Operating Note 01 · Brendan T. Walden · September 2026

How much does discovery cost? How much per document?

Discovery cost rarely shows up as a single usable operating measure. This note explains three different cost-per-document ($/doc) metrics—and which one gives a team the clearest view of matter-level cost.

Discovery is one of the largest and least visible cost drivers in litigation, and its effects on team capacity and matter control often appear before its full cost. Yet discovery spend rarely appears as a single operating measure. It is distributed across vendors, technology, review, processing, hosting, production, and, where included, internal operating resources. When those sources remain spread across systems and reporting cycles, the team cannot measure—or manage—the defined discovery operation as a whole.

Discovery was not designed as a functional system. It emerged as a collection of necessary processes and technologies—not an integrated operating discipline. Work moves across teams and technologies, while responsibility for the overall operating picture remains unclear.

The result is an expanding control gap. High-volume matters put more documents, decisions, vendors, and review activity into motion—often without comparable growth in legal-team capacity or budget. In Thomson Reuters Institute’s 2025 Legal Department Operations Index, 81% of respondents reported increasing matter volumes, 55% reported flat or decreasing legal-department budgets, and 56% described their departments as under-resourced.¹

Legal teams usually have plenty of data. What they often lack is a connected operating record. Spend is measured far more consistently than the signals that explain it: in ACC and Everlaw’s 2025 research, 83% of departments tracked outside-counsel spend, while only 16% tracked litigation outcomes relative to cost.²

Spend is usually visible. The operating picture often is not.

A practical first step is a small, defined set of operating measures—not a new dashboard. Discovery Operations Cost-per-document (CPD)—defined discovery cost divided by processed documents—fills one of the most consequential gaps: a consistent matter-level view of cost against volume. It should be paired with a small number of driver and quality measures that explain what moved and whether action is warranted.

Without a defined unit cost and supporting measures, discovery spend functions more like a receipt system than a control system. It records what happened but offers little ability to explain or steer cost while choices remain open.

But “$/doc” can mean several different things: first-level review cost, full review-operations cost, and matter-wide discovery cost. They are not competing benchmarks; they are nested measures that answer different operating questions. This note separates them into three tiers, explains what each is good for, and shows why matter-wide discovery cost per document is the anchor worth building.

Cost control operates within counsel’s judgments about relevance, proportionality, preservation, privilege, and production—not instead of them.

Why the number is hard to see

The source data usually exists. The operating record does not. Spend, volume, time, and quality are recorded in different places, under different definitions, for different reporting periods. The usual blockers are not mysterious:

Spend is split across buckets. Vendor, hosting, processing, review, outside-counsel, and internal costs often live in separate systems.
The denominator drifts when no one owns its definition. “Document volume” might mean collected files, processed documents, review-set documents, produced documents, families, threads, or an inconsistent mix. If the denominator is not defined, the metric is not real.
Time is not tracked cleanly. Attorney hours, reviewer-platform time, and output-based proxies can all be valid. They are not interchangeable.
Quality is not instrumented. Throughput without overturn rates, QC exceptions, or rework can make speed look like progress while defects move toward production.

Those gaps create operational drift. The work keeps moving, cost accumulates, and visibility gets worse until a deadline, invoice, or dispute makes the gap impossible to ignore.

Three cost-per-document measures

Measure	Costs included	Document population
Tier 1 — First-Level Review CPD	The direct cost of the first-level review pass.	Unique documents receiving that review.
Tier 2 — Review Operations CPD	The full review operation: first-level review, QC, sampling, privilege handling, project management, rework, and technology allocated to review.	Unique documents reviewed under the team’s defined protocol.
Tier 3 — Discovery Operations CPD	Tier 3-Core: collection, processing, hosting/platform licensing, review operations, and production.	Unique processed documents after deduplication and before downstream culling.

The cost boundaries are nested; the per-document figures need not increase from Tier 1 to Tier 3 because their denominators can differ.

Tier 1 — First-level review $/doc

This is the direct, isolated cost of first-level review, per unique document receiving that review—regardless of who performs it. A managed review vendor, contract attorneys, staff attorneys, or an in-house team can all generate a Tier 1 figure. It is usually driven by reviewer hourly cost and assumed review speed: documents per hour.

Vendor rate cards, RFI quotes, and hosted-review invoice lines are common sources for this number. Publicly reported vendor pricing varies materially by staffing model, review complexity, location, privilege exposure, language, and use of AI-assisted workflows.³An in-house or staff-attorney figure should be built from the team’s own direct hourly cost and throughput.

