AI Transforms Investigations Beyond the Search Bar

When Costel Ion, CFE, asked conference attendees to identify key challenges of using artificial intelligence (AI) in their investigations, many were ready to answer. They noted inaccuracies, hallucinations and concerns about data privacy and security.

Ion recalled the beginning of his career in cybercrime investigations in Romania, when Microsoft Excel was considered the best tool. He’s since worked for INTERPOL in Singapore and is now at Deutsche Bank. In one of his investigations, he and a colleague had over 10 years of data to manually sort through.

In his session, Ion demonstrated how AI can transform investigations. With the right mitigation strategies, he showed how these tools can lead fraud examiners to information beyond what they can obtain manually with a search bar.

During his session, Ion emphasized that mitigation strategies for using AI involve keeping humans in the loop. “AI is not going to replace investigators,” he stated. People will make the final determination for any information AI generates.

Ion explained that one way he uses AI is to generate a summary of policies when he conducts policy reviews. He advised session attendees to always verify AI outputs to ensure there aren’t inaccuracies.

He then walked attendees through the full life cycle of an investigation, including scoping, interviews, reporting, analysis, eDiscovery and document review. He noted that AI can be especially helpful in scoping when fraud examiners have many allegations and need to prioritize information. AI is also particularly valuable in eDiscovery and collecting information in emails and text messages. Ion also noted that if fraud examiners use AI for any steps in an investigation, they need to mention that.

Ion also explained that AI is perceptive in recognizing context. For example, AI can flag polarizing moods, such as the difference between the statement, “Here’s the report as requested” and the question, “Why are you questioning this?” Some statements, like the first one, are neutral, while others can be positive or negative, like the question.

Fraud examiners can also use AI for analysis. AI can make timelines of key events and recognize changes in behavior based on patterns. For example, AI can note in messages from a case when someone, instead of continuing to send emails, suggests switching to chat or suddenly stops communicating.

Ion noted that “AI doesn’t ‘know’ what is real — it only knows what sounds real.” He explained that this means AI suggests “the most statistically probable next word based on what it’s trained on.” It’s also important for fraud examiners to consult with privacy officers to protect sensitive information when using AI, as well as to remember that AI is meant to aid the investigator. “The investigator should do the hard work, and AI should support that work,” said Ion.