The verdict in three sentences
Beyond 1,000 contracts to analyse, AI clause extraction costs three to five times less than manual review and fits within a due diligence timeline. The 2026 budget sits between EUR 12,000 and 35,000 excl. VAT, with 85 to 95% accuracy depending on the clause type. The setup is only reliable with human sample-based checks and EU hosting.
The problem: 4,000 contracts and six weeks
The general counsel of an industrial group in Lyon is preparing an acquisition. He must map 4,000 supplier contracts: term and renewal, termination notice, penalties, exclusivity, change of control clauses, governing law. The contracts are heterogeneous: native PDFs, scans from the 2010s, English versions, separate amendments.
At 20 to 40 minutes per contract, manual review means 1,300 to 2,700 hours. Outsourced to a law firm, it would cost several hundred thousand euros. In-house, it would tie up the team for months.
| Approach | Total time (4,000 contracts) | 2026 cost (estimate) | Timeline |
|---|---|---|---|
| In-house manual review | 1,300 to 2,700 h | EUR 90,000 to 190,000 of loaded time | 4 to 8 months |
| Law firm (juniors) | 1,300 to 2,700 h | EUR 200,000 to 400,000 excl. VAT | 6 to 10 weeks |
| SaaS review platform | 300 to 500 h of checks | EUR 25,000 to 60,000 excl. VAT licence | 4 to 6 weeks |
| Custom AI extraction | 250 to 450 h of checks | EUR 12,000 to 35,000 excl. VAT | 4 to 8 weeks |
| Mixed AI and law firm | 200 internal hours and targeted review | EUR 40,000 to 80,000 excl. VAT | 4 to 6 weeks |
What accuracy to expect by clause
Accuracy depends on the type of information. Structured data (parties, dates, amounts) is extracted very reliably. Clauses drafted in variable ways, such as change of control or liability caps, need closer human checks. Scan quality matters too: a poorly digitised document drops accuracy by 5 to 15 points.
| Type of information | Observed AI accuracy | Recommended human check |
|---|---|---|
| Parties, signature dates, term | 95 to 98% | 5% sample |
| Automatic renewal and notice | 90 to 95% | 10% sample |
| Penalties and caps | 87 to 93% | 15% sample |
| Exclusivity and non-compete | 85 to 92% | 20% sample |
| Change of control | 85 to 90% | 100% review of flagged contracts |
| Poor-quality scanned contracts | 75 to 85% | Enhanced review |
The right setup gives each extraction a confidence score and cites the source passage. Lawyers only review uncertain cases and critical clauses, rather than everything.
Architecture and confidentiality
A serious project chains text recognition on scans, splitting contracts and their amendments, extraction by a language model guided by a clause grid, then a validation interface that exports a table usable in the data room. On confidentiality, acquisition contracts are highly sensitive: EU hosting, APIs with no data reuse, access logging, data deletion at the end of the engagement. These requirements align with GDPR and the deal's confidentiality commitments.
The budget varies with volume, share of scans, number of clauses tracked (10 to 30) and planned reuse. Keeping the tool afterwards for contract portfolio management, with expiry alerts, adds EUR 5,000 to 10,000 excl. VAT.
Mini case study
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Isabelle, general counsel of an industrial group in Lyon, must analyse 4,000 contracts in 6 weeks. Manual review: 30 minutes on average, i.e. 2,000 hours, about EUR 140,000 at EUR 70 loaded per hour. With AI extraction at EUR 24,000 excl. VAT on 18 clauses and 350 hours of human checks (EUR 24,500), the total cost reaches EUR 48,500. The saving exceeds EUR 91,000, or 65%, and the 6-week deadline is met. The 140 contracts flagged for change of control are fully reviewed by the team, securing the most sensitive point of the deal.
FAQ
Can AI replace a lawyer's review?
No, it replaces the spotting and data entry work, which is 70 to 80% of the time. Legal analysis of risky clauses is still done by lawyers.
Are scanned contracts usable?
Yes, after text recognition, with accuracy 5 to 15 points lower for poor-quality scans. These documents then go to enhanced review.
Where is the data hosted?
In the European Union, with APIs that do not reuse content for training. Data is deleted at the end of the engagement, usually within 30 days.
How long to launch the extraction?
Allow 2 to 3 weeks to define the clause grid and calibrate the tool on 100 contracts, then 1 to 2 weeks of processing for 4,000 contracts.
Can the tool be reused after the acquisition?
Yes, it can become a contract portfolio management tool with expiry alerts, for an extra EUR 5,000 to 10,000 excl. VAT.
Let's scope your project. Tell us the contract volume, share of scans and clauses to track: we will price an extraction between EUR 12,000 and 35,000 excl. VAT, calibrated on a 100-contract sample. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
