Using Generative AI to Address Internal Investigation Challenges

Investigators are trained to question and verify. So, introducing generative AI into the investigative workflow requires attention to questions such as:

• Who hosts the model and where is it geographically located?
• Does the usage of the model “learn” from my data?
• How might algorithmic bias and others risks such as hallucinations affect the results?

One approach to answer these questions and build confidence in the use of generative AI for certain investigative tasks is to conduct contained experiments. For example, asking a large language model to handle a portion of the review process, such as classifying a subset of documents, while human investigators validate the results. This can foster confidence in the technology without risking substandard results. 

Improving investigative processes

A recent report from the Association of Corporate Investigators (ACi) highlights several common and persistent challenges in internal investigations. These include limited resources within investigative and legal teams, vast datasets that must be included within the discovery process and cross-jurisdictional complexities. With expert oversight and application to targeted tasks within an investigative workflow, generative AI can help solve these challenges in the following ways:

• Allegation triage: Identifying potential issues and streamlining the initial assessment phase by analyzing complaints and whistleblower reports.
• Document drafting: Crafting initial drafts of essential documents, such as terms of reference or briefing notes, saving time and helping to surface key information faster.
• Interview summarization: Enabling investigators to focus on analysis versus administrative tasks by transcribing and summarizing interviews (especially helpful in employment-related investigations).

FTI Technology has worked with clients to leverage generative AI in investigations through IQ.AI by FTI Technology, an advanced suite of AI-driven solutions designed to address data-intensive challenges in legal and compliance matters. Within the context of corporate investigations, IQ.AI offers a flexible and secure platform for accelerating review, identifying relevant content and classifying privileged material. 

Evaluating outputs for accuracy

Assessing the quality of generative AI outputs is critical. Traditional metrics like precision and recall, which are used to determine the efficacy of predictive coding, remain relevant for classification tasks. Metrics for applications like summarization and translation will rely on established natural language processing evaluation metrics to provide performance insights. 

Additionally, investigative teams may use one large language model to evaluate the output from another. Though innovative, such techniques should be applied judiciously, with evaluations still grounded in established standards.

Maintaining data privacy and security 

Investigations often involve cross-border transfer of data. As privacy remains a top concern for organizations in many jurisdictions, it's essential to ensure data that is being loaded into generative AI models is encrypted in transit, not stored at rest and not used to train models. Such measures help to safeguard sensitive information and maintain compliance with data protection regulations.

Additionally, within FTI Technology, models are hosted in multiple regions. This allows the team to use models within designated regions, ensuring that data remains within the region and complies with local data residency requirements where applicable. 

Integrating generative AI into corporate investigations workflows offers a path toward more effective processes. By addressing concerns around reliability, verification and data security, organizations can harness the benefits while continuing to uphold the rigorous standards investigations demand. 

Technology will keep evolving, and so too must the investigative approach — ensuring innovation complements, rather than compromises, the integrity of investigative work.

Related topics:

The views expressed herein are those of the author(s) and not necessarily the views of FTI Consulting, its management, its subsidiaries, its affiliates, or its other professionals.