Scope & boundaries
This guide covers reading the contract rather than filing it — playbook depth, third-party paper and lawyer adoption, in a category where not one vendor publishes a rate.
It does not cover the repository, the approval workflow, the signature and the renewal calendar (Contract Lifecycle Management (CLM)), tracking that obligation as a compliance requirement once it exists (Governance, Risk & Compliance (GRC)), or the invoice that agreement is supposed to govern, and paying it (AI Invoice Processing & AP Automation).
Executive Summary
Two products both called contract AI: one knows where every contract is and what it says, the other reads the one on your screen and tells you what is wrong with it. Buying the first when you needed the second is the most expensive mistake in this category.
This market grew from two directions and the join is still visible. From one side came contract lifecycle management — the repository, the approval workflow, the signature, the renewal calendar — which added AI to a system that already existed. Ironclad positions itself as AI contract lifecycle management; Icertis describes itself as AI-native contract intelligence; Malbek surfaces revenue, risk and commercial opportunities from a contract repository. From the other side came the reading tools, built to help a lawyer with the document in front of them. Definely frames its product around reducing review time and standardizing quality. Luminance describes Legal-Grade agents automating contract activity across a business.
Those are different purchases with different buyers. The repository is bought by legal operations, sold on cycle time and visibility, and implemented as a system migration. The reading tool is bought by lawyers, sold on the hour, and adopted or abandoned within two weeks. A shortlist that mixes them produces a comparison where every product wins on its own axis, which is how these evaluations stall.
Why Contract Review Was the First Real Legal AI Use Case
Contract review was where legal AI worked first, and for structural reasons rather than technical ones. Contracts are semi-structured, they repeat, the same clauses recur across thousands of documents, and there is a well-understood notion of what a deviation from standard looks like. That is close to an ideal machine-learning problem, and it is why this category matured while other legal AI applications were still demos.
The third-party paper distinction is the one most likely to be missed and the most consequential. Reviewing your own template with negotiated changes is close to a diff, and a competent system handles it. Reviewing a supplier's contract written on their paper, in their structure, with their definitions, is a genuinely hard reading problem — and it is also where most legal review time actually goes in a procurement-heavy organization. Ask any vendor to demonstrate on inbound third-party contracts from your own files rather than on a marked-up version of a standard template.
There is also a delivery-model question that this category surfaces more sharply than most. Thomson Reuters describes its AI as built on 175 years of its own knowledge, and LexisNexis positions Lexis+ with Protégé as legal AI built on authority — both leaning on proprietary legal corpora. Harvey describes purpose-built agents that execute complex legal work end to end and covers legal, regulatory and tax research. Legora describes an agent that plans, executes, reviews and delivers complex legal work. Whether a research corpus is a durable advantage in contract review specifically, as opposed to in legal research, is a genuinely open question and worth asking each vendor directly rather than assuming.
Which type of AI Contract Review & Intelligence fits your organization?
Building is not realistic here and the reason is instructive. The models can read a contract; what they cannot do without help is know what your organization considers acceptable — your fallback positions, your risk tolerance on limitation of liability, which indemnity language your general counsel will sign and which they will not. That knowledge exists in a handful of people's heads and in past negotiated agreements, and encoding it is the whole product. Spellbook describes an AI system powering contracts end to end and covers drafting, redlining and communication; the drafting-with-your-standards part is the part that is hard to replicate.
The genuine decision is between four things. A CLM platform with AI attached solves visibility and workflow. A reading tool solves review speed for individual lawyers. A legal research platform brings authority and a broader legal remit. A general-purpose legal agent covers work well beyond contracts. They are priced and adopted completely differently, and only the first is a system migration.
| Approach | What you are buying | What it will not do |
|---|---|---|
| CLM with AI | Repository, workflow, signature and renewal visibility | Make a lawyer faster on the document in front of them today. |
| Contract reading tool | Faster, more consistent review of a document | Tell you what you have signed. It reads; it does not remember. |
| Legal research platform | Authority-backed drafting and research alongside review | Run your contracting process. It serves the lawyer, not the workflow. |
| General legal agent | Work across matters, of which contracts are one type | Match a specialist on contract-specific depth, at least today. |
| In-workflow review assistant | Help inside the document editor lawyers already use | Exist as a platform. It is a productivity tool and priced like one. |
| Your CLM vendor's own AI | Whatever is bundled where the contracts already live | Match a specialist reading tool — but it needs no migration and no new adoption. |
How do you evaluate AI Contract Review & Intelligence?
