
Quick Answer
Answer a weaponized ESI demand with a measured number, not a bare objection. Assess the real collection early, price review at AI-assisted rates, and take that figure to opposing counsel.
A 2023 podcast guest argued the parties usually settle these fights better than a judge. Numbers help. With Relevant e-Discovery you bring a messy collection and a hard question, then watch it read, code and cite your own evidence, with hashed originals and an append-only audit trail behind it. Review runs at cents per document, not dollars.
Key Points
- The 2015 Advisory Committee Note to Rule 26(b)(1) warns that a boilerplate proportionality objection is not enough, by itself, to refuse discovery.
- In the Summer 2024 ComplexDiscovery and EDRM pricing survey , 27.9% reported onsite review above $1.00 per document, while Relevant e-Discovery prices AI-assisted review at cents per document.
- Reed Smith partner David Cohen recommended a protective order against using public GenAI to review produced documents, so AI review terms belong in early negotiation.
An oversized ESI demand lands on a small firm's desk. The answer starts with measuring what is actually in the collection.
A lawyer who cannot measure an ESI collection turns defensive, because there is no defensible position to stand on. A needless fight over scope, one 2023 podcast guest warned, only produces another motion, and the client pays for the response.
I keep coming back to that idea. It was put plainly on a 2023 podcast episode about cooperation in e-discovery: lawyers who do not understand the technology get defensive because they lack a defensible position, and any representation about cost or data volume made before you have talked to your own client undermines trust. Both points land hard on a small firm. The demand arrives on a Tuesday. The client is anxious. You have no number yet.
That is the gap a weaponized request is built to exploit. Early case assessment, the practice of examining a collection before review to see what is actually in it and what reading it would cost, closes the gap with a figure instead of an adjective.
Keep the record of what a lawyer checked and changed, because, as Patrick Barone wrote from the criminal defense side in 2026 about AI-drafted police reports, "approval is not the same thing as authorship." Relevant e-Discovery was built around two things that matter here:
- Every answer traces to its source. Outputs link back to the exact exhibit they came from, so an attorney can verify before filing.
- Privileged evidence stays under your control. Processing runs single-tenant, or inside your own AWS account under your own keys, with no vendor retention and no model training.
Handing the problem to a vendor is the usual reflex. A 2015 trade piece observed that firms were often disappointed by vendor proposals, or by how the vendor performed once the work began. The cost of finding that out late is the first thing worth pricing.
In 2024, the share of pricing-survey respondents quoting remote collection under $250 an hour rose from 15.8% to 31.1%. Gathering data is getting cheaper. Sending an overbroad demand was always cheap.
A cost weapon is an ESI request drafted less to find evidence than to make answering it cost more than the case is worth. Small firms feel it first. The reflex is a Rule 26(b)(1) proportionality objection, and I think that reflex is incomplete, because an objection with no number behind it asks the judge to take your word on burden.
The tools to supply that number exist. Guests on BDO's Legal Tech Talk podcast estimated in 2023 that the criminal bar ran about 5 to 10 years behind the civil bar in using technology-assisted review and early case assessment to cull documents. Practitioners now report that Relativity no longer promotes linear review or TAR at all, only its generative aiR product. AI-assisted review is priced in cents per document rather than dollars.
Having those tools and pointing them at a hostile demand are different things. The first question is a plain one: what would it truly cost to read what they asked for? What follows starts with the number.
How much does e-discovery processing and review cost per gigabyte for a small civil case?
Relevant e-Discovery is built for a solo or small firm on one messy matter, and on that matter the review line, not the processing line, decides whether a demand hurts.
Before I answer any ESI demand, I price it in three plain pieces:
- Collection: what it costs to gather each custodian, laptop and phone.
- Processing and hosting: what it costs per gigabyte to make the data searchable and keep it live.
- Review: what it costs to have someone, or something, read it.
A 2013 Bloomberg Law piece put the arithmetic simply. Collect N bytes and you spend X dollars on collection, processing and hosting, then Y dollars on review over Z time. The same article said those estimates exist to set expectations with the judge and opposing counsel about how long the work reasonably takes. The formula has not aged.
The ComplexDiscovery and EDRM pricing survey supplies the inputs. In 2024, 59.0% of respondents put onsite forensic collection at $250 to $350 an hour. Devices were billed on top: 45.9% charged over $350 per desktop or laptop, and 47.5% charged over $350 per mobile device. Processing was cheaper than most lawyers expect. For collections over 251 GB, 63.9% reported completion costs under $100 per GB in 2024. Remote human review ran $25 to $40 an hour for 49.2% of respondents that year. The survey drew on 61 respondents and called itself non-scientific, a guide rather than a price list.
