Owlin Solution Talks: How Owlin Reduces False Positives When Monitoring Risk

Sjoerd Leemhuis About Real-Time Risk Monitoring in TPRM and Reducing False Positives

Effective third-party risk management is more critical than ever in today’s fast-paced and interconnected world. Companies need to identify risks as they emerge, not after the damage is done. But while a lot of data is available, filtering out noise so you only focus on what matters can be challenging. In a new Owlin Solution Talk, we ask Sjoerd Leemhuis (CEO) how Owlin tackles this challenge.

Why is real-time risk monitoring becoming increasingly crucial for third-party risk management?

“Traditionally, organizations relied on manual adverse media checks, credit ratings, and watchlists, like PEP or sanctions lists, to check if third parties were exposed to risk. However, these only flag entities after events have passed or investigations have started.

In today’s risk environment, you need a scalable solution embedded into your daily risk workflows. Monitoring more real-time data sources—such as adverse media and consumer reviews—is no longer a ‘nice-to-have’ but a necessity. However, filtering out irrelevant or unreliable data is challenging.”

What do you mean by irrelevant or unreliable data?

“One of the reasons we originally chose adverse media as a source for risk insights is that it’s, on average, better-curated content than posts on social media. However, fake news is a significant issue because it can appear on seemingly legitimate platforms, especially newer websites, and spread quickly through social media. Even well-established sources like the BBC might occasionally publish inaccuracies. Furthermore, with the rise of ChatGPT, the number of fake reviews has surged.”

How does Owlin then ensure the credibility of its risk insights?

“Our first line of defense is ensuring we only track reliable sources. We’ve been leveraging most of them for over a decade and have developed a robust system for ensuring transparency. We do severe quality assurance with manual checks and AI for source validation. In this process, we analyze the source’s content to ensure it meets our high standards before we start tracking. This helps us maintain a reliable monitoring ecosystem.

Because we own our own data pipeline, we can also provide complete transparency in where our insights come from. Even if a validated source shares inaccurate information, we can show where this article was first published and which other sources report it online so the client can make informed decisions.

What’s your second line of defense to provide credible risk insights?

“Our second line of defense is the code of our platform, developed by our data scientists and technical engineers. Through ongoing improvements to our AI and machine learning algorithms, we ensure that false positives—instances where a risk alert is incorrectly triggered—are minimized.

Can you give some examples of this?

“One key way we reduce false positives is through contextual AI and NLP (Natural Language Processing). Our platform uses NLP to analyze news and adverse media in context, meaning only the relevant content gets flagged, and generic mentions or unrelated stories don’t get picked up.

In more detail, we use entity recognition and disambiguation to help identify the company (or person) of interest. Next, we identify more than 300 issues ranging from financial crime to ESG risk. With our contextual in-house developed AI models, we also know if the issue is actually about the company we’re tracking. So we look at the context of the sentences to identify the subject (i.e., distinguish between subject and direct object in a sentence.) to ensure that the company you’re interested in is the story’s subject.

For example, company X lost its most significant customer, Y, to competitor Z and is filing for bankruptcy. Three companies are mentioned in this case, but only one is in trouble. Owlin’s models have the intelligence to recognize that this is only bad news for company X. So it won’t give hits on adverse media for companies Y & Z.

We also allow users to customize the platform to their specific needs. You can define keywords, risk categories, and filters focusing on the issues most relevant to your business. Fewer irrelevant alerts mean less risk, precisely what you want when monitoring third-party risk. And, of course, we don’t just do this in English. Our platform analyzes content in 17+ languages so we can capture local context and nuances, which helps avoid errors that might come from mistranslation or misinterpretation.

With these layers in place, we can provide real-time insights that are accurate and actionable while minimizing false positives as much as possible.”

What’s next for Owlin in terms of data coverage and capabilities?

“We’re constantly expanding. For example, while we currently avoid social media due to its low signal-to-noise ratio, we’re exploring other data types like video, podcasts, and external risk signals from financial statements and qualitative data. We’re also working to broaden our language support. The top 20 languages already account for 90% of internet news content, but our end game is to have support for every language in the world to make our coverage truly global.”

What’s Owlin’s vision for the future of risk monitoring?

“We continue working on our mission to shape a better-informed world by further expanding and improving our data ecosystem. And while for all the new sources we might add, the challenge of false positives will never disappear entirely; I know our proactive approach minimizes the noise—leaving only the signals that matter.”