Editor’s Note: This week, we’re diving into how AI helps compliance teams keep up with changing regulatory priorities by reducing alerts so teams can focus on real risk. Read this two-part series to learn how technology helps by eliminating duplicative text, analyzing and contextualizing content, and transcribing multilingual communications.
Once your team has started the process of building an AI-based surveillance system to help cut through the noise of huge data volumes, the next step is to find real risk.
Advanced surveillance systems today favor behavioral monitoring instead of merely hunting for certain words and phrases. The use of conduct risk rankings gives a rounded view of an individual’s behavior, replacing the need to dig through large amounts of data for “red” incidents. Pinpointing risk in this way helps identify trouble before it hardens into more serious breaches.
Benefit #1: Enhance Review of Diverse Data Sets
Having captured and distilled the relevant communications data as outlined in part one of this series, the next step is to analyze the content for actionable breaches. Machine learning algorithms can be trained to sweep conversations for cases where an individual is demonstrating signs they have, or will, break the rules.
Structured data, like metadata including the author, file size, and the date a document was created, is simple for legacy surveillance systems to search and analyze, as it exists in a predefined format such as tables like Excel files and Google Docs spreadsheets. However, analyzing unstructured data is much harder. Log files, audio files, and image files are often impossible for legacy surveillance systems to cope with.
Multi-language and Vernacular Challenges
Any global business will have multiple languages spoken as workers are drawn from a much wider geographic pool, but the dearth of options previously available means most firms do not monitor beyond one language, usually English. Advanced AI products have audio transcription tools and can be trained on various languages and dialects, along with offering multilingual search capabilities.
Lexicon policies are often relied on to find certain words and phrases used in email. Difficult to adjust and insensitive to more subtle misconduct, they are narrow and belong to the previous generation of surveillance. Their success depends on identifying an actual word or sentence, which can occasionally be enough to trigger an investigation, but casts a huge net when applied to the mound of data to be sifted through. Lexicons also miss the mark when it comes to language switching, colloquialisms, market slang, linguistic flexibility, and pidgin English.
AI can step in and fill in the blanks when employees may be using their own made-up codes to collude or rig markets. Unstructured categorization AI can organize documents into named multi-depth groups based on the topics discussed in the text. This helps identify topics that seem out of place, such as heavy use of unidentified code words in sentences.
Audio is not just a different channel, it’s a completely different data source and successfully monitoring it without AI is nearly impossible. Transcribing phone and video calls is far from a perfect science, with various factors—such as a speaker’s annunciation, background noise, or network disruption—affecting the transcription quality. AI-based systems can offer up alternate words and phrases that may add context to these communications and make review even easier.
AI tools such as sentiment analysis, entity recognition, and market manipulation models can be deployed alongside lexicon searches to further reduce the number of false positives dealt with by compliance teams. By filtering the data using AI tools, you can find content most likely to contain actionable breaches. Sentiment analysis, entity recognition, email threading, and other features can be used alongside lexicon searches for the most accurate results.
Benefit #2: Provide Deeper Context for Reviewers
When suspected misconduct is identified, an alert is issued to be reviewed and then potentially investigated further. Here, a compliance officer needs context—an understanding of the conversations that took place and led to the suspected breach.
If the flagged data is an email, a reviewer will want to see the rest of the email conversation in as simple a format as possible. They will want to assess inline comments or attachments in other emails in the thread group that may have made their way in. This unique content will be important to establish context.
NLP tools can parse unstructured documents and extract this key information, presenting it in a simple visualization to show how the conversation unfolded. Key variations in documents can be highlighted, allowing reviewers to move between documents with speed.
Examining specific sentences is extremely valuable when trying to locate similar content, as it can unearth communications a reviewer may not have been sure existed, going beyond the limitations of standard terms-based rules.
Other hidden risks include instances where communications have completely different text, but the topics being discussed are the same. Conceptual similarity capabilities inside modern AI surveillance tools can locate documents contextually, adding the complete text or specific sentence/paragraphs in the full alert.
A reviewer can then identify important information without knowing the specific terms, phrases, jargon, or code words that may be used by employees.
Analyzing the other people in the suspect conversation can lead to other documents important to the event. AI that groups aliases (e.g., email address, phone number, et cetera) to a single individual, paired with visualizations that map participants, can help locate communications between two specific individuals or during off hours and can help understand the participants’ social dynamics.
If a case is to be escalated, either to a larger pool of reviewers or to another department, such as legal, findings may have to be passed on. It is more efficient to use a tool that supports both teams and can transfer the documents with all the original alerts in just a click.
Empower investigators by giving them the full context for a suspected breach, showing relationships, conversations, and more that will help determine if misconduct occurred. Uncover hidden risks through AI tools that can identify contextual similarities and map suspect relationships otherwise invisible to human eyes.
Benefit #3: Adapt to the Changing Surveillance Landscape
Pressure is increasing on compliance teams to rein in misconduct. A new strategy is needed to target conduct threats that have shifted from physical offices to the digital realm.
AI capabilities are being used by innovative businesses to uncover patterns in the conduct of individuals through their communication data, stopping problems before they escalate. Compliance teams swamped in data generated by digital communications can use AI tools to close off new avenues of risk, discard superfluous data, and protect privacy whilst homing in on actionable content.
This marks an evolution of surveillance beyond the “smoking gun” detection process of legacy technology to a “breadcrumb trails” approach, where human compliance staff can harness powerful technologies to confidently and defensibly protect their organization.
With the right strategy and tool, like Relativity Trace, compliance teams that use behavioral analysis and predictive analytics stand to significantly reduce false positives and find real risk.
Mark Taylor is a compliance expert and a writer for Relativity.