Tier 1 is useful for vendor comparison, in-house-versus-vendor staffing decisions, and RFI negotiation. It does not include the review operations around the first-level pass—QC, privilege handling, project management, or rework—and cannot show whether the review population was properly scoped, whether rework is growing, or whether the matter is running well. A poorly run matter with a well-priced first-level review is unnecessarily expensive.

Tier 2 — Review Operations $/doc

Tier 2 is the total cost of the review operation divided by the unique documents reviewed through the stated cutoff date. It includes first-level review, QC, sampling, privilege handling, project management, rework, and technology costs allocated to the review.

Count each document once when it reaches a completed review decision under the team’s protocol, whether through human review or a validated technology-assisted process. Documents merely processed or ranked for later review do not count. Identify the review method when comparing figures.

Tier 2 uses a broader cost boundary than Tier 1 because it measures the full review operation, not only the first-level pass. A team might start with illustrative diagnostic bands for a mid-market commercial matter with mixed staffing and standard privilege exposure:

Working: $1.50–$3.00 per reviewed document
At-risk: $3.00–$5.00 per reviewed document
Red-flag: above $5.00 per reviewed document

These BTLOC diagnostic bands draw on pricing research and analysis for commercial litigation involving substantial document review, including human review. Use them to investigate cost and establish a working range for your own matters, staffing, technology, and quality requirements. A figure outside the bands calls for an explanation; it does not, by itself, establish good or poor performance.

The arithmetic shows why the unit matters. For 100,000 reviewed documents, $4.00 per document—within the At-risk band—produces $400,000 in review-operations cost. Each additional quarter per document adds $25,000. The first question is whether 100,000 is the right review population. If the population is inflated, no rate negotiation recovers what the volume already cost.

These figures illustrate the arithmetic; they are not market benchmarks or expected results.

Measured against the team’s own baseline, Tier 2 tells the team whether review operations look approximately right, or whether something inside review is moving out of range—protocol, batching, QC design, PM allocation, rework rate—before it becomes a budget argument. EDRM’s Metrics Model and accompanying case study provide an established foundation for measuring discovery cost, time, and volume.⁴

Tier 3 — Discovery Operations $/doc

This is the roll-up. Tier 3 (practical default): Discovery Operations CPD. BTLOC’s default Tier 3-Core boundary includes collection, processing, hosting/platform licensing, review operations, and production. Divide those cumulative costs by unique processed documents after deduplication and before downstream culling. Keep the cost boundary and counting rule consistent, and align costs and volume to the same cutoff at each update.

Processed documents provide a common upstream reference point before downstream culling and review. Using that population allows changes in the share sent to review—and the cost of reviewing it—to affect the matter-level measure.

Optional expanded view: Total Allocable Discovery CPD. Tier 3+ adds any specifically named, consistently allocable discovery costs to the core Tier 3.

Why Discovery Operations CPD is the anchor

The instinct in a budget conversation is to start with Tier 1. First-level review is often the most visible number, and it sits inside review, usually the largest discovery cost category. But it does not capture the matter-level operating story.

In RAND’s 2012 study of large-volume matters, review accounted for roughly 73% of eDiscovery production cost, with processing another 19%.⁵ ComplexDiscovery’s long-horizon model puts review at a reconciled 62% of worldwide eDiscovery task spend in 2025, still the largest category even as automation compresses review work and collection takes a growing share.⁶

Sedona’s proportionality commentary emphasizes that assessing the needs of a case requires more than conjecture or unfounded assertions.⁷ A defined cost-and-volume baseline can contribute to that assessment, giving counsel concrete information for weighing scope choices on the matter at hand.

What the number can tell you

It distinguishes volume growth from changes in average cost. At a steady $2.50 per processed document, growth from 250,000 to 400,000 processed documents raises defined discovery cost from $625,000 to $1.0 million—$375,000 from volume. If unit cost also rises to $2.75, total cost reaches $1.1 million, adding $100,000 from the higher unit cost. The calculation shows what changed numerically; the operating drivers explain why.
It brings fragmented costs into one defined boundary. Tier 3-Core costs arrive through different providers, invoices, and systems. A reprocessing run, a few extra months of hosting, and a second QC pass can each look reasonable on its own invoice. Tier 3 shows what they do together to the matter’s unit cost.
It creates a shared matter view across providers. Providers generally account for their own assigned scope. The matter-wide record exists only if someone aligns spend, volume, reporting periods, and operating drivers across participants—which usually makes it a client-side function. Maintained while work is underway, Tier 3 gives the client a common signal while scope, staffing, and provider decisions remain open.

