Recent breakthroughs in generative AI and machine learning technology present one of the biggest leaps in our ability to detect, analyze, and manage risks. Contrary to what many believe, AI is neither a wholesale replacement for human researchers and analysts, nor is it a passing fad that (like blockchain) will fail to live up to many projections.
Instead, AI’s application to risk research solves a series of specific pain points that have plagued researchers and analysts for decades, and that have allowed vast swathes of risk-relevant data to go un-researched and, by extension, risks to go undetected.
These pain points have been summarized below in more detail, but in a nutshell they relate to the ability to analyze high volumes of unstructured data quickly and cost-efficiently, and detect potentially risk relevant information. “Unstructured” is the operative word – while AI can also be effective in analyzing structured data more quickly (such as subscription-based compliance, regulatory, and adverse media databases), it is the messy and often difficult to detect data on the broader internet that has long been the source of key risk and business intelligence findings.
As any seasoned risk researcher can tell you, most material open-source red-flags are spread across local media, blogs, professional and consumer websites, message forums, NGO reports, and other messy unstructured internet data. The process of identifying such information is a complex combination of exhaustive process, creative problem-solving, and “following your nose” sleuthing that represents the “secret sauce” in most high-end open-source risk advisory work.
At Integrus, our integrated team of business intelligence and engineering professionals have spent years researching the best ways to apply AI and other emerging technology to this problem. Tech-driven research models work on a statistical basis, and at root are so wholly different from how a human approaches such problems that a direct imposition of technology onto the traditional research process often gives fairly weak results.
Instead, we have found that the key is to leverage the strength of technology – as summarized below – to mine and process vast amounts of both unstructured and structured data, far more than what a human (or even a team of humans) would be able search and review. We then leverage breakthrough semantic detection capacity of emerging technology, including proprietary analytical models developed by our in-house team of mathematicians, to scan this vast amount of data for material information. Information that is identified – references, statements, allegations, etc. – are methodically analyzed and extracted in a structured way for human review.
This technology is useful for a variety of risk research types: all the major variations of due diligence (DD), screening, business intelligence, industry analyses, compliance checks, etc. This functionality can serve both as a standalone product and an invaluable tool in human-driven research processes. For example, the results of this AI-driven tech allows analysts to hone in on material information on day 1 of a project, rather than having to spend the first days (or even weeks) hunting/scrolling/CTRL+F’ing for needles in a haystack.
This means that more time can be spent on identifying additional context, corroboration, leads, and ultimately evidence than would otherwise be possible on the same timeline and within the same budget.
AI-driven research also improves the quality of source inquiries. For those in the industry, it is widely known that the key challenge with source inquiries is timelines. Beginning source inquiries before having completed open-source research means there is less context based on which to conduct interviews, meaning that they often yield suboptimal results. Conversely, waiting until open-source research is manually completed means that inquiries are conducted later in a project timeline, with less time to capitalize on new information and feed it back into the overall research process. Fortunately, when a researcher is already equipped with the results of AI-driven DD at the beginning of a project, then source inquiries can start to be conducted earlier and more effectively.
Regardless of the various ways that AI’s impact on the risk industry has been underestimated and overestimated, the real point is that AI allows us to identify and analyze more risk than in the past, mainly in the following ways:
How AI Is Transforming Risk Research
Speed
Algorithmic and machine learning models are able to acquire, mine, deduplicate, and analyze large volumes of data far faster than a human (or even a team of humans) can
Consistency
Whereas humans can only perform searches individually, our AI-driven models can algorithmically exhaust search possibilities because there is little concern of having too much data, given how quickly the AI-driven model can review content
Accuracy
AI-driven DD can take into account corporate profiles or biographical details when zeroing in on relevant information out of the vast amount of data fed to it
Scale
Subjects can be run in parallel, meaning that several thousand subjects can be screened within days
Customizable
AI-driven DD can go beyond standard scopes, and scour the internet and other open sources for data relevant to specific risk scopes
Cost effective
While the data feeds and computing capacity necessary to power AI-driven research and analysis processes do incur cost, it is far more cost-effective than a human analyst