Clause extraction accuracy is what gets demonstrated and it is close to table stakes; every serious product finds the limitation of liability clause. What separates these systems is whether they can tell you that a found clause is a problem for you specifically, whether the lawyer can see why, and whether the whole thing happens somewhere a lawyer already works.
Four vectors matter once the pilot is over. Playbook depth is first and is the actual product: how your positions, fallbacks and red lines get in, who maintains them, and whether the system distinguishes must-have from nice-to-have. Third-party paper handling is second, because that is where the difficult volume is. Explainability is third — a flag a lawyer cannot trace to the clause and the rule is a flag they will check manually, which returns the time you were buying. Fourth is where the work happens: an assistant inside the document beats a better assistant outside it, more or less every time.
| Capability | What it does | Buyer translation |
|---|---|---|
| Playbook encoding | Captures your positions, fallbacks and red lines | The actual product. Ask who maintains it after month three, and what happens when it goes stale. |
| Third-party paper review | Reads contracts written on somebody else's template | Where the hard volume is. Demand a demo on your own inbound contracts, not a marked-up standard. |
| Explainability | Shows why a clause was flagged | A flag a lawyer must verify by hand costs the time it was supposed to save. |
| In-editor operation | Works inside the document rather than beside it | Definely frames its product around reducing review time and standardizing quality, in the lawyer's own workflow. |
| Repository intelligence | Answers questions across everything already signed | Malbek surfaces revenue, risk and commercial opportunities from a contract repository — a different product from review. |
| Agentic execution | Carries out multi-step work rather than annotating | Legora describes an agent that plans, executes, reviews and delivers complex legal work. |
| Security posture | How contract text is handled and certified | LinkSquares describes ISO-certified security with enterprise-grade controls built in. Ask specifically about training on your text. |
Which vendors lead in AI Contract Review & Intelligence?
The camps below reflect which end of the problem each vendor started from, which still predicts what the product does well even as everyone converges on similar language. Vendors are extending across the join — the CLM platforms are adding review depth and the reading tools are adding repositories — but the origin shows in what the product assumes about your data.
One caution about reading these pages. Every vendor now describes AI-native, agentic, end-to-end contract intelligence, and the copy will not tell you which camp a product belongs to. The reliable test is what the product assumes you already have: a tool that assumes a repository of signed agreements is CLM-descended, and one that assumes a document open on your screen is a reading tool. Ask each vendor which of your two problems — knowing what you have signed, or reviewing what is in front of you — they would send you elsewhere for.
| Vendor | Approach | Where it fits |
|---|---|---|
| Icertis | AI-native contract intelligence | Enterprises querying obligations and risk across a large signed corpus |
| Ironclad | CLM platforms with AI | Legal operations teams whose problem is workflow and visibility |
| Luminance | Contract reading tools | Legal teams whose bottleneck is review volume, including third-party paper |
| Harvey | General legal agents | Organizations whose legal AI ambition extends well past contracts |
| Definely | In-editor assistants | Teams where lawyer adoption is the real constraint on any tool |
| Thomson Reuters | Legal research platforms | Teams wanting contract work alongside authority-backed research |
One representative of each approach is named here; the category runs to several dozen vendors and the boundary between contract AI and contract lifecycle management is actively dissolving. The camps were written before the vendors were chosen, and no placement here is for sale. Any vendor in this category can speak for themselves in the Spotlight below.
How much should you budget for AI Contract Review & Intelligence?
Not one vendor examined for this guide publishes a rate. Ironclad's page is titled Pricing and Plans and carries no price on it; the only dollar figure it contains is an announcement about Ironclad's own annual recurring revenue. Every other pricing URL either quotes, redirects to a demo request, or does not exist. This guide therefore quotes no figure in this category, because inventing a range would be the precise failure this corpus is built to avoid.
What can be said is which units the market uses, and the important thing about them is that they measure completely different quantities. A per-seat license prices access for lawyers and is indifferent to volume. A per-contract or per-page fee prices the work and is indifferent to team size. A platform tier prices your organization. A repository-size band prices your history — which is the awkward one, because your back catalog is fixed, already exists, and grows whether or not you use the product. Ask which of these you are on before discussing the number, because the same headline can be cheap or ruinous depending on which quantity your organization is large in.
