| Client | Boutique entertainment and media law firm, London |
| Matter Type | Complex commercial litigation |
| Review Universe | 70,773 documents |
| Services Provided | TAR 2.0 (Continuous Active Learning), managed review, validation |
A leading specialist law firm based in London was engaged in a complex commercial litigation matter requiring the review of a substantial cross-border document population. Faced with more than 70,000 potentially relevant documents, tight case deadlines, and mounting review costs, the client needed a defensible and efficient review strategy that could accelerate decision-making without sacrificing quality or accuracy.
A traditional linear review model would have required significant reviewer hours, increased costs, and extended timelines—creating the very inefficiencies and inconsistencies that often define the eDiscovery “quality rollercoaster.” The client sought a more agile approach that could deliver reliable results, improve review precision, and provide transparency throughout the lifecycle of the matter.
Sandline was retained as the independent forensic examiner and implemented a collection strategy rooted in fSandline Global deployed a Technology-Assisted Review (TAR 2.0) workflow powered by Continuous Active Learning (CAL), combining advanced analytics with experienced attorney oversight to create a smarter, more adaptive review process. Unlike legacy TAR 1.0 methodologies that depend on a static seed set, CAL continuously refines the predictive model as reviewers code documents, allowing the system to improve accuracy and prioritization in real time.
Consistent with Sandline’s philosophy of combining leading technology with proactive project management and senior-level expertise, the workflow was designed to maximize efficiency while maintaining defensibility at every stage of the review. This approach enabled the legal team to focus reviewer time on the documents most likely to be relevant, significantly reducing unnecessary review effort while maintaining confidence in the results.
Sandline ingested and processed a large volume of data, leveraging email threading, near-duplicate identification, and nonresponsive analysis to reduce the review population to an initial set of 70,773 documents.
- Documents were then surfaced to reviewers through a dynamic Priority Review Queue, ensuring that the materials most likely to be responsive were reviewed first and maximizing the value of each hour of attorney review time.
- As reviewers coded documents, the TAR model continuously retrained and recalibrated, refining responsiveness predictions after each review cycle and improving prioritization accuracy over time.
- Throughout the engagement, Sandline’s project management team closely monitored relevance rates, model stability, and statistical coverage metrics in real time, providing transparent reporting and proactive recommendations to the client as the review progressed.
- When relevance rates among queued documents began approaching the established 60% threshold, Sandline recommended a formal validation sample to determine whether the review could be defensibly concluded.
To ensure consistency and accuracy prior to validation, the team also performed mass-tag reconciliation between the original relevance coding field and the TAR relevance field, resolving minor discrepancies—typically fewer than 10 documents per cycle—before final validation and closure.
Of the 70,773 documents in the review universe, 35,666 were coded by human reviewers while the TAR model predicted the remaining 35,107 documents. Among the human-reviewed set, 62.1% were found to be responsive—a high rate indicating well-targeted collection and search term efforts.
| Metric | Result |
| Total Documents in Review Universe | 70,773 |
| Documents Coded by Human Reviewers | 35,666 |
| Documents Predicted by TAR Model | 35,107 |
| Human-Reviewed Responsiveness Rate | 62.1% |
Sandline performed a formal elusion sample of 500 documents drawn from the TAR-predicted non-responsive set. The results confirmed the review could defensibly conclude:
| Validation Metric | Result | 95% Confidence Interval |
| Elusion Rate | 4.6% | 2.94% – 6.82% |
| Precision | 100.0% | 100.00% – 100.00% |
| Recall | 93.2% | 89.98% – 95.64% |
| Richness | 33.6% | 32.75% – 34.68% |
Elusion Rate (4.6%): The elusion rate estimates the proportion of responsive documents remaining in the set the model classified as non-responsive. At 4.6%, this translates to approximately 1,613 potentially missed documents (range: 1,029–2,392). This falls within accepted industry thresholds for a defensible TAR review and is classified as a low elusion rate given the responsiveness of the dataset.
Precision (100.0%): Every document the model identified as responsive was genuinely responsive. The model retrieved zero false positives, a statistically robust result confirmed by the confidence interval.
Recall (93.2%): The review captured approximately 93 out of every 100 responsive documents in the universe. This exceeds common industry benchmarks for defensible review.
Richness (33.6%): The model estimated that 33.6% of the full universe was responsive. This figure appears lower than the 62.1% reviewer-coded rate because it accounts for the TAR-predicted non-responsive documents, which are expected to contain very little responsive material. Both figures are consistent with a well-functioning TAR process.
| Efficiency Metric | Result |
| Documents Eliminated from Manual Review | 35,107 |
| Reduction in Manual Review Volume | 49.6% |
| Average Reviewer Speed | 46 docs/hour |
| Estimated Review Hours Saved | ~763 hours |
TAR eliminated the need to manually review 35,107 documents. Based on the team’s average review speed of 46 documents per hour, this translates to approximately 763 hours of review time saved—time that would otherwise have been spent reading documents the model correctly identified as non-responsive.
Upon reviewing the validation metrics, the client confirmed acceptance of the elusion results and authorized conclusion of the TAR review. The final production set was then prepared for disclosure.
The engagement demonstrated a review process that was not only cost-effective and legally defensible, but also reflective of Sandline’s commitment to delivering consistent, high-quality outcomes through agile workflows, senior-level expertise, and proactive collaboration. By leveraging a cross-border CAL-driven TAR 2.0 strategy, the team achieved strong accuracy while significantly reducing the time and expense of manual review—helping the client move beyond the traditional eDiscovery “quality rollercoaster” and toward a more efficient, predictable, and defensible review process.
- Senior level project management support with 30-minute response times with AI expertise.
- Real-time reporting on model health, relevance trends, and review progress.
- Proactive communication: regular status updates including mass tagging reconciliation, relevance rate tracking, and clear recommendations on when to validate and conclude.
- Defensibility-first approach: formal elusion sampling with confidence intervals, documented methodology, and transparent client sign-off.
- Platform expertise: deep experience with Continuous Active Learning workflows and Priority Review Queues, within RelOne and Everlaw, ensuring reviewers spend time where it matters most.
- Scalable global team, including London-based resources with global capabilities across US, EMEA and APAC time zones
