How eDiscovery is priced, and why nobody can quote you a number

There is no honest eDiscovery price in the abstract. A price without a data profile is a guess: it assumes a collection volume, an error rate in the estimate, a review population, a production format, a retention period, and a level of service that the quoted parties may never have examined.

That is why a vendor’s headline unit is not a price. It is a statement about what the vendor is paid to do and, consequently, what the vendor is rewarded for doing more of. The same matter can look inexpensive under one model and expensive under another without either quote being fraudulent.

Any number offered before the data profile is settled is therefore withheld deliberately. A useful quote requires enough shared facts for the buyers and providers to evaluate the same work.

Common eDiscovery pricing units#

Pricing units do not describe interchangeable services. They allocate different parts of the work: ingesting data, retaining it, providing access to it, reviewing it, or managing the process around it. A comparison that places those units side by side without separating their functions is not a comparison.

Per-gigabyte ingestion can reward a vendor for moving every collected file into the review environment, even when many files will later be removed from the review population. Per-gigabyte hosting can preserve that apparent economy for as long as the data remains online. Per-document pricing appears more closely aligned with the burden of review, but its actual burden depends heavily on what enters the review population.

Per-user pricing can make a matter with substantial data look inexpensive when the team is small. It can become the wrong metric when a broad data set requires many reviewers or when user access remains open long after the active work has ended. Hourly professional-services pricing offers visibility into labor, but it also makes the quote sensitive to assumptions about staffing, discipline, and the number of problems encountered along the way.

No unit is inherently best. The stronger commercial fit follows the data profile. A bounded collection processed once may support ingestion pricing. A long-running matter with retained data makes hosting duration central. A tightly scoped review with reliable culling makes per-document pricing easier to test. Unusual sources and workflows may make hourly services the most transparent starting point, provided the estimate identifies the tasks and assumptions behind the labor.

How the same matter changes price under each unit#

Consider a collection containing structured application data, email, attachments, and compressed archives. Processing may expose documents and message families that were not apparent in the initial inventory. Those expanded contents increase the population available for review and may also increase the stored corpus.

Under ingestion pricing, the collection looks inexpensive because the quoted unit is applied once. Under hosting pricing, it becomes a recurring commitment that continues after active review has ended. Under per-user pricing, it may still look inexpensive if only a small team needs access. Under per-document pricing, it becomes expensive if de-duplication, threading, and scope decisions leave a large review population. Under hourly professional-services pricing, it becomes expensive if the provider must investigate unfamiliar containers, correct processing exceptions, and manage repeated decisions.

That divergence is not evidence that the vendor has priced the matter improperly. It shows that the vendors are pricing different portions of the lifecycle. A meaningful comparison either holds the commercial unit constant or translates each quote into a complete statement of what the price includes, what remains billable, and what happens when the matter changes.

Costs that commonly sit outside the headline#

Processing and reprocessing

A processing estimate may assume clean metadata, standard file types, and a stable field mapping. Real workflows can require additional normalization, extraction, analysis, or correction. If a scope decision changes, the affected set may have to be processed again. The quote should distinguish the initial processing scope from later reruns and identify which changes would trigger additional work.

Data expansion

Containers, archives, and embedded content can enlarge the search and review population beyond the shape of the original collection. An archive may contain files that are not individually visible in the initial inventory; a container may hold metadata that points to additional content; and extracted attachments may enter the workflow as separate review items.

A credible estimate therefore needs a collection profile that describes sources and formats, not merely a total data volume. “All custodians” is not a sufficient scope, and a file count alone does not establish how many review items will result after processing.

Hosting for the life of the matter

Hosting is the cost most likely to compound because it can recur after the collection and first production have been completed. The relevant questions include the minimum retention period, the charge for continued access, the treatment of archived or inactive matters, and whether the hosted data is reduced when its relevance ends.

The buyer should also obtain the terms governing export, transfer, and deletion. A low ingestion charge can be outweighed by months of retained capacity, particularly when early scoping and culling have not reduced the active corpus.

Overtime project management

A matter does not pause for weekends, production deadlines, or unresolved custodian issues. Project-management assumptions should distinguish ordinary coordination from work outside the expected operating window. The estimate should identify the included cadence of communication, the treatment of urgent issues, and which events are likely to require new approvals rather than routine administration.

Production reruns

A production can fail for reasons that have little to do with review accuracy. A wrong production format, an omitted family relationship, an incorrect redaction, or a metadata mapping error can require another production and another validation cycle. Under Federal Rule of Civil Procedure 34(b)(2)(B), a responding party must state the grounds for an objection with specificity, and under Rule 34(b)(2)(C), the objection must state whether responsive materials are withheld on that basis. Those requirements help frame disputes, but they do not make the operational cost of correcting a defective production disappear.

