AI E-Discovery for Small US Law Firms: Affordable Tools Compared (2026)

Last updated: October 2026 · Reading time: about 13 minutes

The three numbers, not one

Most e-discovery comparisons compare headline per-GB rates. That is the wrong number, and it is why small firms end up surprised by the invoice.

Every platform charges for three distinct things:

• Hosting — a recurring monthly charge on whatever data is currently sitting in the workspace. You pay this every month the matter stays open.
• Processing — a one-time charge on data as it is ingested. This covers OCR, deduplication and normalisation.
• AI review — a third charge on documents actually pushed through the AI workflow. Sometimes included, sometimes per-GB, sometimes per-document.

The trap is that vendors collapse these three differently. Relativity keeps them maximally itemised — three separate per-GB line items. Everlaw folds all three into an annual per-seat fee. Logikcull folds them into flat per-GB SaaS. Disco folds them into a blended per-matter rate.

Which means a firm comparing "Relativity is $17.50/GB and Logikcull is $30/GB" is comparing a hosting rate against an all-in rate. Those numbers are not the same kind of number.

Four pricing models, one matter

The clearest way to see the problem is to price the same matter four ways. Take a 500 GB matter running six months with a ten-person review team.

Read that table again and notice what it is actually saying: the same matter, the same data, the same six months — and an eight-fold spread in platform cost. None of those vendors is lying about its rate. They are selling different shapes of contract, and the shape matters more than the rate.

For a small firm, that produces a simple decision rule:

The tools actually built for small firms

Here is the tier almost no mainstream comparison covers, because the mainstream comparisons are written for Am Law 200 firms. These are the platforms with published prices and small-firm economics.

Two things stand out about this tier.

First, the pricing transparency is the differentiator. GoldFynch is one of very few e-discovery tools where you can price a matter before you talk to anyone[reference:27]. For a firm that has never bought e-discovery software before, that is worth more than a feature comparison — it means you can model a matter without a sales call.

Second, the model is per-case, not per-seat. That matters enormously for a small firm, because you are not paying for the reviewer who logs in for two hours on a Friday. A 3 GB case costs $27 whether it holds 0 GB or 3 GB, and whether one person or five access it[reference:28].

The mid-tier options

Once your matters exceed roughly 25 GB, the small-firm tier starts to feel constrained — usually on AI depth rather than on capacity. That is where the mid-tier platforms come in.

Logikcull (now part of Reveal)

Logikcull kept its self-service identity after the Reveal acquisition. It is the only major platform where you can sign up and start uploading a 25 GB custodian collection immediately, with flat per-GB pricing and no seat fees on the entry tier. Reported rates run $20–$50 per GB depending on volume and contract length.

The honest limitation: Logikcull's AI features are genuinely useful but not at parity with Relativity's aiR or Everlaw AI for complex privilege review. It works best for investigations and pre-litigation matters rather than bet-the-company cases.

Everlaw

Everlaw is a browser-native platform with collaborative review and AI included in every seat. It sells direct, prices per seat per year, and includes hosting and processing in the seat cost up to a reasonable data threshold. Verified 2026 figures put it at roughly $3,000–$12,000 per seat per year depending on tier and volume commitment.

The catch: the per-seat model becomes punitive if you need 40 reviewers for two months. Everlaw does offer flex licensing, but it is rarely publicised.

Disco

Disco's blended per-matter model is the best fit for unpredictable matter loads. A boutique that takes on a sudden 800 GB matter does not need to pre-purchase anything. All-in rates typically run $10–$30 per processed GB, with hosting charged separately after the first 90 days.

The catch: the platform is genuinely cheaper than Relativity, but total project cost closes the gap once you add labour — Disco bills review-time hours separately if you use its bundled review services.

Why Relativity isn't on the table

Relativity is the reference platform for complex litigation, and its aiR add-on is a genuine generative review tool rather than a predictive-coding wrapper. But it is not a small-firm product, for a structural reason rather than a pricing one.

Relativity does not sell direct to law firms below a certain size. You buy through a Relativity Certified Partner — a service bureau like Lighthouse, Consilio or Epiq — which means the headline per-GB figure is rarely what you actually pay. The partner marks it up and bundles project management on top.