What it cannot tell you

Tier 3 is not an automatic savings metric. A lower number is not, by itself, proof that a matter was run better; it may reflect simpler data, a lighter privilege burden, a shorter hosting window, or a different review protocol.

Processing additional low-cost material can also lower the average while increasing total spend, so read Tier 3 alongside total cost, the share of processed documents sent to review, and the matter’s stage of completion.

Tier 3 supports two practical questions: Is this matter running approximately as expected? If not, which cost component or operating driver needs attention?

Cost per document should be paired with at least one quality signal—QC error rate, rework rate, or overturn rate. If cost improves while quality degrades, there is an operating failure hiding behind a better-looking number.

What to compare—and what not to

Comparability is not all-or-nothing. The more closely two figures align on cost boundary, denominator, measurement basis, and work profile, the more confidently the team can compare the overall result. When they differ, the comparison can still be useful—but as a diagnostic, not a performance verdict. In practice, that means comparing phase over phase within a matter, matter over matter when definitions align, and driver against driver when they don’t.

A slower review rate may reflect a denser population after effective culling; a faster rate may reflect a larger, noisier review set. The team can normalize what it can, identify the variables that changed, and compare narrower measures—such as culling yield, review speed, QC exceptions, rework, hosting duration, or vendor variance—without pretending the matters are identical.

Label every figure before comparing it. For example: Discovery Operations CPD · Tier 3-Core · Matter X · Through June 2026 · Unique processed documents after deduplication and before downstream culling.

Start with one recent baseline

Don’t start with a full portfolio dashboard. Instead, establish a reliable baseline on one active or recent matter first: a written cost boundary, shared cutoff date, defined processed-document denominator, named owner, review cadence, and decision triggers. Use current volumes, committed rates, and expected work to forecast cost to complete; the recorded CPD shows cost to date.

It does not need to be perfect to be useful; it needs to be consistent enough to explain movement and support a decision.

The baseline has the most leverage when established early. During the Rule 26(f) process and related ESI protocol discussions, a defined cost model can help counsel estimate how alternative scope, phasing, preservation, and production choices may affect downstream volume and cost. After those choices are made, the same measures can track actual scope and spending against expectations.

Fix the cost boundary. Start with Tier 3-Core: collection, processing, hosting/platform licensing, review operations, and production.
Align the cutoff date and processed-document count. Use an authoritative processing report or another defined source; align the cost and volume cutoffs; and document deduplication and any nontraditional object-counting rule.
Calculate the initial Tier 3 figure. It will be rough. It will still be useful.
Identify its two or three largest drivers. Quantify cost components as a share of the included cost where possible, identify the principal drivers, and state what would have to change to move each one.
Assign the owner, review cadence, and decision triggers. Define who maintains the baseline, when the team reviews it, and what movement requires approval, investigation, or escalation. Expect the cumulative figure to move with matter phase, invoice timing, and cost recognition; rising is not itself evidence of drift.

An AI-assisted workflow may change the review population and cost mix. Where technology materially excludes or deprioritizes documents, use method-appropriate validation.

Turning metrics into matter control

The number is the signal, not the control. The control is the operating rhythm built around it.

The problem is not simply whether a budget exists. In the 2026 CLM National Litigation Management Study, only 13.8% of responding claims and litigation executives described litigation budgets as very effective; 62.1% said they are mostly communication tools.⁸

A baseline is only a control if it can trigger a decision. That requires a named owner, shared cutoff date, defined decision rights, recurring review while decisions can still change, and an escalation point when the number or one of its drivers moves outside the expected range.

Counsel owns legal judgment. The operating layer makes scope, vendor, staffing, QC, and budget decisions visible in time to act.

With that, scope changes become explicit tradeoffs, QC becomes a control loop, and operating choices become more objective. Stage-specific metrics help optimize workflow parts. The anchor metric connects those improvements to matter-level economics, efficiency, and risk.

The invoice still matters. But it should not be the first place the team sees what the work is costing or why.

Related reading: The One-Minute Discovery Reality Check → Six-question self-diagnostic built around the same anchor metric: Discovery Operations CPD.

BTLOC provides litigation operations consulting focused on discovery operations. BTLOC is not a law firm and does not provide legal advice. This Operating Note is informational and does not create a consulting relationship. Any operating figures, bands, ranges, or examples are directional and matter-dependent; they are not commitments, guarantees, or estimates for any specific matter. Actual metrics, thresholds, and operating ranges should be established case by case based on the matter, data, workflow, staffing model, and quality requirements.