Your single source for new lessons on legal technology, e-discovery, compliance, and the people innovating behind the scenes.

Find Real Risk and Ignore the Rest: AI in Surveillance

Cassandra Morrison

Communication surveillance is hard. Data is growing exponentially these days, with employees using a myriad of different communication channels spanning chat, voice, mobile, and email platforms.

But this is nothing new to compliance leaders—and Aileen Tien and Peter Haller were not at Relativity Fest this year to share a bunch of old facts and figures on how to deal with these ongoing problems. Instead, during a session entitled “Finding Real Risk and Ignoring the Rest: Artificial Intelligence in Surveillance,” they debuted a new report by Opimas.

Aileen and Peter were full of lessons about how artificial intelligence is impacting communications monitoring, and how Relativity Trace’s data cleansing and new email thread deduplication. Here’s a recap of what they shared.

AI in Surveillance

“People are becoming better at behaving badly; just take a look at the headlines from around the world about financial misconduct,” Aileen says. “At the same time, regulators are coming out with new regulations to these evolving types of misconduct—from FINRA to the SEC, regulatory bodies are constantly updating their requirements and punishments.”

As Aileen goes on to explain, one of the biggest challenges with meeting these regulatory requirements is that the data that needs to be monitored is unstructured. Most corporate stores are comprised of unstructured data—up to 80 percent of enterprise data, based on many estimations. With multiple authors, and text that’s full of shorthand and abbreviations, conversations can also take place over multiple days, making it difficult to pinpoint misconduct.

Anna Griem, senior analyst at Opimas, joined to provide further insight around specific pain points that compliance teams are facing and how they’re dealing with them based on a survey of 150 financial institutions.

“Firms must be able to show both the reactive and proactive steps they’re taking to protect their organization from bad actors and misconduct. Fines levied reached their peak about five years ago,” Anna explained. “In the US alone, they’ve averaged about $15 billion per year over the past 20 years. Over 50 percent of these fines for financial offenses have centered around toxic securities abuses and investor protection violations.”

Anna went on to dive into the benchmarking report, listing the key components of a successful communications surveillance program: “While financial institutions are generally on top of recording and archiving requirements, few operate communications surveillance programs that are sophisticated across the board. For example, very few firms have programs with robust aComms and multilingual surveillance. Many also operate very siloed systems, with limited integration between systems.”

Eliminating noise from alerts is an increasingly difficult and important task firms must face.

Acknowledging that firms have long been hesitant to employ AI for both cost and defensibility reasons, Anna says the tide is changing.

From metadata to AI, finding ways to reduce false positives is becoming a high priority for both buy- and sell-side firms. Still, as Anna reported at Fest, the Opimas research found that buy-side lags behind the sell-side in terms of the sophistication of communications surveillance.

Where AI Comes In

Compliance teams are stretched thin to meet these regulatory requests. High false positive numbers can cause review fatigue, causing teams to miss real risk—which is where AI can help.

With Relativity’s recent acquisition of TextIQ, AI has never been more prevalent in our strategy. As Peter explained, Relativity Trace’s AI approach is to “deliver a robust set of purpose-driven AI capabilities that solve targeted challenges and, when combined, deliver acute detection of risk with significantly reduced false positives.”

With over 20 different AI capabilities analyzing each communication, combined with over 50 out-of-the-box policies, and flexibility to build custom risk detection solutions based on an organization’s unique needs, when Trace was measured against other solutions in Opimas’ report, it came out in the top tier. Announcing an ambitious plan to continue to up-level AI solutions to compliance teams, Peter and Aileen went on to describe the newest capabilities that have been added to the tool.

“Our newest data cleansing improvements, combined with our proprietary email thread deduplication, works to help analysts by reducing alert volumes and detecting risk. The three ways we go about this are by removing irrelevant content, pinpointing risk, and increasing the reviewer’s understanding,” Peter said.

Let’s dive deeper into how these three processes work to help compliance teams detect misconduct:

1. Remove irrelevant content. What is irrelevant content? From spam to newsletters to research reports, communications that are blasted to many people can result in a large number of false positives. Most analysts would agree that a newsletter from CNN does not contain any risky content, just like a marketing blast from Venmo doesn’t. But with many systems, they’d push an alert the same way an authored email would.

Another headache for compliance team members comes with other non-authored content, like confidentiality disclaimers, headers, footers, and email signatures. By utilizing algorithms to identify and remove these irrelevant alerts, teams can really dig into what matters.

2. Pinpoint risk. “Now that we have the unique authored content, we can identify whether this communication contains any risk that the compliance team should be alerted on. To effectively identify risk, we need a deep understanding of the people involved, we need to understand the communication itself, and we must understand the greater context of this message,” Peter said.

AI helps break down who the participants of the conversation are: who they work for, what their role is, what risks they are susceptible to, and their current risk profile. With this information in one place, teams can understand the relationships between those in the conversation and the type of risk that a given communication may contain. And then, the message itself must be analyzed. Using specific lexicons and out-of-the-box policies, analysts can begin to evaluate the true risk of the message. Lastly, understanding the context of the communication within a larger conversation provides the most accurate representation of risk for reviewers.

3. Increase reviewer understanding. As Aileen explained, in some cases, a segment of an email will have already been ingested and alerted on, meaning a reviewer has already read through that segment and decided whether it was risky or not—but it could still be served up as a fresh alert. The big announcement for this session was Relativity Trace’s debut of its proprietary email thread deduplication. The new feature automatically identifies these previously alerted email segments and removes them from the body of a newly ingested email, so that new and unnecessary or duplicative alerts are not generated.

With a continued focus on reducing false positives to highlight real risk for compliance teams and an invigorated devotion to further exploring how AI can help, the session revealed a lot of the headache-saving innovations built into, and on the roadmap for, Relativity Trace.

Some of the benefits, presenters shared, are already being realized: Customers have already seen false positives being reduced by up to 92 percent.

Artwork for this post was created by Kael Rose.

Download an Analyst Report on the Use of AI in Communication Surveillance


Cassandra Morrison is a senior specialist in content marketing at Relativity, with a special focus on the platform's communication surveillance solution: Relativity Trace.