Three costs sit outside every quote and one of them is genuinely large. Playbook construction is the first: encoding your positions, fallbacks and red lines is the work that makes the product function, it requires your most senior lawyers, and no vendor does it for you at any meaningful depth. Historical contract ingestion is the second — getting a back catalog of signed PDFs into a queryable state is a document project with its own budget, and it is the one most often discovered after signing. The third is the parallel-running period: for the first months lawyers will check the AI's work, which means the review time goes up before it goes down, and a business case that does not allow for that arrives late.
| Basis | You are charged for | Grows with | Where it goes wrong |
|---|---|---|---|
| Per seat | Each lawyer or user with access | Legal headcount | Occasional users, who cost the same as daily ones and use it twice. |
| Per contract reviewed | Each document processed | Deal volume | Success. More contracts reviewed is the outcome you wanted and the bill you did not. |
| Per page | Document length | Contract complexity | Long-form agreements, where the expensive contracts are also the long ones. |
| Repository size band | How much signed paper you hold | History, which you cannot reduce | Large back catalogs, priced before the product delivers anything. |
| Platform tier | A band of capability and capacity | Whatever the band counts | Bands discovered at renewal rather than at signing. |
| Bundled in CLM | Nothing incremental, until the tier moves | The CLM's pricing | Renewal, when the AI you adopted turns out to sit one tier up. |
No rate appears in this section because no vendor examined publishes one. Where a page is titled Pricing and carries no price, the evidence ledger records exactly that.
How long does implementation take for AI Contract Review & Intelligence?
Two things determine whether this purchase works and neither is the software. The first is the playbook, which is senior lawyer time and cannot be delegated. The second is adoption, which is decided in the first two weeks by whether the tool fits how lawyers already work.
Positions, fallbacks and red lines, by contract type, with must-have separated from preferred. This requires your most senior lawyers and it is the input that determines whether the output is useful. Organizations that discover their positions are inconsistent gain something from this phase even if no tool is bought.
Hide the outcome, compare each tool's flags against what your lawyers actually raised, and count both misses and noise. Include inbound third-party paper deliberately — a tool that only performs on your own template solves the easy half of the problem.
In the editor if possible, and with a named lawyer responsible for playbook upkeep. Expect review time to rise for the first weeks as people check the AI. That is the trust-building cost and it is normal; a business case that has not budgeted for it will look like a failure at week six.
Once review is working, the back catalog becomes worth loading — obligations, renewal dates, unusual terms. This is where the repository camp's value actually arrives, and it is a separate project with a separate budget from the review tool.
Contract review AI is not high-risk in the regulatory tiering sense, and the interesting obligations are professional rather than statutory. A lawyer remains responsible for the advice regardless of what a tool produced, which makes explainability a duty rather than a feature: a flag nobody can trace is a flag a lawyer cannot rely on. Two data questions attach as well and should be settled before the pilot, not after. Contracts contain the counterparty's confidential information, which is usually subject to a confidentiality obligation you owe them, so where the text is processed and whether it trains a model are questions with a contractual answer rather than a preference. And privilege can attach to the surrounding legal analysis, which affects retention and access design.
Classified under professional responsibility for legal advice, together with confidentiality obligations owed to counterparties over contract text
What should you ask vendors about AI Contract Review & Intelligence?
The first question sorts this category cleanly, and evaluations that skip it tend to compare products that are not alternatives.
-
Can you say today what obligations and renewals you have signed up to?Yes Your problem is review speed. Shortlist reading tools and in-editor assistants, and do not migrate anything.No Your problem is the repository. No reading tool fixes it, and this is a CLM decision with AI attached.
-
Is most of your review volume on third-party paper?Yes Weight the evaluation entirely on inbound contracts from your own files. It is the hard case and the common one.No Your own template with redlines is a simpler problem, and more of the market will handle it well.
-
Will lawyers have to leave their editor to use it?Yes Expect adoption to be the binding constraint, whatever the capability. Ask for twelve-month weekly-active figures from a reference.No Adoption risk is much lower, which for this category is worth more than a marginal capability advantage.