AI for Legal Work
Contract review that finds the important clauses fast.
What is AI legal?
AI for legal work reads agreements, extracts the terms that matter into a consistent summary, flags wording that differs from your standard position, and tracks renewal and notice dates, without offering advice or approving anything.
Reading contracts is slow, and the real exposure is missing one clause buried in a long document that nobody had time to read closely.
We set up a system that reads agreements and pulls out what matters: payment terms, notice periods, liability caps, penalties, renewal dates, and anything differing from your standard wording.
It does not advise and it does not approve. It finds and summarises, so your lawyer spends time on judgement rather than on searching.
Why this is a safety net rather than a substitute
Duara implements contract review from Nairobi for businesses signing many similar agreements: suppliers, leases, service contracts and employment terms.
We are firm that this is a safety net rather than a substitute for reading. Anything the system flags gets read by a person, and genuinely important agreements should still be read in full. A client who treats extraction as a replacement for legal review has bought the wrong thing.
Who this is for
- You sign many contracts with similar terms
- Renewal dates get missed
- Legal review is a bottleneck on volume
- Nobody can find what you agreed two years ago
Industries we serve
- Property and leasing
- Construction
- Distribution and supply
- Financial services
- Professional services firms
- Public sector procurement
Problems this solves
These situations are where extraction saves genuine time rather than adding a layer of false comfort.
- One unusual clause in a long document. Most terms are standard, so attention drops, and the clause that matters is the one nobody reached.
- Renewal dates missed entirely. An agreement renews automatically because a notice deadline passed unnoticed, which is often more costly than the terms themselves.
- Legal review blocking the business. Volume exceeds review capacity, so contracts either wait or get signed without proper reading.
- No searchable record. Finding what you agreed with a particular supplier two years ago means opening files one at a time.
What we deliver
Consistent extraction of key terms, flagging of anything non-standard, tracked renewal and notice dates, and a searchable record of what you have signed.
Key terms extracted into a consistent summary format
Clauses differing from your standard wording flagged
Renewal and notice dates tracked with reminders
A searchable record of your executed agreements
Comparison between versions during negotiation
Confidence levels so uncertain extractions are visible
Features
- Comparison against your standard. The system knows your preferred wording and flags departures from it, which is where the exposure sits.
- Date tracking with reminders. Notice periods are calculated backwards from renewal dates, so the reminder arrives while action is still possible.
- Version comparison. During negotiation, changes between drafts are surfaced rather than needing manual comparison.
- Confidence flagging. Where extraction is uncertain the system says so, rather than presenting a guess as a finding.
- Search across executed agreements. Finding every contract with a particular clause becomes a query rather than an afternoon.
Benefits
- Fewer missed deadlines. Automatic renewal because a notice date passed is a common and entirely avoidable cost.
- Faster review cycles. Lawyers start from an extracted summary rather than a blank first read, which changes throughput materially.
- Consistent attention. The last contract of the day gets the same extraction as the first, which is not true of human reading.
- Institutional memory. What you agreed survives the departure of the person who negotiated it.
Technologies we use
- Clause extraction. Identifying and pulling defined term types from agreements in varied formats and drafting styles.
- Standard position comparison. Matching extracted wording against your preferred clauses to surface departures.
- Date calculation. Deriving notice deadlines from renewal terms so reminders fire while action remains possible.
- Version differencing. Highlighting what changed between contract drafts during negotiation.
- Secure document storage. Agreements held under your own access rules, never used to train public models.
Extraction accuracy varies by document quality and drafting style. We measure it on your own agreements rather than quoting a general figure.
How the process works, step by step
Contract extraction is usually running against live agreements within four to five weeks.
-
Step 1
Free volume review
We establish how many agreements you handle and how similar they are, since similarity is what makes extraction reliable.
-
Step 2
Standard position capture
We record your preferred wording for each term type, which is what departures are measured against.
-
Step 3
Accuracy testing
Extraction is tested against agreements you have already reviewed, so accuracy is measured rather than claimed.