The quote should therefore identify the production format, whether load files are included, the validation steps, and whether a rerun is treated as an ordinary production, a correction, or separately charged work. Native, TIFF, and PDF outputs are not interchangeable commitments.

The cost of leaving

Switching providers late in a matter can create work that does not appear in the original estimate: migrating the corpus, transferring access and audit information, reconciling processing settings, preserving privilege and work-product designations, and revalidating the production history. The cost is not limited to vendor fees. It can include internal staff time and the risk that differences in processing or review configuration change the production set.

Early culling also affects what can be left behind. Data retained without a defined purpose may remain subject to hosting charges and preservation duties. Legal holds should follow the organization’s preservation obligations, not a vendor’s convenience, but the retention plan should distinguish material subject to a hold from data that is merely retained as a technical convenience.

Culling is the main pricing lever#

Culling is the process of reducing the population that requires expensive attention while preserving the material that must remain available for review, production, and analysis. It is not simply deletion. Its objective is to remove redundancy, immaterial material, and data outside the agreed scope before the billable review population takes shape.

De-duplication removes exact or otherwise redundant copies so reviewers do not assess the same content repeatedly. Email threading groups messages and attachments into reviewable conversations rather than treating each message as an isolated item. Near-duplicate detection identifies materially similar documents that exact de-duplication may miss. Each mechanism can affect the per-document price and, if the hosted set includes redundant material, the hosting price as well.

Date and custodian scoping reduce the population before or during processing. Search-term negotiation further reduces the set that reviewers must evaluate. Rule 26(b)(1) places the proportionality factors within the definition of the scope of discovery and adds consideration of the parties’ relative access to relevant information. Search terms and review limits should therefore be negotiated against the information available to the parties, rather than selected from a generic keyword list.

Rule 26(f) requires the parties to confer before the scheduling conference, and Rule 16(b) governs the scheduling order. Those provisions make early scope discussion the proper setting for testing search terms, data sources, custodians, and any technology-assisted review method before substantial processing and review costs accumulate.

Culling also requires quality control. A reduction in review volume is not meaningful if the removed set includes responsive material. The quote should identify what de-duplication and threading will do, how near-duplicate clusters are treated, who validates the scope, and how decisions are documented.

Technology-assisted review affects cost and risk#

TAR encompasses methods that assist a reviewer in identifying relevant material. Predictive coding applies statistical analysis to document content and decisions. CAL, or continuous active learning, uses reviewer decisions to update an assisted-ranking process during review. Boolean or keyword search supports more traditional query-driven retrieval. The pricing consequence depends on how the method is charged and what work remains for human reviewers.

In Da Silva Moore v. Publicis Groupe, Magistrate Judge Andrew Peck held that computer-assisted review was an acceptable way to search for relevant ESI in appropriate cases, with the parties’ agreement mattering; District Judge Andrew Carter affirmed in April 2012. The opinion is widely described as the first judicial opinion approving predictive coding.

In Rio Tinto PLC v. Vale S.A., Judge Peck described judicial willingness to permit technology-assisted review as black-letter law, and Maura Grossman was later appointed special master to resolve TAR disputes. In Hyles v. City of New York, Judge Peck held that a court would not force a producing party to use TAR when it preferred keyword searching, consistent with Sedona Principle 6’s recognition that the responding party is best situated to decide how to search for and produce its own ESI.

Some language in Rio Tinto and Hyles is dicta rather than holding, so it signals judicial attitude rather than a binding rule. Pricing discussions should not treat any of these decisions as a universal mandate. Rule 37(e) separately addresses lost ESI that should have been preserved, including prejudice-based curative measures and more severe consequences when intent to deprive another party is found. Cost control does not displace preservation duties.

Getting comparable vendor quotes#

A comparable quote begins with the same data profile supplied to every vendor. The profile should identify the custodians and data sources, relevant date ranges, collection formats, archives and containers, expected attachment and message relationships, file types, rough collection volume, known duplicates, required production formats, privilege and redaction needs, expected review population, hosting requirements, and the operational assumptions on which the estimate depends.

All vendors should then quote in the same unit. If a provider insists on a different structure, require a worked example on your numbers that translates the result into the common unit. The worked example should show how collection, processing, culling, review, hosting, project management, validation, and production contribute to the total.

Each quote should also state exclusions and triggers. Ask which activities are included, which are optional, what happens after the hosted data is culled, whether professional services are capped, and what would trigger a rerun. Without those terms, the quotes describe different scopes rather than competing prices.

The final comparison should preserve the distinction between price and cost. The per-unit charge is the price. The total cost of the matter includes internal staff time, avoidable processing, unnecessary hosting, repeated review, correction work, and the expense of changing direction later. A defensible eDiscovery budget is built from the data profile, not from a generic market number that pretends the matter has no shape.