How E-Discovery and Technology-Assisted Review Work

What e-discovery actually is

E-discovery, usually called disclosure in England and Wales, is the process of identifying, collecting, reviewing and producing electronic documents relevant to a legal dispute or investigation. It covers emails, chat messages, spreadsheets, contracts, mobile data and cloud files. Two or three decades ago disclosure meant photocopying paper. Today a mid-sized commercial dispute can generate several million potentially relevant documents once mailboxes, shared drives and messaging apps are collected, far more than any team of lawyers could read individually within a sensible budget or timeframe.

The civil courts recognise this. The disclosure rules for the Business and Property Courts (Practice Direction 57AD, made under the Civil Procedure Rules) require parties to agree a proportionate approach to searching for documents and to record it in a Disclosure Review Document. The rules expressly contemplate the use of technology, including keyword searches and technology-assisted review, as legitimate ways of narrowing an enormous data set to something a legal team can actually review and a court can rely on.

The standard workflow (EDRM)

Most providers and law firms follow a version of the Electronic Discovery Reference Model, a widely used industry framework:

• Identification, working out which custodians (people) and systems might hold relevant material.
• Preservation and collection, securing data so it cannot be altered or deleted, then extracting it forensically.
• Processing, converting collected data into a searchable format, removing duplicates, and extracting metadata such as sender, date and file type.
• Review, lawyers or trained reviewers examine documents for relevance, privilege and confidentiality.
• Analysis, using the review output to build the case narrative.
• Production, disclosing the relevant, non-privileged documents to the other side in an agreed format.

Technology touches every stage, but it is the review stage where the most significant efficiency gains, and the most scrutiny, tend to arise.

Keyword search versus technology-assisted review

The simplest way to cut down a data set is keyword searching: agreeing search terms with the other side and running them across the document population. This is cheap and easy to explain but blunt. Poorly chosen terms miss relevant documents (a false negative problem) or capture huge volumes of irrelevant ones (a false positive problem), and it does nothing to help prioritise what is found.

Technology-assisted review, often called TAR or predictive coding, goes further. A senior lawyer reviews a sample of documents and codes them as relevant or not relevant. The software builds a statistical model from those decisions and then scores the remaining documents by likely relevance, effectively predicting how a human reviewer would have coded them. The review team can then prioritise the highest-scoring documents first, or use the model to eliminate the batch of documents it is confident are irrelevant, subject to quality checks.

There are broadly two approaches in use. Simple passive learning trains the model once on a random sample and then applies it. Continuous active learning retrains the model constantly as reviewers keep coding, which tends to be more efficient and is now the more common approach among specialist providers.

Why courts accept it

TAR has been accepted by courts in England and Wales and elsewhere as a legitimate disclosure method, provided the process is transparent and defensible. Defensibility generally means the parties agree the methodology in advance, the seed set and training process are documented, and the output is validated using statistical sampling, checking a random sample of documents the model excluded to measure the likely recall rate (the proportion of truly relevant documents actually found). A validation exercise showing that a large majority of relevant material has been captured, with the shortfall explained and proportionate, is usually what persuades a court or opposing party that the process was sound. This is why parties are expected to cooperate and disclose their methodology rather than treat it as a black box.

Cost, proportionality and governance

The attraction of TAR is largely cost and speed. Linear manual review of millions of documents at a reasonable per-document rate becomes prohibitively expensive and slow, and the courts’ overriding objective requires disclosure to be proportionate to the value and complexity of the case. TAR does not remove the need for lawyers; qualified reviewers still make the relevance and privilege calls, supervise quality control, and take responsibility for what is ultimately produced. Data protection also matters throughout: collected data often includes personal information, so firms need a lawful basis for processing it and appropriate security, particularly where data crosses borders.

Practical points for anyone involved in a document-heavy dispute

Agree the disclosure methodology early, ideally before collection begins, since retrofitting a defensible process is far harder than building one from the start. Keep a clear audit trail of decisions, sampling and validation statistics. Treat privilege review as a distinct, careful step, since technology can flag likely privileged material but a lawyer must confirm it. Finally, remember that TAR is a tool for prioritisation and proportionality, not a substitute for legal judgement about what a case actually needs.

Anyone dealing with disclosure obligations should check the current Civil Procedure Rules and practice directions, since the detailed requirements are periodically updated.