E-Discovery Cost Reduction: How a Litigation Firm Saved $1.2M on a Single Case
A litigation boutique used AI-powered e-discovery tools to review 2.3 million documents in a fraction of the time and cost of traditional review.
Background
Brennan & Cole LLP is a 20-attorney litigation boutique in Chicago specializing in complex commercial disputes. The firm was retained to defend a mid-size manufacturing company in a breach-of-contract and trade-secret misappropriation case brought by a former supplier. The stakes were high — the plaintiff sought $18M in damages — and the discovery obligations were massive.
The client's document universe included 2.3 million files: emails, Slack messages, shared drive documents, engineering specifications, and financial records spanning seven years of the business relationship. All of it was potentially discoverable.
The Challenge
The firm obtained estimates from two traditional e-discovery vendors. Both projected a review cost of approximately $1.8M, based on a linear review model using contract attorneys at $45 to $65 per hour. The timeline was equally daunting: 8 months for first-pass review, with production deadlines set at 6 months by the court. The numbers simply did not work.
Beyond cost and time, quality was a concern. Linear review — where contract attorneys read documents one by one — is prone to inconsistency. Different reviewers apply different judgment to the same privilege and relevance questions, especially over an 8-month engagement with reviewer turnover. The risk of inadvertent privilege waiver on a dataset this large was significant.
The Solution
The litigation team selected an AI-powered e-discovery platform that combined predictive coding (technology-assisted review, or TAR) with advanced analytics including email threading, near-duplicate detection, and concept clustering. The workflow proceeded in three phases.
In Phase 1, the platform ingested and processed all 2.3 million documents, automatically de-duplicating the set down to 1.4 million unique items. Email threading further reduced the review set by grouping conversation chains into single review units. In Phase 2, a senior associate and two partners reviewed a seed set of 2,500 documents, coding each for relevance, privilege, and key issues. The AI model trained on these decisions, then scored the entire corpus. After two additional rounds of active learning — where the model surfaced the most uncertain documents for human review — the system achieved a recall rate above 90% and precision above 85%. In Phase 3, a targeted human review of the AI-identified relevant and privileged documents was conducted by a team of six attorneys over four weeks, focusing on the approximately 180,000 documents the AI flagged as potentially responsive.
The Results
The entire review was completed in 10 weeks — well within the court's 6-month production deadline and roughly 75% faster than the traditional estimate. Total cost came in at $600,000, saving the client $1.2M compared to the linear review estimates.
Quality metrics were equally strong. The AI-assisted review achieved higher consistency than linear review benchmarks, with inter-reviewer agreement rates above 92%. A post-review quality control sample of 1,500 documents found only 11 coding errors — a 0.73% error rate, compared to the 3 to 5% error rates typical of large-scale linear reviews. No privilege documents were inadvertently produced.
The case ultimately settled favorably for the client, with opposing counsel acknowledging that the speed and thoroughness of production put pressure on their own, more disorganized discovery process.
By the Numbers
$1.2M
Cost savings vs. traditional review
2.3M
Documents reviewed
10 weeks
Total review time (vs. 8 months)
Key Takeaways
- TAR is no longer experimental. Courts have accepted technology-assisted review for over a decade. The question is no longer whether to use it, but how to use it well.
- Senior attorney involvement in seed sets is critical. The quality of the AI model depends entirely on the quality of the training decisions. This is not a task to delegate to the most junior person on the team.
- Cost savings scale with document volume. The larger the dataset, the greater the advantage of AI-assisted review over linear review. At 2.3M documents, the savings were dramatic.
- Speed is a strategic advantage. Producing documents quickly and thoroughly shifts litigation dynamics in your favor and demonstrates confidence in your position.