In a highly-manual world, AI is the true transformer.
Indeed, many workflows, actions, and processes that have historically relied upon wholly-human activation are now ripe for intelligent, labor-saving automation that can turbocharge efficiency.
In one area in particular – regulated communication surveillance – AI can punch way above its weight.
Here, in this most high-risk and resource-heavy of environments, Machine Learning, for example, can deliver disproportionate levels of cost savings whilst simultaneously mitigating many times more risk than the once-essential human equivalent.
Those so-called ‘reviewers’ - highly-trained surveillance experts tasked with spotting potential compliance breaches in regulated communications - are expensive and often required in large numbers. Banks and law firms, for example, may employ hundreds – all reading thousands of emails, messages, and voice call transcriptions every day, looking for risk.
AI can now do much of that heavy-lifting; automatically, 24/7, and with its eyes wide shut.
To capitalize, firms (and/or their technology service providers) must partner with a vendor whose solutions come with all of that capability baked-in.
“In the past, regulated firms would have a rarely-updated lexicon of phrases; a dictionary if you will, of words and phrases which, to their trained human reviewers, indicated potential non-compliance - today, thanks to AI, those businesses can work so much smarter,” says Chris Stapenhurst, Senior Principal Product Manager at leading data management experts Veritas, whose discovery, surveillance, and file analysis products leverage AI to the max.
“Machine Learning, for example, combs through every communication automatically and turns potential regulatory violations into alerts so that human reviewers can scrutinize them more closely.
However, most importantly, the Machine Learning can be programmed to ignore millions of words or phrases in, say, junk emails, which may appear risky but which the firm has identified as innocuous, such as ‘Out of Office’, ‘prohibited’, or ‘unauthorized’. That can reduce the number of false positives by up to 98% - and support a significant reduction in reviewer headcount.”
Of course, the best surveillance strategy is multi-layered: a mix of old-school, manually pre-built lexicons plus AI-powered add-ons. Firms and their service providers should ask whether vendors’ solutions go beyond the ability to search for just basic terms and phrases. Do they model behaviours? Do they leverage sentiment analysis?
Whilst ticking all those boxes, the Veritas solution also – and uniquely – constantly teaches and updates itself 24/7 based on human reviewer interactivity. Its Machine Learning engine sits inside customers’ UC platforms and, as reviewers are marking and labelling items, it begins to understand the difference between an item marked relevant or irrelevant; risky or innocuous.
“It looks at a variety of different parameters and basically conducts its own classification,” says Stapenhurst.
“It examines the communication’s metadata, when it was sent, who it was sent between, what the subject line was, what the content is.
“Many of our competitors take a different approach. For example, in financial services, they may hire an expert in market manipulation and put them with an AI expert to create a surveillance model based on their joint expertise. They might work together for weeks or months to create a super-accurate lexicon but, the very next day, it’s out of date because it is a non-evolutionary entity which might only get re-visited every quarter.