Then comes review, and review is the bill. Our own pricing shows manual review at dollars per document and roughly $19K per GB, while AI-assisted review runs cents per document. That gap is the whole game.
The common assumption is that volume alone makes a demand expensive. It does not. Volume becomes expensive only when every page is read by a person billing by the hour. An analysis of 3 sources points one way: the gigabyte charge is modest, and the reading is where the money goes.
So the cost weapon is aimed at one place. A demand letter costs its author an afternoon. Answering it by hand can cost a small firm the season, and I think most of us have felt that cold arithmetic at least once, late, with the office empty.
What this means for you is practical. A burden argument built on manual review rates concedes the other side's number. A burden argument built on what review actually costs on your collection is a different conversation, and I have written about how that shift can turn e-discovery from a cost center into case leverage. The next question is how to get that real number early enough to use it.
Does a proportionality objection or a measured number push an oversized ESI demand back?
When Rule 26 was amended in 2015 to put proportionality back into the scope of discovery, the drafters' note included a warning for lawyers who answer an oversized demand with boilerplate.
"Nor is the change intended to permit the opposing party to refuse discovery simply by making a boilerplate objection that it is not proportional."
Advisory Committee Note to Federal Rule of Civil Procedure 26(b)(1), 2015
A small firm facing a demand built to cost more than the case is worth cannot simply object. The same note says where the leverage sits: a party claiming undue burden "ordinarily has far better information," the committee wrote, "perhaps the only information." The requesting side "may have little information about the burden or expense of responding," and the note expects such gaps to be "addressed and reduced in the parties' Rule 26(f) conference."
Read that way, the rule hands the responding party a job: say what compliance will cost, and say it early. The remedy points the same way. Rule 26(c)(1)(B) lets a court order "the allocation of expenses," but only after the movant certifies a good-faith attempt to confer. Any shift in cost starts with a conversation.
Practitioners describe what that conversation should contain. A 2013 Bloomberg Law article for new associates reduced the estimate to a formula: "if we collect N bytes of information from our client, we will spend X dollars in collecting, processing and hosting fees and Y dollars to review the information in Z time frame." Such figures help lawyers "set expectations with the judge and opposing counsel." On a 2023 legal technology podcast, a defense lawyer said that within a couple of weeks of receiving a complaint, they call the plaintiff, name the five most relevant custodians and key data sources, and put that in the early joint status report. Hiding everything, the same lawyer warned, brings motions to compel and spending beyond plan. Another guest said forcing the other side to search two decades of backup tapes is very rarely necessary.
| Boilerplate objection | Measured burden estimate | |
|---|---|---|
| What the 2015 note says | Not enough, by itself, to refuse discovery | Supplies burden information the responding party ordinarily holds alone |
| What the other side learns | Nothing about the burden | Volume, cost, time, custodians and data sources |
| What it invites | Motions to compel, if it amounts to hiding data (practitioner view) | Scrutiny of the method: who supervised it and whether it can be rerun |
In a cost-driven demand, Y is the number to watch. In the Summer 2024 pricing survey by ComplexDiscovery and EDRM, 63.9% of respondents put completed processing above 251 GB under $100 per gigabyte, while 27.9% reported onsite review above $1.00 per document. Our own figures put AI-assisted review at cents per document, against roughly $19K per gigabyte for manual review. We sell that software, so test the gap on your own collection.
Once a number is on the table, opposing counsel will attack it. In a 2016 Bloomberg Law commentary, an executive at an e-discovery services company set the standard: "To be defensible, the legal team must be able to replicate the processes and steps they used and achieve the same results with the same raw set of documents." Commentary in The Legal Intelligencer on the recent Trailsedge case adds that counsel "cannot abdicate responsibility for ESI collection and must supervise search methodologies." An unsupervised, unrepeatable estimate hands opposing counsel a new target.
When the method holds up, a court has accepted technology as the proportionate answer to volume. In Pyrrho Investments v. MWB Property, one party faced 3.1 million files from backup tapes. On February 16, 2016, Master Matthews of the UK High Court approved technology-assisted review, relying on disclosure rules that support "automated methods of searching if a full review of each and every document would be unreasonable." The ruling rested on its own facts and one particular approach, so it shows what one court accepted and nothing broader.
Line these sources up and they point to one posture. The rule says the objection alone falls short and the burden figure is yours to supply. Practitioners say to supply it early, in writing, with named custodians and sources. The defensibility commentary says the figure is only as strong as a supervised method someone could rerun. Proportionality still matters, and a measured, repeatable number is what gives it weight. None of our sources counted how often that approach wins a narrower scope or a cost order, so treat it as stronger footing, with no guaranteed result.
Before you answer the next oversized demand
- Measure the collection first, then write down four numbers: volume, processing and hosting cost, review cost, and time.