-
Step 4
Historical backfill
Existing executed agreements are processed so the searchable record starts populated rather than empty.
-
Step 5
Live running
New agreements are extracted as they arrive, with flagged items going to your lawyer.
Quality assurance and how we work
The limits we set on anything touching legal documents.
- It never gives advice. The system finds and summarises. Interpretation, advice and approval remain entirely with your lawyer.
- Flagged items are always read. Anything the system flags goes to a person. It is a safety net, not a substitute for reading.
- Accuracy is measured on your documents. We test against agreements you have already reviewed rather than quoting a general accuracy figure.
- Uncertainty is shown, not hidden. Low confidence extractions are marked as such, because a confident wrong summary is worse than no summary.
- Documents never leave your control. Agreements are stored under your access rules and are never used to train public AI models.
Frequently asked questions about AI legal
Does this replace a lawyer?
No. It finds and summarises. Advice, interpretation and approval stay with your lawyer, and anything flagged is read by a person.
Is it a substitute for reading contracts?
It is a safety net, not a substitute. Genuinely important agreements should still be read in full, and we say so before starting.
How accurate is extraction?
It varies with document quality and drafting style, so we measure it against agreements you have already reviewed rather than quoting a general figure.
What happens when it is unsure?
The extraction is marked low confidence rather than presented as a finding, because a confident wrong summary is worse than none.
Are our contracts kept confidential?
Yes. They stay in your systems under your access rules and are never used to train public AI models.
Can it track renewal dates?
Yes, and notice deadlines are calculated backwards from renewal so the reminder arrives while action is still possible.
Does it compare contract versions?
Yes, surfacing what changed between drafts during negotiation rather than requiring manual comparison.
What is a standard position?
Your preferred wording for a given term. Departures from it are what the system flags, because that is where exposure sits.
Can it handle scanned agreements?
Yes, though extraction confidence is lower on poor scans, and those are flagged rather than guessed at.
Will it process our existing contracts?
Yes, as a historical backfill, so the searchable record starts populated rather than building up slowly from new agreements.
How long before it is reviewing documents?
Four to five weeks, including capturing your standard positions and testing accuracy on agreements you have already reviewed.
Can it draft contracts?
That is a separate question and we are cautious about it. Extraction and flagging carry far less risk than generation.
Does it work with agreements in Swahili?
Yes, though most commercial agreements here are in English. Mixed language documents are handled.
What term types can it extract?
Whatever you define, commonly payment terms, notice periods, liability caps, penalties, renewal terms and governing law.
Can we search across all agreements?
Yes. Finding every contract containing a particular clause becomes a query rather than an afternoon of opening files.
What if a contract is unusual?
Extraction confidence drops and it is flagged for full human review, which is the correct outcome for anything genuinely unusual.
Who from the legal team should take part?
Whoever handles contracts and whoever holds the standard positions, usually legal and commercial together.
Does it integrate with document storage?
Yes, with the common systems, so agreements stay where your team already keeps them.
Can it flag missing clauses?
Yes, and an absent liability cap or notice provision is frequently more significant than an unusual one.
What does legal document work cost to run?
Agreement volume sets it. We project the monthly cost from your own contract counts and fix it before any work is agreed.
Glossary of terms
Contract terms as used in this work.
- Clause extraction
- Automatically identifying and pulling specific term types out of a contract.
- Notice period
- How much warning must be given before terminating or declining renewal.
- Auto-renewal
- A term causing an agreement to continue automatically unless notice is served in time.
- Standard position
- Your preferred wording for a given term, used as the benchmark for flagging departures.
- Liability cap
- The contractual limit on what a party can be required to pay if something goes wrong.
- Redline
- A marked-up draft showing proposed changes between contract versions.
Related services in AI Implementation
Talk to us about AI legal
Tell us roughly how many agreements you sign a year and how similar they are. Similarity is what makes extraction reliable, and it decides feasibility quickly.
Start your project