AI Contract Review for Indian Businesses: A Practical Guide

AI contract review reads a contract, compares it against a defined standard, and returns a structured view of what the document says and what it fails to say. It is fast, consistent and tireless on mechanical work. It does not exercise legal judgment and it does not carry professional accountability. For an Indian business the useful question is therefore not whether software can review contracts, but which parts of your current review are mechanical enough to hand over, and which parts must stay with a qualified professional.

What AI contract review actually does

An AI contract reviewer performs five mechanical jobs, each of which a junior would otherwise do by hand, and each of which software does more consistently at volume.

• Extraction: pulling parties, term, renewal mechanics, notice periods, payment terms, liability caps, indemnities, governing law and forum into a structured record you can sort, filter and search.
• Comparison: measuring the draft in front of you against your own standard position and reporting where it deviates and by how much.
• Flagging: marking clauses that carry disproportionate risk, such as uncapped indemnities, automatic renewal with a long notice window, unilateral price escalation or one-sided termination rights.
• Absence detection: identifying what is missing. A great deal of contractual pain in practice comes from clauses that were never drafted, not from clauses drafted badly.
• Drafting: producing a first-cut rewrite or a tracked-change redline that a reviewer can accept, reject or edit.

None of these is a legal opinion. Each is a well-defined information task, which is precisely why software handles them reliably.

What it does not do

AI does not decide whether a risk is acceptable. Acceptability depends on the counterparty, the deal size, the commercial relationship, the client's risk appetite and the litigation posture of the sector, none of which appear within the four corners of the document.

• It does not advise. Advice to a client remains the responsibility of the advocate or professional engaged; software output is an input to that advice, never a substitute for it.
• It does not know your commercial context: the concession already given on price, the fact that this counterparty is your largest customer, or the board resolution limiting the signatory's authority.
• It does not verify facts outside the document, such as whether the signatory was authorised, whether the correct stamp duty was paid in the state of execution, or whether the counterparty is solvent.
• It does not carry accountability. The person whose name is on the advice, the filing or the negotiating position still owns the outcome.

Where it fits in an Indian legal workflow

The best results come from inserting AI at three specific points rather than treating it as a general assistant sitting alongside the team.

Intake and triage

Many in-house teams receive contracts through email, WhatsApp and a shared drive, with no consistent view of what is pending. An automated first pass that extracts key terms, returns a risk indication and routes the document by contract type turns an unsorted queue into a prioritised one. The reviewer then spends their first hour on the three agreements that matter rather than on the three that arrived most recently.

First-pass review and redlining

For high-volume, moderate-value agreements such as NDAs, vendor terms and standard order forms, an AI first pass removes the mechanical layer of work. The reviewer receives a marked-up draft with the clauses and the omissions already located, checks them against the written playbook the team maintains, and spends their time on the deviations that are genuinely contentious. Whether a particular tool can be configured to weight its own output to your playbook is a question to put to the vendor rather than an assumption to make. Tracked-change redlines exportable to Word matter here, because Indian negotiation still runs on Word documents circulated by email.

Portfolio and renewal questions

Questions such as which of our contracts auto-renew in the next ninety days, which ones cap liability below one year's fees, and which ones lack a data protection addendum are impractical to answer manually across several hundred agreements, because the cost and elapsed time scale with the size of the portfolio. This is where AI produces value that adding reviewers does not produce economically, because the task is search and comparison at scale rather than judgment.

Indian-law specifics the review must account for

Templates written for other jurisdictions fail in predictable ways in India. A reviewer, human or machine, should be checking against the following.

