Disclosure and predictive coding: how courts scrutinise technology-assisted review

Why disclosure needs technology at all

In civil litigation, parties must search for and disclose documents relevant to the issues in a case. In anything beyond a small dispute, that pool of material is now almost entirely electronic: emails, messaging apps, shared drives, contracts, spreadsheets. Reviewing every document by hand is often impossible within reasonable time and cost, which is why technology-assisted review (TAR) and predictive coding tools have become standard in substantial commercial litigation and investigations.

But in England and Wales, the use of this technology is not simply a private choice between a law firm and its software supplier. Disclosure is governed by the Civil Procedure Rules, and for many business and property disputes in the Business and Property Courts, by a dedicated disclosure regime that expects parties to be transparent about how they searched, what they used, and why they believe the approach was reasonable and proportionate.

The legal framework: Practice Direction 57AD

Disclosure in the Business and Property Courts is governed by Practice Direction 57AD, which replaced an earlier pilot scheme and sits alongside Part 31 of the Civil Procedure Rules for other courts. The practice direction requires parties to agree, early in a case, a Disclosure Review Document that sets out the proposed scope of disclosure, the data sources to be searched, and the methodology to be used, including whether keyword searching, technology-assisted review, or a combination will be applied.

This matters because it moves the decision about using predictive coding out of the law firm’s back office and into a document the other side and, if necessary, the court can see and challenge. A party that wants to rely on TAR to reduce the volume of documents needing manual eyes-on review must generally be prepared to explain and justify that choice to the opposing party, and ultimately to a judge if it is disputed.

What the court actually looks at

When a disclosure methodology is challenged, or when the court is asked to approve a proposed approach, it tends to focus on a handful of practical questions rather than the underlying algorithm itself:

• Was the training of the review tool properly supervised by lawyers with genuine understanding of the issues in the case?
• Was the process validated, typically through statistical sampling of documents the tool classified as not relevant, to check the error rate is acceptable?
• Is there a clear, documented audit trail showing how search terms, seed documents and review decisions were arrived at?
• Is the overall approach proportionate to the value and complexity of the case, reflecting the overriding objective in the Civil Procedure Rules to deal with cases justly and at proportionate cost?

Courts in England and Wales have generally been willing to accept predictive coding as a legitimate alternative to manual review or simple keyword searching, provided the process is transparent and defensible. The direction of travel in case law and professional guidance has been towards acceptance of TAR as reasonable, not scepticism of it, so long as parties do the groundwork of agreeing methodology and documenting validation.

Keyword search versus predictive coding

Many disclosure exercises still rely heavily on agreed keyword searches across custodians’ mailboxes and document stores, often combined with date ranges and deduplication. Predictive coding goes further: lawyers review a sample set of documents, tag them relevant or not relevant, and the software learns from those tags to rank or classify the remaining, much larger, document population. Reviewers then typically focus their time on the documents the system considers most likely to be relevant, with sampling used to check what has been left unreviewed.

The practical difference for a law firm is accountability. A keyword search is relatively easy to explain and reproduce. A predictive coding exercise requires the firm to be able to describe, and defend, a statistical process, including the sample sizes used for training and validation and the confidence levels achieved. Firms that cannot explain their own methodology in plain terms to a judge are taking on real risk, because an adverse finding on disclosure can affect costs orders or even the admissibility of a party’s evidence on the point.

Costs, proportionality and professional duties

Disclosure exercises, particularly in large commercial disputes, can be one of the most expensive phases of litigation. Costs budgeting under the Civil Procedure Rules requires parties to estimate and justify the cost of disclosure in advance, which puts pressure on firms to choose a methodology, whether manual review, keyword search, TAR, or some blend, that is proportionate to the sums and issues at stake, not simply the most thorough option available.

Solicitors also carry a professional duty of competence and of candour to the court. Relying on a review tool does not transfer that responsibility to the software. A solicitor who signs a disclosure certificate is personally vouching that reasonable steps were taken, so firms using TAR need internal processes that keep a qualified lawyer, not just a technologist, accountable for the final judgement calls.

Practical takeaway

For firms and clients, the message is straightforward: technology-assisted review is accepted and often expected in substantial disclosure exercises, but it must be agreed, documented and validated, not simply switched on. Anyone running or receiving disclosure in a Business and Property Courts matter should check the current Practice Direction 57AD requirements and the latest Civil Procedure Rules guidance before finalising a methodology.