E

eDiscovery AI

The models are the product, with nothing beneath them to license. Every capability sold is a classification or extraction produced by a language model: relevance coding against written instructions, privilege calls with named attorneys and privilege elements, early case intelligence, detection and extraction of personal information, and image comparison, filtering and facial recognition. No repository, processing engine, review platform or production tool is sold alongside them.

The delivery design settles it. The product is a plugin that receives documents from Relativity, classifies them and writes results back to a mapped field. The platform of record belongs to someone else, and this vendor supplies the judgment applied to the documents. The vendor presents AI review as the successor to Boolean search and technology assisted review, not a feature added to either.

Grounding is real and documented for each document, and a figure is published without the test set that would make it testable. The grounding is built into the output. Relevance restates the instruction it was given for each classification and explains why it categorized the document as it did, consistently across every document. Privilege goes further: it names the attorneys it identified in a separate field, sets out which elements of privilege are present, and drafts its reasoning as a privilege log entry.

The published workflow ends in validation, with the fourth step being to review results and generate standard performance metrics. The vendor says the product consistently achieves 90 percent recall with excellent precision. No corpus, sample, matter type or method is given for the recall figure, precision is described but not quantified, and no failure mode is identified.

The review steps and the route back to human judgment are published as part of the workflow, short of any stated threshold. Oversight is built into the four published steps: the reviewer sets the criteria, the model classifies, and the fourth step is to review results to validate the classifications and generate standard performance metrics. That is a validation gate with a measurement attached. Every classification carries a restated instruction and an explanation, so a reviewer can audit a call rather than accept it, and privilege output is a draft, with the log entry field available to use or edit.

The named attorney field exists so results can be validated. What is missing is the boundary. Nothing published states a confidence threshold, what the system does with an ambiguous document, an escalation path, or what share of a set a reviewer should sample. Nothing separates a confident classification from a doubtful one, which matters on privilege, where a single wrong call can waive it.

The headline result is stated with no basis, and no customer is named on the pages read. The published claim is about performance, not deployment: the product consistently achieves 90 percent recall with excellent precision, and reviews hundreds of thousands of documents in hours at a fraction of the time and cost of manual review. No matter, client, engagement, volume or method is attached, so a reader cannot tell whether the figure comes from a benchmark, a customer project or an internal test.

The product is listed on the Relativity App Hub and delivered as a Relativity plugin, which is a partner relationship, not a customer. The vendor publishes case studies alongside fact sheets, whitepapers, videos and books. These were not opened, and any of them could hold a named engagement with a figure and a method.

Nothing is published on how client confidences are handled by a product built to take them in, and both published documents were read in full. The Privacy Notice, last updated 8 January 2025, applies only to information collected on or through the vendor's websites or other interactions with the vendor and its clients. It covers names, addresses, identifiers, browsing data and marketing preferences, and says nothing about customer documents submitted for classification.

The Terms and Conditions, last updated 23 May 2024, are a generic website terms of use with no data provisions. So nothing published addresses training on client documents, retention of a reviewed collection, segregation between matters or clients, or what any model provider may keep. The product's main function is to identify privileged material so it can be withheld, yet the vendor publishes nothing on how it protects the privilege of the documents it reads.

Nothing published addresses the advice line for a product that makes privilege calls. The audience is clear: the product is sold to legal teams and eDiscovery service providers and delivered inside Relativity, so there is no consumer confusion to resolve. Nothing else is addressed. No statement is published that the vendor is not a law firm, that a classification is not a legal determination, or that a privilege call needs attorney review before it is used to withhold a document.

Nothing covers supervision, competence or the reviewer's own professional duties, and the two published documents, both read in full, contain no disclaimer on the point. That matters because a privilege designation is a legal conclusion with a waiver consequence, and the published workflow has the model make it first. The validation step does put a person after the machine, but that is a control, not a professional responsibility position.

No governance position is published for a system whose output affects legal outcomes. There is no responsible AI page, principles, named framework, certification, accountable person, description of what is evaluated before a model changes, or testing regime beyond the project metrics a customer generates itself. Neither published document mentions AI. Bias is unaddressed, and it is a live issue here. The vendor ships facial recognition as a named feature, alongside detection of personal information, applied to litigation collections that contain images of real people.

Facial recognition is the AI capability with the most documented demographic performance gaps, and nothing published addresses accuracy across populations, thresholds, or what a false match in document review would mean. The relevance and privilege classifiers raise a quieter version of the question, since a model reading correspondence to infer legal significance can perform unevenly across writing styles, languages and formats.

Nothing is published on retention, deletion or access for a system that holds client documents, and both documents that would carry it were read in full. The Privacy Notice's security section is one paragraph. It says the vendor takes a number of steps to protect against loss and misuse of information under its control, warns that no internet transmission is completely secure, says information is sent at the user's own risk, and disclaims responsibility for illegal acts such as hacking by third parties.