That has two consequences a small firm should understand before evaluating it:

• You are buying a partner relationship, not software. The partner's project management fee is part of the cost, and it is not optional in the way it is on a self-service platform.
• The economics only work at volume. Entry-level partner programmes start around $25,000 per year, and a fully-loaded enterprise deployment runs $250,000+ annually. For a firm with one or two matters a year, that floor is the whole budget.

Which is not a criticism of Relativity. It is the right tool for multi-district litigation and regulatory matters with hundreds of thousands of documents. It is simply not the right tool for a five-attorney firm with a 40 GB contract dispute — and the comparison articles that put it in a table alongside GoldFynch are not helping anyone.

The ethical flip: competence is now the argument

Here is the part that changes the calculus for a small firm from "can we afford this" to "can we afford not to."

The cost of manual review has collapsed. RAND priced document review at $18,000 per gigabyte in 2012. The 2026 EDRM Winter Pricing Survey puts generative-AI-assisted review at $0.11 to $0.50 per document[reference:40]. That is not an incremental improvement — it is a structural shift in how review is priced.

And the adoption gap is stark. Only 17.7% of solo practitioners use AI tools for any part of their work, against 47.8% at firms with 500 or more lawyers[reference:41].

That gap now has ethical teeth, for three reasons.

1. Technological competence is mandated. Forty-two jurisdictions — 40 states plus D.C. and Puerto Rico — now require technological competence under ABA Model Rule 1.1, Comment 8[reference:42].

2. The sanctions are real and personal. In DR Distributors v. 21 Century Smoking, Judge Iain Johnston split $2.5 million in fees and held the attorneys personally liable for over $1.25 million because they lacked what he called "the basic knowledge, training, and skills to handle properly ESI"[reference:43]. That was 2021, and the bar has risen since.

3. Refusing the technology can weaken your proportionality argument. Under FRCP 26(b)(1), discovery must be proportional. A court that sees you choosing a $200,000 manual review when AI-assisted alternatives exist at a fraction of the cost is unlikely to sympathise when you argue the request is disproportionate[reference:44].

The case law on technology-assisted review is settled, and it has been for over a decade. Da Silva Moore (2012) was the first published decision approving TAR. Rio Tinto (2015) declared it "now black letter law," and established two points that matter especially for small firms: TAR cannot be held to a higher standard than manual review, and the producing party chooses the methodology unilaterally[reference:45].

Hyles v. New York City (2016) set the standard that matters most for a resource-constrained firm: "the standard is not perfection, or using the 'best' tool, but whether the search results are reasonable and proportional"[reference:46]. That is a lower bar than most small firms assume.

A defensible small-firm workflow

Affordability is only half the question. The other half is whether the process survives a challenge. Five steps, in order.

1. Preserve and collect defensibly first. The DR Distributors sanctions were triggered by preservation and collection failures, not by review methodology[reference:47]. No AI tool fixes a bad hold.
2. Pre-cull before running AI. Livingston v. City of Chicago endorsed keyword pre-culling on a 1.5 million document collection — filtering down to 192,000 before applying Relativity's Active Learning. Narrow first, then apply AI. That is the workflow that cuts cost[reference:48].
3. Run AI-assisted categorisation, not just keyword search. Modern tools answer natural-language queries with results anchored to specific passages in the source documents. That anchoring is what makes the output reviewable[reference:49].
4. Validate with a documented sample. The In re Broiler Chicken protocol provides a court-approved template: a 3,000-document validation sample across three strata — responsive, non-responsive, and AI-excluded. It is a replicable process you can follow step by step[reference:50].
5. Document the process as you go. Courts expect a reasonable, proportional, documented review. Not a perfect one[reference:51]. If your tool exports its reasoning to case notes, use that — it is the record that answers a challenge.

Two supervision points from ABA Formal Opinion 512 that bear directly on this workflow. First, the competence standard is lower than most attorneys assume — Opinion 512 requires "a reasonable understanding of the capabilities and risks of any GenAI tool" you use, or the ability to draw on someone else's expertise. Second, the supervision duty treats AI much like an unsupervised paralegal: you would not let a paralegal sign off on privilege calls, and the same logic applies[reference:53].

For the broader compliance picture — BAAs, SOC 2 reports, subprocessor chains — see our guide to HIPAA and SOC 2 compliance for legal AI tools, and for the accuracy-versus-traceability question in medical records specifically, our guide to AI medical record summaries.