• Stamp duty is a divided field: rates for the instruments listed in Entry 91 of the Union List are fixed centrally, while rates for most commercial documents are fixed by the states under Entry 63 of the State List. Stamping depends on the state of execution and the nature of the instrument. In some states duty is charged under the Indian Stamp Act, 1899 as amended in its application to that state; other states have their own legislation, such as the Maharashtra Stamp Act, 1958 and the Karnataka Stamp Act, 1957. An instrument that is not duly stamped is inadmissible in evidence under Section 35 of the Indian Stamp Act, 1899 or the corresponding provision of the applicable State Stamp Act. An insufficiently stamped instrument can generally be impounded and then admitted on payment of the deficient duty and the prescribed penalty, subject to the exceptions in the applicable Act, so the defect is usually curable but at cost and delay at exactly the wrong moment. Software can flag a missing stamping clause; it cannot confirm what was actually paid.
• Electronic execution has statutory backing with exclusions. Section 10A of the Information Technology Act, 2000 recognises contracts formed through electronic means, while the First Schedule to that Act excludes certain classes of instrument. Confirm the class of document, and the current content of that Schedule, before defaulting to e-signature.
• Section 27 of the Indian Contract Act, 1872 makes an agreement restraining a person from exercising a lawful profession, trade or business void to that extent, subject to the statutory exception for the sale of goodwill. A post-termination non-compete copied from a foreign precedent usually does not survive in India.
• Section 28 of the Indian Contract Act, 1872 voids agreements that absolutely restrict a party from enforcing rights before ordinary tribunals, with a saving for arbitration. An exclusive jurisdiction clause works only where the chosen court would otherwise have jurisdiction; parties cannot confer jurisdiction on a court that has none.
• Section 74 of the Indian Contract Act, 1872 governs stipulated sums. Where a figure is named as compensation for breach, a court awards reasonable compensation not exceeding that amount. The named figure is a ceiling, not an automatic entitlement.
• Section 19(1) of the Copyright Act, 1957 requires an assignment of copyright to be in writing signed by the assignor or by his duly authorised agent. A services agreement stating that the customer 'shall own' deliverables, with no express written assignment, is weaker than it appears.
• Payment and tax mechanics change the economics: whether fees are inclusive or exclusive of GST, the withholding position under the governing income-tax legislation, and the statutory payment timelines under the Micro, Small and Medium Enterprises Development Act, 2006 where the supplier is a registered micro or small enterprise. Note that the Income-tax Act, 2025 has replaced the Income-tax Act, 1961, so any section number carried over from an older template or precedent should be treated as a claim to verify against the current bare Act rather than reproduced. Verify current thresholds and timelines against the notification in force.

What to check before you trust the output

Treat every AI-generated review the way you would treat a first draft from a new junior: useful, worth reading closely, and not yet advice.

• Does every flag point to specific words? A risk flag that cannot be traced to a clause and a sentence is not reviewable, and an unreviewable flag cannot be used in advice.
• Have you tested for missed issues, not just wrong ones? A false flag costs two minutes; a missed uncapped indemnity costs the negotiation. Run the tool over ten contracts already reviewed by hand and count what it failed to find.
• Are statutory references checked against the bare Act? Treat any section number in generated output as a claim to verify, particularly where a statute has recently been replaced. The same discipline should apply to human drafts.
• Is the whole instrument in front of it? Schedules, annexures, statements of work, purchase orders, side letters and incorporated policies carry a great deal of the operative risk, and are often the pages nobody uploaded.
• Has a person read the numbers and dates? Notice periods, cure periods, cap amounts, currency, effective dates and renewal windows are cheap to verify directly and expensive to get wrong.
• Is the output benchmarked against your playbook or a generic standard? Ask the vendor directly which of the two it is, because a deviation report only means something once you know what the system is comparing the draft against.
• Who is signing off? Name the professional accountable for the final position on each reviewed contract. The tool produces an input; a person produces the advice.

How LexVio approaches contract review

LexVio treats contract review as a structured first pass rather than an opinion. A reviewed contract returns a 0-100 Legal Health Score, clause-level risk flags, and, where a rewrite is wanted, tracked-change redlines exportable to Word. The Legal Health Score is LexVio's own assessment of the document as reviewed; it is not weighted to your playbook, so your record of which deviations you accept remains a discipline your team maintains. Vio works on a single document as a copilot; Nexus works across the portfolio, with search by clause type, clause benchmarking and drift alerts when executed positions begin moving away from your standard. Because review sits in the same platform as Indian court research across the Supreme Court, High Courts, NCLT, ITAT, CCI and CESTAT, the clause in front of you and the authority you want to check it against are not in two different tools. Whether an authority remains good law is a judgment for the professional reading it, made against the bare judgment and the current position.

A sensible way to roll this out

Start narrow. Pick one contract type you sign often and write down your standard position on the ten clauses that matter in it. Run the tool over agreements you have already reviewed manually and compare. Where the tool and the reviewer disagree, the useful question is which was right and why, and that conversation is how a playbook actually gets built. Once the first contract type is stable, add the next. A narrow scope with a written playbook gives you something measurable to compare against; switching the tool on across every contract type at once gives you a volume of flags nobody has time to read.

The right test for an AI contract reviewer is not whether it is correct more often than a lawyer. It is whether it makes the lawyer faster and harder to surprise, while leaving the judgment and the accountability exactly where they were.