Ask five vendors for a demo of AI contract review and you will see five products that appear to do the same thing and are built on entirely different assumptions. Most disappointing purchases in this category come from buying the right product for the wrong problem.
There are four distinct categories here. They are priced differently, they are implemented differently, and they fail differently.
1. Playbook review
The problem it solves: a high volume of inbound third-party paper that has to be checked against your standard positions.
You encode your positions once. Your preferred limitation of liability, your fallback, your walk-away. The system reads an incoming contract and flags every clause that departs from the playbook, with a suggested redline.
This is the highest return category in legal technology when the conditions are right, and the conditions are specific: you need real volume, contracts that are broadly similar, and positions that can actually be written down. A team reviewing thirty NDAs and forty vendor agreements a month is the ideal case.
Where it fails: bespoke, heavily negotiated agreements where every deal is different. There is no playbook to encode, so the tool has nothing to measure against.
The hidden cost: building the playbook. This is legal work, it takes senior time, and it is where these implementations stall. Budget several weeks of a knowledgeable lawyer’s attention. A vendor who tells you the playbook builds itself is selling you a disappointment.
2. Full contract lifecycle management
The problem it solves: nobody knows where the contracts are, what is in them, or when they renew.
CLM is a system of record. Intake, approval workflow, negotiation, signature, repository, obligation tracking, renewal alerts. AI sits on top for extraction and search.
The value is rarely in the review speed. It is in being able to answer “which of our contracts have a change of control clause” in an afternoon rather than a fortnight, and in nothing auto-renewing by accident.
Where it fails: when it is bought as a review tool. CLM is a business process implementation that touches procurement, sales and finance. If the wider business will not change how it requests contracts, you have bought an expensive filing cabinet.
The hidden cost: migrating and extracting the historic contract set, and integration with the systems the business actually uses. Implementation regularly runs six to twelve months in a mid-sized company.
3. Drafting assistants
The problem it solves: producing a first draft faster.
These sit in the drafting environment, usually Microsoft Word, and generate or rewrite clauses on request. They are the easiest category to adopt because they change nothing structural, and the easiest to over-value for the same reason.
Where it fails: the output needs the same review as any first draft, so the saving is real but bounded. It also degrades quietly where the drafting requires knowledge the tool does not have, such as a client’s commercial history or a regulator’s current posture.
The hidden cost: low, which is genuinely a point in favour. The risk is not cost, it is that light-touch adoption produces light-touch verification. See the sanctions record for what that looks like at the extreme.
4. Negotiation and comparison tools
The problem it solves: tracking positions across versions and counterparties.
Version comparison, position tracking across a portfolio, market-standard benchmarking. Useful for teams doing repeated negotiations of the same agreement type at scale.
Where it fails: low volume. If you negotiate twelve agreements a year, a person with a spreadsheet is better and cheaper.
Choosing between them
Two questions settle most cases.
How similar are your contracts to each other? Highly standardised means playbook review. Highly bespoke means drafting assistance, because nothing else has purchase.
Is the pain in reviewing or in finding? If the complaint is that review is slow, buy review. If the complaint is that nobody can find the executed version or knows what was agreed, buy CLM. These are different products and the second one is a much larger project.
What to test in a demo
Vendor demos run on documents the vendor chose. Insist on yours.
- Bring three of your own contracts, including one deliberately awkward one with unusual formatting or a scanned page.
- Ask it to find something that is absent. Missing clauses are where these tools are weakest and where the risk actually sits.
- Include a document in a second language if you operate across borders.
- Ask what happens when the system is unsure. A tool that flags uncertainty is safer than one that always sounds confident.
- Ask specifically what the false negative rate looks like, and what testing supports the answer.
On accuracy claims
Published accuracy percentages in this category are close to meaningless without knowing the test set, who built it, and what counts as correct. Treat any number a vendor quotes as marketing until they will tell you the methodology.
The number that matters is your own. Run a set of contracts you already know the answers on, and count what the tool missed. That measurement takes a day and is worth more than every benchmark in the deck.
Before you buy anything
Work out what the current process costs. Volume per month, hours per contract, cost per hour. Many teams discover the honest answer is that their volume does not justify a platform, and that a drafting assistant plus a better template set solves most of it.
That is a good outcome. It is much better than a six-figure CLM nobody uses.
Related: how to draft a legally binding contract and our guide to running a legal AI pilot that produces a decision.
Where the legal industry reads first.
Enjoyed this article? Get the biggest legal industry updates, deals, appointments, insights and expert interviews in your inbox, free.
No spam. Unsubscribe anytime.