Skip to content

5 platforms priced · Last updated July 2026

Legal Document Review Software for Law Firms: AI Document Review, eDiscovery Platforms and 2026 Costs Compared

Legal document review software is the platform a litigation team uses to load, cull, search, code and produce the documents collected in a case. Unlike almost everything else a law firm buys, it is priced by data volume rather than by seat: 2026 pricing surveys put processing at roughly $3 to $10 per gigabyte, hosting at $5 to $15 per gigabyte per month, and review and production at $15 to $30 per gigabyte. The reason firms care is that review is the largest single line in a document intensive matter, commonly reported at 60 to 80 percent of total project spend. This page compares five platforms on 2026 pricing, works through what a real 100 GB matter actually costs, and covers technology assisted review and what makes an AI assisted review defensible.

Run the live demo
Exhibit A The short answer

Why review, not collection, is where litigation budgets die

Collecting and processing data is a fixed, predictable cost. Reading it is not. Every hour of human attention on a document set is billed, and the set is always bigger than anyone estimated at the Rule 26(f) conference.

01 · THE SHARE

60 to 80 percent

The portion of total eDiscovery project spend that review alone consumes on a document intensive matter, according to industry cost analyses. Software licensing is a rounding error next to it.

02 · THE RATE

About 50 docs an hour

The average throughput of an experienced reviewer on straightforward responsiveness coding. Privilege and issue tagging run considerably slower. This number sets your whole budget.

03 · THE LABOR

$50+ an hour

Typical contract attorney review rates, with onshore US reviewers priced meaningfully above offshore. Partner time on the same documents is priced in multiples, which is why nobody wants partners in the review tool.

04 · THE UNIT

Per gigabyte

Nearly every platform in this category bills by data volume, monthly, for as long as the matter is open. A case that settles in year three has paid hosting thirty six times.

Exhibit B The decision

Four different products are sold as document review software

Shortlists in this category go wrong because buyers compare an eDiscovery platform against a contract AI tool against a filing system. Pick the shape first. Only then does a vendor comparison mean anything.

01
Litigation review platforms. Relativity, Everlaw, DISCO, Logikcull and Reveal. You load a collected data set, deduplicate and cull it, run search and machine learning over it, code documents for responsiveness and privilege, and produce a Bates numbered export with a privilege log. This is what most people mean by document review software, and it is billed per gigabyte.
EDISCOVERY
02
Contract review AI. Tools that read an agreement you are about to sign, flag deviations from your playbook and suggest redlines. Completely different job from litigation review: one document at a time, forward looking, no production obligation. If this is what you actually want, see our guide to AI contract review software instead.
CONTRACTS
03
Document management systems. iManage, NetDocuments and the rest hold the firm's own working files with versioning and permissions. They are permanent infrastructure, not a per matter workspace, and they do not produce. The distinction matters at budget time and is covered in our legal document management software guide.
STORAGE
04
Agentic review inside the case file. The newest shape, and where Caseagent sits. Rather than hosting an opponent's data set for production, an agent reads what actually lands on a matter, pleadings, orders, correspondence, records, and extracts the obligations, dates and facts a lawyer needs. It does not replace an eDiscovery platform on a big production. It removes the reading on everything else.
CASE FILE
Exhibit C The chart

Legal document review platforms compared, with 2026 pricing

Every vendor in this category negotiates, and most publish nothing. The figures below are the ranges reported by 2026 eDiscovery pricing surveys and cost trackers, not vendor list prices. Treat them as a sanity check on a quote, never as the quote itself.

Platform Shape AI review Reported 2026 cost Best for
Relativity (RelativityOne) The enterprise standard, cloud or on premise, huge partner and vendor ecosystem. Mature TAR plus generative assistants; the workflow most service providers already staff for Commonly reported around $50 to $150 per GB per month, with annual commitments in the tens of thousands Large firms, regulatory matters, anything with a service provider already in place
Everlaw Cloud native review with a markedly better reviewer interface and story building tools. Predictive coding, clustering and generative summaries built into the review pane Reported around $2,000 to $5,000 per month base plus roughly $18 to $35 per GB hosted Mid size firms running complex litigation in the 10 to 100 GB range
DISCO Speed focused cloud platform, the most aggressive AI first positioning in the category. Cecilia, marketed as an agentic reviewer that codes documents with stated reasoning Reported around $15 to $25 per GB per month hosting, with processing charged separately at roughly $20 to $50 per GB Litigation teams willing to lean on AI coding to cut reviewer hours
Logikcull Self service, upload and go, deliberately minimal configuration. Automated culling, deduplication and search; lighter on advanced TAR than the platforms above Reported from around $25 per GB with no minimum commitment Solos and small firms with occasional matters and no eDiscovery staff
Reveal Analytics led platform assembled from several acquisitions, strong visualization. Machine learning and generative AI wired directly into the review workflow Quote only; positioned with the enterprise platforms rather than the self service tier Investigations and matters where finding the story matters more than raw throughput

Two structural warnings. First, per seat quotes exist in this market too, commonly $150 to $250 per user per month and above $400 for enterprise analytics tiers, and a vendor will quote whichever model looks cheaper for your stated matter, so model both. Second, hosting is a recurring monthly charge on data that sits there whether anyone opens it or not. Ask what it costs to archive a closed matter before you sign, not after.

