What you pay before anyone reads a document

On most eDiscovery platforms the bill starts before the first document is opened, and it is not one number. It is a stack of meters: per gigabyte to process, per gigabyte per month to host, per user per month, per page to OCR and image, per gigabyte to produce, and per hour for the people who run the software. Each one is small on its own. That is the design.

A rate card, line by line

This is a public one: Consilio's master contract with the State of Washington, in force through 2028. It is a large provider selling Relativity to a large customer, which means these are good rates, not bad ones.

Other public sheets read the same way. One lists TIFF processing at $350 a gigabyte and a $35 per gigabyte charge just to switch analytics on. Another charges $125 a gigabyte to turn natives into searchable PDFs and $15 a gigabyte to export your own database. Epiq's RelativityOne card for 2026 runs to a dozen billing codes before review: all-in processing at £45.50 a gigabyte, three tiers of hosting, a separate fee to load data that was processed somewhere else, and a per-gigabyte production charge on the way out.

One small matter, added up

Take a 50-gigabyte collection, five users, 100,000 scanned pages, a 20-gigabyte production, and forty hours of a project manager's time over the year. At the rates above: processing is $3,750. Hosting is $7,200 a year. Five seats are $4,500. OCR is $3,000. Producing 20 gigabytes of natives is $5,500. The project manager is $7,000. That is $30,950 in the first year, before a single hour of review, on a matter most firms would call small.

Per-page fees are where quotes drift furthest from the sheet. In quotes we have seen this year, OCR, text search and PDF conversion together came to 15 to 20 cents a page, five to seven times the contract rate above. On the same 100,000 pages that line alone is $15,000 to $20,000, and the year lands closer to $45,000.

Courts notice. A federal magistrate in Florida last year allowed ten cents a page for electronic duplication and struck the platform's hosting and processing charge as a non-taxable “data hosting fee.” Hosting is the meter you pay and never recover.

What those items cost to provide

Very little. Storage is cheap. Processing is a solved problem. OCR is a commodity. The one thing that genuinely costs money in modern eDiscovery is AI compute, and none of the line items above is that. The hours are the most telling: a platform that needs a professional to operate it is charging you for its own complexity.

How Marbury takes the meters off

The engine does the processing: it expands containers, hashes every file, screens out system files against the NIST list, deduplicates without breaking families, runs AI OCR on scanned pages and images so they are searchable, converts Slack, Teams and text exports into readable threads, and indexes the result. Every excluded file lands on a named list. The collection stays here for review, or exports as a Concordance load file, text and coding included, so the next platform has the OCR text on day one instead of running it again.

There are no hours. No project management fee, no professional services, no hourly rate for someone to operate the software, because it is built to be run by the lawyer on the matter and the agent does the repetitive work. There is no vendor in between, no separate products, and no per-page surprise. Storage and processing cost very little to provide, so they are not where the money goes. The real cost is AI compute, and that is where Marbury puts it. Pricing is set with our first partners, and we will walk you through it.