- Bring those numbers, plus the custodians and data sources you see as most relevant, to the Rule 26(f) conference or the first joint status report.
- Record who supervised the search and how it ran, so someone else can rerun the estimate and get the same results.
- If the demand reaches backup tapes or millions of files, ask whether automated review would be the proportionate method, and raise it in writing.
- If talks stall, ask whether a protective order allocating expenses under Rule 26(c)(1)(B) fits your facts, and keep a record of your good-faith conferral.
How we checked this
We read the text of Federal Rule of Civil Procedure 26 and its 2015 Advisory Committee Note. We also used two Bloomberg Law articles, a pricing survey, a 2023 podcast episode and a Legaltech News summary of commentary on Trailsedge. We read that summary, not the Trailsedge opinion itself. The survey had 61 respondents, and its publisher calls it non-scientific. The podcast speakers are unnamed in the transcript. The 2016 commentary was written by an executive at an e-discovery services company, and the podcast comes from an advisory firm with its own e-discovery practice. The figures comparing AI-assisted and manual review are ours. We sell AI review software, so we have a stake in this answer. Pyrrho is a UK decision that rested on its own facts. Still unknown: how often a measured estimate actually wins a narrower scope or an allocation of costs. None of our sources tracked outcomes.
- Legal Information Institute, Federal Rule of Civil Procedure 26 and its Advisory Committee Notes, retrieved November 16, 2026.
- Bloomberg Law, article on eDiscovery for new associates, October 9, 2013.
- Bloomberg Law, commentary on analytics and the Pyrrho decision, March 25, 2016.
- ComplexDiscovery, Summer 2024 eDiscovery pricing market report, August 5, 2024.
- Legal Tech Talk podcast, episode on e-discovery cooperation produced by an accounting and advisory firm, June 21, 2023.
- Legaltech News, e-discovery coverage summarizing commentary on Trailsedge, October 20, 2026.
- Relevant e-Discovery, product figures comparing AI-assisted and manual review costs.
How do I use early case assessment to prove an ESI request is disproportionate?
Run early case assessment on the real collection, price review at what it actually costs with AI assistance rather than by hand, and put that number in writing early.
Early case assessment (ECA) is the work of sizing and sorting a collection before review begins: who holds the data, what dates it covers, and how much of it is likely to matter. A defense lawyer on BDO's Legal Tech Talk podcast in 2023 preferred another name for it, early case strategy. I think the second name is the truer one.
The workflow I recommend has three steps:
- Call the other side within the first couple of weeks. The same podcast guest described naming the five most relevant custodians and the key data sources, then putting them in the early joint status report.
- Load the collection and measure it. Count volume, custodians, date ranges, and how much survives deduplication and a first relevance pass.
- Price review two ways and show both. One figure assumes a person reads every page. The other reflects AI-assisted review, which runs at a fraction of the manual cost on our own pricing.
Step two must stand on a sound collection. The questions are old ones: were the right custodians interviewed, was the time frame accurate, and were the methods defensible. That is also why forensic imaging is not for your IT person.
The conventional wisdom says to file a Rule 26(b)(1) objection and wait. Start with a custodian questionnaire, the kind a 2023 podcast guest recommended, which records how each client uses email and whether they run Office 365. Adjectives invite motions to compel. The BDO guest said as much in 2023: hiding everything leads to motion practice, and you spend more than you planned. Another speaker on that episode said it is very rarely necessary to make the other side search two decades of backup tapes, and that discovery is the largest contributor to litigation expense.
Ask whoever runs the assessment for its log. A sound one keeps a plain record of what the software set aside, how sure it had to be before it did, and which lawyer read the borderline documents. ECA gives you that kind of record for the burden itself, not just for the review.
I would like to give you our average culling rate by matter size here. We have not published one, and I will not guess at it. What the measurement gives you instead is your own figure, on your own matter, and a judge will weigh that more heavily than any industry average.
A number on a page is cheap to make and hard to argue with. In 2023, one of the podcast guests put settlement at 90% of these cases, which means the early figure often does its work before anyone files a motion. What the other side does next, when it can no longer fight the size of the burden, is go after the process that produced it.
How do I respond when opposing counsel challenges the defensibility of my ESI collection?
Answer a defensibility challenge with records, not reassurance: show who supervised the search, how the method ran, what was hashed and logged, and where each AI answer came from.
Once the cost lever fails, the attack moves to process. Patricia E. Antezana, writing in The Legal Intelligencer on October 20, 2026, read the Trailsedge case to mean that counsel cannot abdicate responsibility for ESI collection and must supervise search methodology. The same piece said best practice is a reasonable inquiry, early, into the client's retention policies, systems and ESI sources. The duty sits with you. Not with the vendor. Not with the software.