What to ask before you sign

1. Which of the three charges does this quote include? Hosting, processing, AI review — get all three confirmed in writing. A quote that names only one is not a quote.
2. Is the rate per GB per month, or per GB one-time? A $20/GB one-time processing fee and a $20/GB/month hosting fee differ by a factor of six on a six-month matter.
3. What is the data quota, and what happens above it? Per-seat models include a quota — typically 25 to 100 GB per seat depending on tier. Overage rates are where the budget breaks.
4. Does the AI review include source citations? If the output does not link to the specific page in the source document, supervision becomes impractical and the review is hard to defend.
5. Is there a minimum term or annual commitment? Logikcull's entry tier has no annual minimum; Everlaw effectively does.
6. Can you export everything on termination, in a documented format? And at whose cost. This is the same exit-terms question worth asking of any legal software.
7. What is the data residency and security posture? SOC 2 Type II, HIPAA where relevant, and the specific jurisdictions your data can sit in.

For the printable version of the vendor due-diligence list, use the Legal AI Vendor Compliance Checklist.

Frequently asked questions

What is the cheapest AI e-discovery tool for a small law firm?

On published pricing, GoldFynch starts at $27 per month for a 3 GB case with unlimited users, processing and OCR included. DocumentReview.Law is $50 per month with no data hosting at all, so there are no storage fees. Which is cheaper depends on how long your matters stay open — GoldFynch is per-case per-month, so a case open for 18 months costs 18 times the monthly rate.

Why is e-discovery pricing so different between vendors?

Because they collapse three separate charges — hosting, processing and AI review — into different contract shapes. Relativity itemises all three, Everlaw folds them into a per-seat fee, Logikcull into flat per-GB, and Disco into a blended per-matter rate. The same 500 GB matter can cost roughly $17,500 on one model and $143,750 on another before labour.

Can a small firm use Relativity?

Not economically. Relativity does not sell direct below a certain firm size — you buy through a certified partner, which marks up the headline rate and bundles project management. Entry-level partner programmes start around $25,000 per year, and fully-loaded enterprise deployments run $250,000+ annually.

Is AI-assisted document review defensible in court?

Yes, and it has been for over a decade. Da Silva Moore (2012) was the first published decision approving technology-assisted review. Rio Tinto (2015) called it "black letter law" and held that TAR cannot be held to a higher standard than manual review. The standard courts apply is whether the search results are reasonable and proportional, not whether they are perfect.

What does the 2026 EDRM pricing survey say about AI review costs?

It puts generative-AI-assisted review at $0.11 to $0.50 per document, against RAND's 2012 figure of $18,000 per gigabyte for manual review. That is a structural shift in the pricing of review, not an incremental improvement.

How much data does a typical small-firm matter involve?

It varies enormously, but the small-firm tier of tools is built around cases of 3 GB to 100 GB. GoldFynch's published tiers run from 3 GB at $27/month to 100 GB at $652/month, with 50 GB at $342/month as a common mid-point[reference:64]. Matters above roughly 300 GB generally move into mid-tier platform territory.

Do I still need a human to review documents if I use AI?

Yes. ABA Formal Opinion 512 treats AI supervision much like supervising a non-lawyer assistant — you would not let an unsupervised paralegal make privilege calls, and the same logic applies[reference:65]. What changes is the volume of documents a human reviews: AI narrows the corpus, and the human reviews the narrowed set plus a validation sample.

Does using AI e-discovery reduce my billable hours?

It reduces the hours needed, which is a different thing. Under ABA Opinion 512 you cannot charge clients for time spent learning a generic AI tool, and you cannot bill for time the AI saved you. If per-matter AI costs are hard to isolate, treat them as overhead rather than passing them through.

What is the minimum I need to do to be defensible?

Three things: preserve and collect defensibly (the DR Distributors sanctions came from preservation failures, not review methodology), pre-cull before applying AI (endorsed in Livingston), and validate with a documented sample (the In re Broiler Chicken protocol provides a court-approved template)[reference:67][reference:68].

Related reading on Dedon Tech: best AI legal software for US personal injury firms · Clio vs Filevine vs Litify · AI medical record summaries: accuracy and traceability · Legal AI Vendor Compliance Checklist · full site map