Future of eDiscovery: Balancing Technology and Human Judgement

AI is top of mind for many legal teams, but is it really going to make document review redundant? The short answer is no. While AI is an increasingly powerful tool in eDiscovery, legal professionals remain critical to unlocking its full potential and ensuring defensible, high-quality outcomes. 

AI as an Extension of Established eDiscovery Technology 

In the eDiscovery world, most teams have been actively using Continuous Active Learning (CAL) on large matters for quite some time, moving away from the traditional linear style review.   

CAL has been widely adopted for its ability to significantly reduce review volumes by prioritising potentially relevant documents and discounting irrelevant material. This enables teams to meet tight deadlines more effectively while delivering meaningful cost savings for clients. Importantly, CAL has gained trust because it is trained and validated by lawyers who understand the facts of the case, the legal issues at stake, and the threshold for relevance. 

Crafting Prompts That Power Smarter eDiscovery 

AI performance is directly influenced by the quality of the instructions it receives. Prompt development must therefore be a structured, iterative process, refined until results are stable and meet agreed accuracy thresholds. 

Lawyers will be critical to ensuring the system understands the key facts of the case and the threshold for relevancy in the review. To ensure the prompts are adequate, lawyers should run the prompts over representative samples of documents, and the AI generated results should be considered against legal team coding to ensure accuracy.   

This workflow will give the legal team the option to refine prompts and review the results to ensure they are confident AI clearly understands the parameters for review.  

Quality Control in the Age of AI: Human Validation as the Final Checkpoint 

Once the prompts have been run over the entire population, a second level review of the documents should be conducted by the legal team over the relevant population. This confirms if the system has categorised the documents correctly and ensures that the case team have a deep understanding of the client’s case, including any high-risk documents in the review set.   

It is also crucial that lawyers conduct checks over a sample of documents that AI has considered as Not Relevant, to ensure that no documents are missed from production. If the team identifies an issue with documents being incorrectly coded as Not Relevant, then they may need to make additional amends to the prompts and re-run AI.  

Some providers have helped assist with this element of review by providing the predictions on relevancy, the rationale for the choices, any counterarguments against the prediction and citations to drive further efficiencies in second level review.  

These additional elements make it easier for teams to have a greater understanding of how the system came to the conclusion, removing the veil from how AI made the decision.   

Beyond Review: AI in Action 

Once the second level review has concluded additional prompts can be run to generate chronologies, witness summaries and deposition outlines.   

As always, it will be important for the team to check the underlying documents have been referenced correctly.  

AI Amplifies, Humans Define 

Using these capabilities can drive efficient, timely reviews and help teams prioritise very large review sets – however it is clear that human review will still need to be at the heart of the process, whether that be prompt iteration, second level review or post review tasks.  

Streamline your legal workflows with Johnson Hana’s blend of outsourced legal expertise and human-guided AI for eDiscovery. To discuss how we can support your next review, contact deirdre.ryan@johnsonhana.com.