract these terms from a commercial lease if configured with a lease clause library. Useful fields include:
- Commencement, expiry and option exercise dates
- Rent review dates and mechanism (fixed, CPI, market)
- Termination rights for each party and notice periods
- Make-good and demolition clauses
- Indemnities and releases, and their carve-outs
- Insurance obligations
- Assignment and subletting conditions
For organisations with many sites, the output belongs in a lease register with automated reminders for option and notice dates. Retail leases also carry state-specific legislation, so a person should review anything the AI flags as non-standard.
Contract review AI for conveyancing
In conveyancing, AI can read contracts of sale and special conditions, extract settlement dates, deposit terms, finance and building-inspection conditions, and compare special conditions against a firm's standard positions. It can also cross-check vendor disclosure documents for missing items. Because conveyancing is regulated and time-critical, the practitioner must review every flag and sign off. See our guide to AI for conveyancers.
The four clause categories that matter most
1. Liability clauses
- Caps: contract value, fixed amount or uncapped
- Carve-outs: gross negligence, wilful misconduct, IP infringement, confidentiality
- Mutual versus one-sided limitations
- Insurance cross-checks: exposure against required cover
Portfolio reviews commonly find a meaningful share of supplier contracts with caps below company policy, some of them with critical suppliers.
2. Termination terms
- Termination for convenience
- Notice periods
- Termination for cause triggers
- Cure periods
- Consequences: data return, transition assistance, survival clauses
Long notice periods buried deep in long agreements are found in seconds.
3. Payment terms
- Payment timing and milestones
- Late payment interest and fees
- Price escalation (CPI or fixed increases)
- Currency
- Invoice requirements
Large businesses reporting under the Payment Times Reporting Scheme can use extraction to track payment terms across suppliers. See our payment times reporting guide.
4. Compliance flags
- Privacy Act clauses: data handling, breach notification, cross-border transfers
- Modern slavery supply chain obligations
- Unfair contract terms risk in standard form contracts
- Australian Consumer Law for consumer templates
- Industry rules: financial services, building and transport
Australian legal considerations for contract review AI
Unfair contract terms regime
Since 9 November 2023, unfair terms in standard form contracts are prohibited, not just void. The regime covers contracts where a party employs fewer than 100 people or has turnover under $10 million, and penalties for companies can reach the greater of $50 million, three times the benefit obtained, or 30 per cent of adjusted turnover. Each unfair term can be a separate contravention.
Configure AI to flag:
- Unilateral variation rights
- Automatic renewal without adequate notice
- Excessive termination fees
- Broad indemnities
- One-sided limitation of liability
A practical approach is to flag any clause resembling the ACCC's published examples of potentially unfair terms. Our unfair contract terms AI audit guide goes further.
Privacy Act clauses
For contracts involving personal information, AI should identify:
- Data handling obligations
- Breach notification: under the Notifiable Data Breaches scheme, entities must assess a suspected breach within 30 days and notify the OAIC and affected individuals as soon as practicable for eligible breaches. Supplier contracts should require prompt notice to you.
- Cross-border transfers and safeguards
- Flow-down of privacy obligations to subcontractors
From 10 December 2026, privacy policies must disclose certain uses of automated decision-making. See our guide to automated decisions under the Privacy Act.
Government AI model clauses
The Digital Transformation Agency has published AI model clauses for Commonwealth procurement covering human oversight, transparency, risk management and restrictions on using buyer data to train models. They signal broader expectations, so contracts with government suppliers should be checked against them.
Implementation: what actually happens
- Week 1-2ConfigureDefine clause library, set standard positions, upload examples
- Week 3-4Pilot and refineLower accuracy at first, calibrate risk scores, filter false positives
- Week 5+Production useAccuracy improves, team reviews exceptions only
- OngoingContinuous improvementExpand clause coverage, refine thresholds, add contract types
Weeks 1 to 2: define the clause library (start with liability, termination, payment and compliance), set standard positions, provide example contracts and configure risk thresholds. Expect several hours of legal team time to define business rules.
Weeks 3 to 4: run real contracts. Expect missed or miscategorised clauses and noisy risk scores. A common early issue is every liability clause flagged as high risk until acceptable caps are configured.
Week 5 onwards: the team shifts from reviewing everything to reviewing exceptions. Ivo, another vendor, reports users saving around 45 minutes per contract at this stage (vendor-reported).
The limitations you need to know
Complex interpretation. AI extracts text accurately but can miss the practical effect of nested exceptions and cross-references.
Clause relationships. An indemnity may be modified by a liability cap, which depends on an insurance clause. AI does not always connect them.
Unusual structures. Defined terms in schedules, documents incorporated by reference, heavy redlines, letter agreements and multi-document packages cause problems.
Enforceability. AI flags clauses; assessing enforceability under Australian law needs a lawyer.
What is the return on AI contract review?
LegalOn has published estimates that a team handling 500 contracts a year at about three hours per manual review could save around 200 working days with AI assistance (vendor-reported). Measure your own baseline instead of relying on vendor figures.
| Manual | AI-assisted | Impact | |
|---|---|---|---|
| Initial contract review | Hours per contract | Exceptions reviewed in a fraction of the time | Faster |
| Portfolio clause comparison | Weeks | Hours | Much faster |
| Due diligence projects | Weeks | Days | Faster |
| Compliance audits | Days | Hours | Faster |
- Internal review hoursFocused on exceptions
- External counsel spendReserved for genuine legal questions
- Missed renewals and notice datesTracked automatically
- Unfavourable terms found before signingThe biggest benefit
Getting started
What's your current contract volume and review burden?
- Hundreds of contracts a year and significant review timeStrong case, pilot now
- Moderate volume but high risk exposurePilot for risk visibility regardless of volume
- Low volume, standard termsManual review may be sufficient for now
- Portfolio review needed once (insurance, M&A, audit)Project-based AI extraction
- Audit your current state: volume, time per contract, clause types searched, pain points, near-misses.
- Define requirements: priority clauses, standard positions, deviation thresholds, applicable compliance frameworks.
- Evaluate tools: Australian conventions, integration with your document management system, accuracy by clause type, configuration effort and whether data is stored in Australia.
- Start small: one contract type, one clause category, 20 to 30 contracts.
How Solve8 can help
We build contract extraction workflows for legal, procurement and property teams: clause libraries matched to your positions, lease and supplier registers with date alerts, and review queues where a person approves every flag. Models run on your infrastructure in Australia, so contracts stay in your environment. See our professional services work and custom AI development, or talk to our team.
