Contract Review with AI: What It Flags, What Your Counsel Still Decides
By Akshit Agrawal, Co-founder, Bitvyas · Published 3 November 2026 · Facts as at 3 November 2026
At Bitvyas, we build contract review as a workflow where AI reads each incoming contract, compares every clause with your playbook, and flags what is missing or unusual, with a risk level. Clean contracts go to the business owner; flagged ones go to counsel. A lawyer or authorised signer still decides and signs every contract.
In many companies, contracts still arrive from every team at once and get read by hand, in whatever order they land. The same unusual clauses get missed when time is short, and the quality of the review depends on who happens to read it. This post maps the process the way our hub post, What to automate with AI first, does: the business process, where AI takes it over, and what stays with a person. It applies to in-house teams and growing businesses in India, the UK and the US.
What are the key facts about AI contract review?
| Fact | Detail | Source |
|---|---|---|
| What the AI does | Extracts text, splits it into clauses, labels each clause by topic, pulls out parties, governing law and amounts, compares clauses with a playbook, proposes edits | Law Insider, "How AI Contract Review Works" |
| What a person does | Accepts or rejects each proposed change and keeps control of the final text | Law Insider |
| Score on our 4-question test | Volume: high. Rules: medium (a written playbook is needed first). Cost of error: high. Data: medium | Bitvyas, hub post |
| Published time-saving claims | Vendor figures of 40–90% depending on contract type; no method or sample size given | Sirion, page updated 27 May 2026 |
| Safe starting point | Standard NDAs and service agreements, with review on every output | Bitvyas view |
| As of | 3 November 2026 |
Which part of contract review can AI take over?
AI can take over the first read: finding clauses, extracting the key terms, and checking each one against your playbook. It cannot take over the judgement about whether to accept a risk.
The first read is where hours go and where attention slips. The judgement is where a lawyer's training matters. The workflow below keeps that split clear.
How does the process run, step by step?
1. A contract arrives. From email, a shared folder or a contract system. AI logs it, identifies the type (NDA, service agreement, licence, lease) and the counterparty. This is intake, and it needs no judgement.
2. AI reads and splits the document. It extracts text from PDF or Word files, keeps headings and numbering, and breaks the text into clauses. Scanned files need a reading step first, and poor scans are the first thing to check in a pilot.
3. AI labels the clauses and pulls the key terms. Each clause is tagged by topic, such as liability, indemnity, termination or payment. Parties, governing law, dates and amounts are extracted into a short summary.
4. AI compares each clause with your playbook. The playbook is your written position: what you accept, what you push back on, and what you never accept. Missing clauses are noted, and misaligned ones get a proposed edit that reuses the contract's own wording where it can. This step is only as good as the playbook, which is why writing it down comes first.
5. AI marks risk and routes the contract. Clean contracts go to the business owner with the summary. Flagged contracts go to counsel with the flags, the proposed edits and the reason for each.
6. A person reviews. Counsel accepts, edits or rejects each proposed change. Nothing goes to the other side without that step.
7. A person signs. The authorised signer signs. AI can track the signature and file the executed copy, then record key dates such as renewal and notice periods.
What stays with a person?
Every legal judgement and every sign-off stays with a person. That includes whether to accept a non-standard liability cap, how to word a negotiation position, and anything the playbook does not cover.
On our hub test, contract review carries a high cost of error, so we run it only with review on every output. In a first month, we would have counsel check the AI's reading of every contract, then relax the check by contract type only as the evidence builds.
What can AI get wrong in contract review?
It can misread a scanned page, miss a clause that sits in an unusual place, or apply a playbook rule too literally. It can also sound confident while being wrong.
Three controls cover most of this. Show the source text next to every flag, so a reviewer checks the clause rather than trusting a summary. Keep a log of what the AI flagged and what counsel changed. Run new contract types in parallel with the manual process before relying on the output.
What do vendors claim about time saved?
Vendors claim large savings, but the published figures come with no method. One vendor page, updated 27 May 2026, reports 85–90% time savings on standard NDAs and 40–55% on complex M&A documents, with no baseline, sample size or method.
Treat these as the vendor's claims, not measured results. The direction is plausible: the simpler and more standard the contract, the more of the first read AI can handle. For your own numbers, time a sample of your contracts by hand before you start, then again with the workflow. Our post What AI automation costs, and how to work out payback shows how to turn that into a payback figure.
What does a worked example look like?
The numbers below are an illustration with stated assumptions, not a client result.
| Assumption | Value |
|---|---|
| Contracts reviewed per month | 40 |
| Manual first read, per contract | 45 minutes |
| First read with AI plus counsel check | 20 minutes |
| Counsel time saved per month | 40 × 25 = 1,000 minutes, about 16.7 hours |
Multiply the hours saved by your counsel's or manager's cost per hour, then subtract the build, run, review and upkeep costs from the cost post above. If the answer is small, start with a smaller scope, such as NDAs only. If the answer is large, check that the 20-minute figure holds on your own contracts before you rely on it.
Which contracts should you start with?
Start with standard, high-volume contracts where you already have a preferred position: NDAs, vendor service agreements, and standard customer terms. Leave bespoke, high-value or regulated agreements to a full manual read until the workflow has a track record.
Two rules help. Do not start until the playbook exists in writing. And do not let any contract skip the person who reviews it.
What about contracts under Indian law?
Contract review for Indian law firms needs tools built on Indian law and Indian case law, and that is not what Bitvyas does. Our sister product Lawgix AI builds legal AI for Indian law firms and in-house legal teams; both are products of Bitvyas Private Limited, so treat that as a disclosure. For a business that wants its own contracts triaged against its own playbook, the workflow above is what we build.
On validity, Section 10A of the Information Technology Act, 2000 provides that a contract is not unenforceable solely because it was formed in electronic form (Mondaq summary of the Act, checked 3 November 2026). Check the Act itself for the full text. This is general information, not legal advice.
How does this fit with your other automations?
Contract review sits beside the other document-heavy processes. Signed contracts feed vendor statement reconciliation with the agreed terms, and the same intake step works for client document collection. You can see the legal workflow in action on our AI triage incoming contracts use-case page, and our services page explains how we scope a first workflow.
Sources
- Law Insider, "How AI Contract Review Works: From Raw Text to Actionable Redlines", 30 June 2025: https://www.lawinsider.com/resources/articles/how-ai-contract-review-works-from-raw-text-to-actionable-redlines
- Sirion, "Playbook-Driven AI Redlining Benchmarks 2026" (vendor page, last updated 27 May 2026): https://www.sirion.ai/library/contract-insights/ai-redlining-benchmarks-legal-ops-playbook/
- Mondaq, "Validity of Electronic Contracts in India": https://www.mondaq.com/india/contracts-and-commercial-law/699022/validity-of-electronic-contracts-in-india
- Bitvyas, What to automate with AI first
Figures and sources checked 3 November 2026.
See it working
Keep reading
Still doing this by hand?
Bitvyas builds AI that takes this kind of work off your team. It reads the documents, does the matching and drafts the next step, inside the tools you already use. A person still approves anything that matters.
Tell us what your team repeats