Exhibit D The arithmetic

What a 100 GB matter actually costs, line by line

Vendors quote per gigabyte because it sounds small. Here is the same number carried through a mid size matter, using the midpoint of the 2026 survey ranges and an eighteen month case life. Your numbers will differ; the shape will not.

Line Basis Midpoint rate Cost
Processing 100 GB, charged once on ingest $6.50 per GB $650
Hosting 100 GB per month for 18 months $10 per GB per month $18,000
Human review Roughly 100,000 documents at 50 per hour, so about 2,000 reviewer hours $55 per hour $110,000
Production Export, Bates numbering, privilege log preparation Varies by volume produced Low thousands
Same matter with TAR culling 60 percent before review About 800 reviewer hours instead of 2,000 $55 per hour $44,000

Read the two review lines against each other and the entire buying decision falls out. The platform fee is under twenty thousand dollars; the reading is over a hundred thousand. A tool that reduces reviewer hours by half is worth more than a tool that halves its own per gigabyte rate. This is why every serious vendor now sells AI review rather than storage, and why the only demo question that matters is how much of the set the tool can defensibly take off your reviewers.

Exhibit E The defensibility question

Technology assisted review, and what makes it hold up

The objection every buyer raises is whether a judge will accept it. That fight was settled years ago. The current fight is about process documentation, which is where firms actually lose.

Technology assisted review, also called predictive coding, works by having senior lawyers code a training set, then letting a model apply that judgment across the full collection and rank everything by likely responsiveness. Reviewers work down the ranked list and stop when the yield falls below a defensible threshold. Federal courts have accepted the approach since 2012, and by the time of Rio Tinto v. Vale in 2015 the court observed that using TAR was no longer controversial. Most federal judges will approve a reasonable TAR protocol today without argument.

What still gets challenged is the process around it. Proportionality under Federal Rule of Civil Procedure 26(b)(1) means you have to be able to justify the scope of what you did and did not review, and a Rule 502(d) order is the standard protection against a privileged document slipping into a production. Neither of those is a software feature.

In practice, a review survives challenge when you can hand over four things: the search and review protocol agreed with opposing counsel, the composition of the training and control sets, recall and precision estimates from a validation sample, and a privilege log consistent with the coding decisions in the system. Firms that get into trouble almost never lost because they used machine learning. They lost because nobody wrote down what they did while they were doing it.

Exhibit F How it works

How to evaluate a document review platform in one week

Run this with a real, messy data set from a closed matter. Vendor demo data is curated and tells you nothing.

STEP 01

Model the whole matter, not the rate

Take your realistic gigabyte volume and your realistic case duration, then ask each vendor for a total. Include processing, monthly hosting to the end of the case, user fees, production and the cost of archiving at close.

STEP 02

Load your own ugly data

Give it the mixed mailbox with the nested attachments, the scanned exhibits and the phone extraction. How a platform handles bad input is the difference between the demo and the matter.

STEP 03

Test the AI against known answers

Use a set you already coded. Ask the tool to rank it and check how much of your known responsive material it surfaces early, and what it wrongly buries. Ask the vendor to state recall, not adjectives.

STEP 04

Produce something and check it

Run a full export with Bates numbering, redactions and a privilege log, then have someone who did not build it check the load file. Production defects are discovered at the worst possible moment.

Exhibit G Use cases

Which tier fits which firm

Volume, not firm size, decides this. A three lawyer firm with one document heavy case has a bigger review problem than a forty lawyer firm doing transactional work.

Under 10 GB, occasional matters

Self service, no minimum commitment, no eDiscovery staff. You want upload, cull, search, review and produce with as little configuration as possible, and you want the meter to stop when the case closes.

10 to 100 GB, regular litigation

This is where TAR starts paying for itself and where reviewer interface quality shows up in your bill. Mid market cloud platforms are built for exactly this band, and the annual commitment is usually worth negotiating.

Over 100 GB or regulatory

Second requests, multi district litigation and investigations. Here the deciding factor is often which platform your service provider and opposing counsel already run, because interoperability beats features.

Personal injury and plaintiff work

The volume is medical records rather than custodial email, and the job is chronology building, not responsiveness coding. See case management for personal injury firms for the workflow that actually fits.

In house legal departments

You are usually buying to control outside counsel spend rather than to run review yourself. Pair the platform decision with legal matter management software so the budget and the data live together.

Firms with no production obligation

If you never produce, you do not need this category at all. What you need is a way to read what arrives on a matter faster, which is a case file problem and a much cheaper one.

Exhibit H Asked and answered

Legal document review software, common questions

The questions US litigation teams search for most before they buy, answered directly.

