Technology Assisted Review vs Linear Review: Real Cost and Accuracy Comparison

A large litigation firm tracked detailed metrics across 12 matters where they used both technology-assisted review (TAR) and traditional linear review on comparable document sets. The data eliminates the usual apples-to-oranges problem in comparing the two approaches because each matter involved the same case, same documents, and same relevance standards. The results were not close.

TAR produced an average recall rate of 89.4% compared to 67.2% for linear review. TAR cost an average of $0.31 per document compared to $0.78 for linear review. TAR completed review in an average of 4.3 weeks compared to 11.7 weeks for linear review. Across every metric the firm tracked, technology-assisted review outperformed manual review.

The Methodology Behind the Comparison

The firm's comparison was not an academic exercise. On several of their larger matters, they ran TAR on the full document collection while simultaneously assigning a subset to linear review teams for quality benchmarking. On other matters, they used TAR for initial review and linear review for quality control sampling, generating side-by-side accuracy data.

The document collections ranged from 200,000 to 6.8 million documents. Richness rates (the percentage of documents that were actually relevant) ranged from 2.1% to 11.4%. The variety of case types included antitrust litigation, securities fraud, patent infringement, and employment class actions.

The firm measured five key metrics across every matter: recall (percentage of relevant documents found), precision (percentage of documents coded relevant that actually were), cost per document reviewed, calendar time to completion, and defect rate on quality control samples.

Why Linear Review Underperforms on Accuracy

The finding that surprises most people is not that TAR is cheaper or faster. It is that TAR is more accurate. The assumption has always been that human review, while expensive, at least catches everything. The data does not support this assumption.

Linear review accuracy suffers from several structural problems. Reviewer fatigue is the most significant. A contract attorney reviewing their 300th document of the day simply does not apply the same attention as they did to document number 30. Studies on reviewer consistency show that the same reviewer will code the same document differently 20-30% of the time when re-presented with documents they have already reviewed.

Reviewer inconsistency across a team compounds the problem. In a typical linear review with 20 contract attorneys, each reviewer develops their own interpretation of the relevance standard, no matter how detailed the review protocol. Borderline documents, the ones that could reasonably be coded either way, get inconsistent treatment. Some reviewers are aggressive about coding documents as responsive; others are conservative. This inconsistency introduces systematic noise into the review.

Distraction and context switching also play a role. Linear reviewers see documents in essentially random order. A contract reviewer might look at an email about a product launch, then a financial spreadsheet, then a personal email, then a regulatory filing. The constant context switching makes it difficult to maintain a coherent understanding of how individual documents relate to the case narrative.

TAR avoids these problems because the algorithm applies the same criteria to every document with perfect consistency. It does not get tired, it does not lose focus after lunch, and it does not develop idiosyncratic interpretations of the coding protocol.

The Cost Comparison in Detail

The per-document cost difference ($0.31 vs $0.78) reflects several factors. TAR requires less total attorney time because most documents are scored by the algorithm rather than reviewed by humans. The human review that TAR does require is focused on the highest-value documents, meaning senior attorneys spend their time on substantively important materials rather than plowing through obviously irrelevant emails.

The cost comparison looks even more favorable for TAR when you include indirect costs. Linear review requires more extensive project management because larger review teams create more coordination overhead. Quality control for linear review requires larger sampling sizes because the base error rate is higher. Hosting costs are similar for both approaches, but the shorter timeline for TAR means less total hosting expense.

On matters with document collections above 2 million, the cost gap widens further because TAR's fixed costs (technology licensing, model training, validation) get spread across more documents while linear review costs scale linearly with volume.

Where Linear Review Still Has a Role

TAR does not eliminate the need for human review entirely. Several scenarios still favor some degree of linear review.

Privilege review is the most important. While TAR can flag potentially privileged documents, the final privilege determination requires careful human judgment because the consequences of producing a privileged document can be severe. Most firms use TAR to identify likely privileged documents and then apply linear review to that subset.

Hot document identification also benefits from human review. TAR can rank documents by relevance, but identifying which responsive documents are particularly important for case strategy requires legal judgment that algorithms cannot replicate. Many firms use a hybrid approach: TAR identifies all responsive documents, and then senior attorneys review the top-ranked documents to identify the ones that will drive depositions, motions, and trial strategy.

Small document collections (below 100,000 documents) may not justify the setup costs for TAR. The technology licensing, model training, and validation process has a minimum cost floor that can exceed the cost of a focused linear review on smaller sets.

For law firms evaluating their review technology investments, the data from these 12 matters makes the case clearly. TAR is not just a cost-cutting tool. It is a quality improvement tool that happens to also cost less. The firms that recognized this early have already built the institutional knowledge to deploy TAR effectively; the firms that have not are competing at a structural disadvantage on both price and quality.

What the Trend Data Shows

The firm's data also showed that TAR performance improved across successive matters. The first matter they handled with TAR showed a 76% recall rate. By the twelfth matter, recall had climbed to 94%. This improvement curve reflects the team's growing expertise in training models, selecting seed sets, and calibrating validation protocols rather than improvements in the underlying technology.

The implication is that there is a meaningful learning curve for TAR adoption. Firms that run their first TAR project may not see results as strong as firms with established TAR practices. But the improvement happens relatively quickly, typically within 3-4 matters, after which the process becomes routine and the benefits become consistent. The investment in building that competency pays for itself within the first year for any firm with a regular volume of large-scale document reviews.