AI adds a second line of attack. Kelly A. Lavelle, in a November 5, 2026 analysis, described courts drawing distinctions between attorney-directed AI use and a client's own research, and between privilege and the broader work-product doctrine. Her reading was that consumer AI platforms may defeat confidentiality under their privacy policies, while work product can survive AI assistance if it was made for litigation and kept from adversaries.
Producing parties have noticed. Doug Austin of eDiscovery Today reported that Reed Smith partner David Cohen closed a Legalweek panel by recommending a protective order against using public GenAI to review produced documents. After a reader objected, Austin narrowed the advice to public models only, not every AI tool. For a small firm, that narrowing matters a great deal. A blanket ban on AI review would push you straight back into the manual cost the demand was built to impose.
I keep three records ready before anyone asks:
- Supervision: who set the scope and search terms, and when they were approved.
- Integrity: hashed originals, an append-only audit trail, and a documented chain of custody.
- Source: every AI-generated answer tied to the exhibit it came from, checked by an attorney before filing.
Relevant builds the second and third into the platform, with immutable originals, a fail-closed privilege gate on production, and outputs that link back to the exact exhibit for one-click verification. Processing that runs in your own cloud account, with no vendor retention and no model training, answers the confidentiality worry before it is raised.
People often ask whether AI issue coding or technology-assisted review is easier to defend. I have not seen evidence that courts rank one above the other. What they test is supervision and the record. If the dispute narrows to search terms, a 5-point search equivalence test gives you a disciplined place to start.
A record is a quiet thing. It sits in a log for months, unread, until the afternoon someone across the table asks how you know, and then it is the only thing in the room that does not need to be believed. Which leaves the other temptation small firms face when the boxes arrive: handing the whole thing to someone else.
What should a small firm do the next time a demand arrives heavier than the case?
Ask your client which systems hold the data, then get a per-document review price in writing before the Rule 26(f) conference. Keep each draft estimate and who changed it.
I think the weapon is losing its edge. A guest on a 2023 legal tech podcast warned that collected data, piling up without efficient ways to analyze it, was "going to come to a head." Phones hold more every year. The demands grow with them.
What changes the arithmetic is the price of reading. When review runs cents a document rather than dollars, the old threat of a ruinous review turns into a figure you can check, and a figure you can check is one you can answer in a letter. Relevant e-Discovery shows that gap on your own matter, so it is something you see rather than something you are told.
Contrary to the habit of filing first, order matters here. Run the assessment. Write the objection around what it found. Then let the other side explain, in front of the judge, why its demand still costs what it said.
Summarize This Article With AI
Open this article in your preferred AI engine for an instant summary.
Frequently Asked Questions
What else do small firms ask when an ESI demand is built to cost them?
Most follow-up questions are about price: processing, review method, vendors and fees. Each one is easier to answer once the collection has been measured rather than guessed at.
What does processing a large collection cost before anyone reads a page?
Less than many burden letters imply. For volumes over 251 GB, a 2024 pricing survey found 49.2% of respondents reporting ingestion under $25 per GB, up from 32.9% earlier that year. Processing is the cheap part. The reading is where a demand does its damage.
Is technology-assisted review cheaper than reading every document?
Technology-assisted review (TAR) means software ranks documents by likely relevance so people read the strongest candidates first. In 2024, 36.1% of the same survey's respondents priced it under $75 per GB for datasets under 250 GB, and 42.6% did for larger sets. Onsite human review, by contrast, ran over $1.00 a document for 27.9%.
Should I just hand the whole collection to a full-service review vendor?
I would be slow to. Practitioners in an e-discovery forum reported in 2025 that many review vendors were owned by private equity and some were offshoring review to India. State bars, they said, claimed jurisdiction only over lawyers and law firms. Your duty to supervise the search does not travel with the hard drive.
Is AI-assisted review always the cheaper choice?
Not always. One practitioner in late 2025 called it "a bit of an undertaking and pretty expensive" and said not every case was a good fit. Another expected Relativity's aiR to be folded into hosting from April 2026, which would change that arithmetic for platform users.
Do project-management hours add much to a burden estimate?
They add hours, and those hours belong in the number. In 2024, the share of providers charging $100 to $200 an hour for basic project support rose from 56.6% to 70.5%. Leave them out and the estimate looks thin when the other side checks it.
What does Relevant e-Discovery charge for review?
AI-assisted review is priced at cents per document, not the dollars per document that manual review costs. You see the difference on your own matter, not on a slide. Reach the team through the contact page at relevantediscovery.com, and bring the collection that has been keeping you up late.
Written by
Michael
Kansky
Michael Kansky is a serial software entrepreneur who has spent more than two decades building and bootstrapping profitable SaaS and services companies.
Connect on LinkedIn