What is legal document review software?

Legal document review software is the platform a litigation team uses to load, search, tag and produce the documents collected in a case. It ingests email and files, removes duplicates, applies search and machine learning to rank what is likely relevant, gives reviewers a coding interface for responsiveness and privilege, and produces a defensible export with a privilege log. In most of the market it is priced by data volume rather than by seat, which makes it behave unlike any other software a firm buys.

How much does legal document review software cost?

Most platforms bill by gigabyte. Pricing surveys for 2026 put processing at roughly $3 to $10 per GB one time, hosting at $5 to $15 per GB per month, and review and production work at $15 to $30 per GB. Self service tools start around $25 per GB with no minimum, while enterprise platforms carry annual commitments in the tens of thousands. Per seat quotes also exist, commonly $150 to $250 per user per month, so model both structures against your real volume.

What is the best AI for legal document review?

It depends on data volume, not on firm prestige. Under about 10 GB, self service tools like Logikcull are usually the cheapest defensible route. Between 10 and 100 GB of complex litigation, Everlaw and DISCO are the common mid market choices and both ship generative review assistants. Above that, or on regulatory matters, Relativity remains the platform most service providers and opposing counsel already run, and interoperability tends to outweigh feature differences.

Can AI do document review?

AI can do most of the first pass, but not the final call. Machine learning reliably ranks and clusters documents so reviewers see likely relevant material first, and generative models now draft issue summaries and suggest coding decisions. A lawyer still validates the sample, makes privilege determinations and signs the production. Courts accept AI assisted review. What they expect is that you can explain and validate the process you used.

What is technology assisted review?

Technology assisted review, or TAR, is the use of machine learning to classify documents based on coding decisions made by human reviewers on a training set. The model learns what responsive means in your specific matter and applies that judgment across the whole collection, ranking everything by likelihood. It is also called predictive coding. Federal courts have accepted it since 2012, and Rio Tinto v. Vale in 2015 confirmed its use was no longer controversial.

Is AI document review defensible in court?

Yes, when the process is documented and validated. Defensibility does not come from the tool. It comes from being able to show a judge your search and review protocol, the composition of your training and control sets, recall and precision estimates from a validation sample, and a privilege log consistent with your coding. Firms that lose these fights almost always lost on process documentation rather than on the technology they chose.

How many documents can a lawyer review per hour?

An experienced reviewer averages about 50 documents per hour on straightforward responsiveness coding, and materially fewer on privilege or issue tagging. At that rate a 100,000 document collection is roughly 2,000 reviewer hours. That arithmetic, not the software license, is why review is consistently reported as 60 to 80 percent of total spend on a document intensive matter, and why culling before review is the highest leverage decision in the process.

What is the difference between document review software and document management software?

Document management software is where your firm stores its own working files day to day, with versioning, permissions and firm wide search. Document review software is a temporary workspace for the evidence in one case: it ingests a data set, deduplicates it, supports coded review for responsiveness and privilege, and produces a Bates numbered export. One is your filing cabinet and it runs forever. The other is a litigation workflow you spin up and shut down.

Exhibit I Being straight

Where Caseagent fits, and where it does not

Better to draw the boundary here than to have you find it halfway through a production.

Caseagent is not an eDiscovery platform. We do not host collected data by the gigabyte, we do not run TAR against a custodial collection, and we do not generate Bates numbered productions or load files. If you have a discovery obligation on a large data set, buy one of the platforms above. Nothing on this page is an argument against that.

What the agent does is the reading problem on everything that is not a production. Pleadings, orders, correspondence, medical records, expert reports and the rest arrive on a matter continuously, and someone bills time to read each one and work out what it changes. The agent sits inside the case file, reads what lands, extracts the dates, obligations and facts, and drafts what comes next for a lawyer to check. That is a different job from responsiveness coding, and it is the job most firms are actually spending their unbillable hours on.

The pairing that makes sense for a litigation practice is a review platform for discovery and an agent for the case file. Caseagent is in early access, launching 2026, priced per firm rather than per gigabyte.

Exhibit J Keep reading

Related buying guides

The rest of the document and litigation stack for a US law firm.

Legal document management software

Where the firm's own files live: iManage, NetDocuments and the rest, priced and compared.

AI contract review software

The other half of AI reading: playbook deviations and redlines on agreements you are about to sign.

Legal document automation software

Generating documents rather than reading them, from templates and matter data.

Legal practice management software

The pillar guide to the platform layer that runs the rest of the firm, with every major vendor compared.

Legal research software

Reading the law rather than the record, and where AI research tools currently stand.

Best legal AI tools

The wider AI landscape for US firms, from research to drafting to review, with honest limits.

EARLY ACCESS · 2026

Discovery is one reading problem. The case file is the other.

Caseagent reads what lands on a matter, pulls out the dates and obligations, and drafts what comes next for a lawyer to review. Early access for US firms; early users lock in launch pricing.

Run the demo first