Burnet, Duckworth & Palmer LLP

For BD&P, a Calgary-based Canadian law firm, the volume and complexity of modern litigation data had made traditional document review increasingly costly and difficult to scale. Manual review required sustained lawyer time to train reviewers, identify key issues, and maintain consistency across large document sets. Beyond efficiency, the firm faced strict requirements around data security, residency, and defensibility — particularly for Canadian clients who expect that sensitive matter data remains hosted domestically and is never used to train public AI models. The status quo imposed high per-matter costs, slower time-to-insight, and growing risk that legal judgment would be squeezed by operational burden.

BD&P partnered with Epiq, an Alternative Legal Service Provider (ALSP), to design and test a hybrid e-discovery workflow before committing to broader adoption. Using a completed matter as a controlled benchmark, the team compared traditional manual review against a workflow combining Continuous Active Learning (CAL) — trained by BD&P's own legal experts — with generative AI for issue identification and document triage. The integration kept legal judgment central: lawyers defined the relevance criteria and trained the model, while Epiq provided scalable delivery infrastructure. Data governance requirements were addressed through secure, access-controlled environments with clear data residency protections, ensuring client data remained segregated from public training models throughout.

The proof of concept produced a clear headline result: approximately 70% faster review timelines compared to equivalent manual review on the same matter. Beyond speed, the workflow reduced the burden on lawyers responsible for training the system, freeing senior time for higher-value analysis. Faster document triage enabled earlier identification of key issues, supporting strategic decisions at an earlier stage of litigation. The partnership model also introduced more predictable cost structures per matter. Qualitative outcomes included increased confidence in defensibility — a critical threshold for litigation work — and a repeatable delivery model that could scale to future document-intensive engagements without rebuilding the workflow from scratch.

• Collaboration between law firm, ALSP, and technology vendor is the core unit of deployment — AI tools deployed in isolation by any single party consistently underdeliver relative to integrated partnership models.
• Data governance is a prerequisite, not an afterthought — Canadian clients in particular require confirmed data residency, access controls, and explicit segregation from public AI training pipelines before adoption can proceed.
• Benchmark against a real matter before scaling — using a completed case as a proof-of-concept baseline produced credible, defensible results and accelerated internal buy-in.
• CAL effectiveness depends on lawyer input quality — the model's accuracy is tied directly to the precision of legal experts training it, making legal judgment an input, not just an output.