Discovery Document Review Time for One Million Pages

Discovery document review time for one million pages could range from about 10,000 to 33,333 reviewer-hours, assuming 30–100 pages per hour. These are planning scenarios, not industry benchmarks; preparation, quality control, privilege review, and production work add time. A useful estimate starts with the review task, document complexity, and measured pilot throughput.

How do you calculate discovery document review time?

The basic calculation is straightforward:

First-pass reviewer-hours = pages requiring review ÷ pages reviewed per hour.

For one million pages, different assumed speeds produce very different workloads:

Assumed review speed	First-pass reviewer-hours	Eight-hour equivalents
30 pages/hour	33,333	4,167
60 pages/hour	16,667	2,083
100 pages/hour	10,000	1,250

Eight-hour equivalents are accounting units, not realistic daily productivity promises. Reviewers also attend briefings, resolve coding questions, document decisions, and take breaks.

The table assumes every page needs first-pass review. It excludes collection, processing, redaction, second-level review, and production. Do not present a first-pass estimate as the complete discovery schedule.

Why is page count an imperfect measure?

A page is not a consistent unit of review effort. A repetitive account statement and a dense conversation involving several participants may each occupy one page, but they impose different reading and interpretation demands.

Document boundaries also matter. A 100-page attachment might receive one responsiveness decision, while 100 one-page messages could require separate decisions and context checks.

Before estimating hours, establish what the million-page count represents:

• Rendered pages: Pages generated from native files, potentially affected by print settings.
• Scanned pages: Images that may need OCR and legibility checks.
• Document families: Parent messages and attachments that need contextual review.
• Repeated material: Duplicate files, quoted messages, and overlapping productions.
• Mixed evidence: Transcripts, spreadsheets, messages, and other formats with different review demands.

Where possible, track pages, documents, families, and file types together. For spreadsheets and message exports, page counts alone can seriously distort the workload.

What changes the number of review hours?

How demanding are the review decisions?

Simple issue tagging is not the same task as responsiveness review combined with privilege analysis, confidentiality coding, and redaction. Each additional decision can increase handling time and the need for escalation.

Define the review protocol before assigning a speed. Specify the issues, coding fields, treatment of attachments, and circumstances requiring a second reviewer.

How difficult is the evidence to interpret?

Poor scans, handwriting, unfamiliar terminology, multiple languages, and fragmented conversations can slow review. A reviewer may need surrounding records to understand an otherwise short message.

Cross-document reasoning also takes time. Establishing who knew what, when they knew it, and how a transaction or relationship developed is different from finding a keyword.

How much repeated content can be removed safely?

Deduplication and related-document grouping may reduce repetitive work. However, identical content can have different custodians, recipients, timestamps, or evidentiary significance.

Preserve the metadata and relationships needed to explain any reduction. Excluding material from individual review should follow an approved protocol, not an assumption that repeated text is irrelevant.

How should you build a defensible estimate?

Run a pilot that resembles the actual assignment. Avoid selecting only clean, short documents or letting reviewers skip difficult categories.

A useful pilot includes:

1. Representative material: Sample across custodians, date ranges, file types, languages, and complexity levels.
2. A stable protocol: Give reviewers the coding instructions intended for the full project.
3. Recorded effort: Separate active review, escalation, technical delays, and quality-control time.
4. Accuracy checks: Examine disagreements and missed issues rather than rewarding speed alone.
5. Stratified results: Calculate separate throughput for materially different evidence categories.

Use the resulting rates to create low, expected, and high estimates. If the sample is too small to capture unusual or complex material, identify that limitation explicitly.

For example, suppose 700,000 pages are routine and 300,000 are complex. If pilot rates are 100 and 30 pages per hour respectively, first-pass review requires 17,000 hours: 7,000 routine-review hours plus 10,000 complex-review hours. That is more informative than assigning one optimistic rate to the entire collection.

How do reviewer-hours translate into a deadline?

Reviewer-hours measure labor. Calendar duration depends on staffing, productive review hours per day, sequencing, and available specialists.

Use this scheduling formula:

Working days = total review-related hours ÷ reviewers ÷ productive hours per reviewer per day.

Consider an illustrative baseline:

• One million pages at 60 pages per hour: approximately 16,667 first-pass hours.
• A provisional 25% allowance for calibration, quality checks, and escalations: approximately 4,167 additional hours.
• Total modeled workload: approximately 20,834 hours.
• Twenty reviewers averaging six productive hours daily: 120 team-hours per day.
• Estimated duration: approximately 174 working days, or 35 five-day workweeks.

The 25% allowance is an example, not a standard. Replace it with pilot findings and separately budget work it does not cover, such as extensive redaction or production preparation.

Adding reviewers does not always shorten the schedule proportionally. Onboarding, inconsistent coding, and limited senior-review capacity can become bottlenecks. Schedule specialist work explicitly rather than assuming every reviewer can handle every task.

Can AI reduce review time for a million pages?

AI can help organize evidence, surface related material, and prioritize where reviewers should look. Whether that reduces total labor depends on the task, evidence quality, validation requirements, and how the team uses the output.

Distinguish three separate measurements:

• Machine processing time: How long the system takes to process submitted material.
• Human review time: Time spent reading, deciding, verifying, and documenting.
• Elapsed project time: The schedule from preparation through final deliverables.

A fast processing job does not establish that a million pages have received adequate legal review. Measure the complete workflow, including checking AI-generated findings against source records and investigating material the system did not prioritize.

Lantern ingests search warrant returns, discovery, transcripts, call records, and digital evidence, then produces searchable timelines, entity/relationship graphs, pattern detection, and AI recommendations with concrete next investigative steps. These capabilities can support chronology building and investigative follow-up; they do not establish a guaranteed pages-per-hour review rate.

Live ingestion today covers Facebook, Instagram, and WhatsApp only. Other source integrations—including telecom, banks, payments, crypto, identity/DNA, cloud storage, and OSINT—are on the roadmap. Distinguish supplied evidence from live source integrations when scoping a project.

Lantern can be deployed on-prem or in a private cloud it stands up, with role-based access controls (CJIS support coming soon), audit trails, and court-ready exports. Teams should still confirm their own access, handling, and validation requirements. This article provides planning information, not legal advice; discovery procedures vary by jurisdiction.

Frequently asked questions

How many hours should we initially budget for one million pages?

An initial first-pass scenario is approximately 10,000–33,333 reviewer-hours at assumed speeds of 100–30 pages per hour. Add preparation, quality control, privilege work, redaction, and production separately, then refine the estimate through a representative pilot.

Can ten reviewers finish in one month?

At six productive hours per day over 20 working days, ten reviewers provide 1,200 hours. Reviewing every page would require roughly 833 pages per hour per reviewer, well outside the illustrative assumptions here and not a sound basis for substantive page-by-page review.

Should estimates use pages or documents per hour?

Use the unit that best reflects the material, and track both where practical. Documents per hour can help with coding throughput, while pages, families, and file complexity reveal workload that a document count may hide.

Does AI processing replace human review hours?

No one-to-one conversion should be assumed. Evaluate AI by the verified work it helps complete, including the time required to validate outputs and address gaps.

Start with a representative sample, measure the full workflow, and revise the schedule as evidence complexity becomes clearer. To explore Lantern with your evidence-review workflow, start free with 5 credits; 1 credit equals 1 minute of processing.