No encryption, access control, monitoring or incident practice is described. Its retention section sets criteria rather than periods and covers personal information collected through the website, not a reviewed document collection. There is no subprocessor list, security page, trust center or data processing addendum, and the full site navigation has no security page. The Terms and Conditions contain no data provisions. For a product that takes in whole litigation collections, that is the entire published record.

Nothing is published on who bears the loss when the system is wrong. The Terms and Conditions were read in full. They run to about a dozen short clauses on content ownership, access to external resources, acceptable use, no waiver, service interruption, reselling, intellectual property, changes to terms, assignment, contacts and severability. They contain no limitation of liability, warranty disclaimer, indemnity or cap of any kind, so nothing is allocated in either direction.

Nothing addresses a relevant document coded nonresponsive and not produced, or a privileged document classified as not privileged and produced, which is the costliest failure in this workflow and the one the product exists to prevent. No warranty attaches to the published 90 percent recall figure, no insurance position appears, and no customer agreement is published.

One integration, documented at the level of what actually moves. The product is a Relativity plugin listed on the Relativity App Hub, a reviewed partner channel rather than a self declared compatibility claim. The workflow is described: the reviewer selects documents in Relativity, supplies relevance, issue, privilege and category instructions, submits the documents through the plugin, and receives results written back to a mapped field.

The vendor says the only setup needed is the prompt instructions and the field to map results to, and that the same instructions can run on more documents without new indexing or model training. Breadth and documentation are missing. No other platform is named, so a firm on Everlaw, Reveal, DISCO or Nuix has nothing published to work from. No API, developer documentation or connector specification is published, and nothing describes authentication, volume limits or how results are versioned.

Cloud delivery is implied, and neither the tenancy model nor the region is stated. Documents go from a customer's Relativity workspace to eDiscovery AI for classification, and results return to a mapped field, so processing happens on the vendor's side through a hosted service. Nothing more is published. No cloud provider, region or data center is named, no residency option is offered, no tenancy model is described, and nothing says whether one customer's collection is isolated from another's. Nothing separates where documents are held during processing from where inference runs, and no on premises or in tenant option is mentioned.

The only geographic statement is in the Privacy Notice: the vendor operates in various countries, including the United States and countries in the EU and EEA, and personal information may be accessed from outside the United States. That concerns website personal information, not document collections.

No independent security attestation is published. The site navigation and footer show Solutions, Partners, Thought Leadership, Company and Contact, with Privacy Policy and Terms and Conditions in the footer. There is no security page, trust center or compliance page. The Privacy item in the Solutions menu is a product for detecting and extracting personal information, not a policy page. No certification of any kind is claimed: no SOC 2, ISO 27001 or HIPAA, and no badge, auditor, report or request route.

No penetration testing or vulnerability disclosure program is described. The only security statement is the Privacy Notice paragraph, which describes no control and disclaims responsibility for third party hacking. There is nothing to request.

The vendor refers to advanced AI without identifying anything underneath it. The language across the site stays at the level of capability and is consistent: advanced AI technology, powerful and efficient AI technology, generative AI, and AI review technology that understands grammar and context. No model is named, no version is given, no provider is identified, no hosting arrangement for inference is stated, and nothing commits to notifying a customer when any of it changes.

There is no subprocessor list on any surface, and neither published document names a processor. One architectural fact is disclosed: the vendor states that no additional indexing or model training is required to run the same instructions against a further document set. The product thus works by instruction rather than by fine tuning on the customer's collection, a statement about method rather than supply. Documents leave the customer's Relativity workspace and go to the vendor for classification, so a litigation collection reaches a model no one has named.

No pricing information is published at any level, including the unit of charge. There is no pricing page. No plan, tier, rate, band, minimum or term is stated, and nothing says whether the product is charged per document, per gigabyte, per matter, per classification run or by subscription, the first question a litigation support manager would ask. The only commercial route is a demo booking, repeated on every page, with a phone number and a support address as the only contacts.

The vendor's central commercial claim is that review is done at a fraction of the time and cost of manual review, and no figure lets a buyer test that against what contract attorney review would cost.

Coverage is described by capability rather than by who the product serves, and the breadth is claimed without evidence. What is published is a product list: four suites covering early case intelligence, review, privacy and multimedia, each with named parts. The buyer is described only as legal professionals and legal teams, a Partners page shows eDiscovery service providers as a channel, and the Relativity plugin implies a firm or provider already on that platform.

No practice area is named, no matter type is called suited or unsuited, no firm segment or size is addressed, and no line is drawn between what a law firm, corporate legal department or service provider would use. There is no jurisdictional statement at all. That matters on a product making privilege calls, since privilege doctrine and the treatment of in house counsel communications differ between federal, state and foreign practice, and nothing says which the privilege classifier is built for.