Video surveillance data: Useful or not?

The widespread adoption of video communications platforms such as Zoom and Microsoft Teams during the pandemic brought to the forefront new concerns about compliance and conduct risk for traders working remotely. Now over a year later, as traders and bankers start making their way back to their offices, video communication platforms are here to stay.

“What we are seeing is that those behaviors of using these tools are now ingrained,” says Mark Whiteman, CEO of UK-based communications technology provider Citycom Solutions. “Everybody is used to collaborating and communicating this way, which they were not pre-pandemic. Even if people do go back into the office full-time, which I very much doubt because there are a lot of benefits of remote and home working—both financial and well-being—they are going to use these new tools.”

Last summer, Citycom launched a solution that combines different monitoring and audio-visual recognition tools to address the compliance needs of traders and other financial markets professionals working from home. The offering includes facial recognition, which can help confirm the identity of a video caller, as well as object recognition to determine whether the person under surveillance is holding any item that is prohibited or presents a security concern. Whiteman says some Citycom clients are in the early stages of deploying the vendor’s solution, while others are still testing it or are in talks about it.

Video communication can present risks such as insider trading, new cyber threats, regulatory compliance issues, and conduct risks. Video surveillance could be helpful, for example, if a trader on a call says they will keep their stock positions as they are, but simultaneously holds up a card signaling to sell a stock. “You’ve got the message, but if you were just monitoring the audio, you have not got the message,” he says. “And there are other hand gestures, mobile calls, and so on that can be counter to what is going on the audio.”

Vinod Jain, senior analyst at Aite Group, says communications surveillance pre-pandemic was focused around the trading desk, the phone network, and chat. But now that surveillance must include video conversations. 

“The challenge is that it is all unstructured data; you can’t follow the conversation,” he says. “There are no numbers or text associated with it that you can put into a database and query and look at more formally. I think that’s the larger challenge every firm has: to identify what’s going on across all these channels.”  

The skeptic 

The usefulness of video surveillance for compliance depends on who you ask. Oliver Rooney, CTO at compliance vendor VoxSmart, says no one is really sure why they want to surveil video calls, or what they want to get out of analyzing video data. He says in the last couple of years, his firm has only had one concrete use case for video processing and surveillance.

“That was a very niche case where someone was using WhatsApp video calls in a know-your-customer setting. They were onboarding customers, it was a wealth management kind of platform, and they were doing their KYC via video call. So they would say, ‘OK, now hold up your passport, and now I need to see your face,’ to verify it’s the right person,” Rooney says.   

He says in the medium-term, video surveillance will be used to better understand the context within a communication, rather than as a deep source of additional information. “Personally, partly by nature, and partly because of the technical challenges, I am quite skeptical that there will be a lot of detailed analysis that happens if that video stream itself is the actual video data, certainly in the near-term,” Rooney says.  

He says firms could spend an enormous amount of effort gathering and analyzing video and not get much out of it. “I think those use cases are limited to things like face recognition, sentiment analysis from face recognition, expression recognition and that kind of thing. But again, that is trivial stuff to do,” he says.  

Rooney says the technology needs improvement. “If you look at the kind of trials of face recognition by the police, most of them have not resulted in a single conviction, because the technology just isn’t good enough to do it,” he says. “So you are asking a lot, I think, to get really useful, actionable intelligence from the video stream at the moment. Perhaps I am too cynical, but that’s what I think.”

However, Whiteman says Citycom’s facial recognition has a high degree of accuracy. The vendor’s process of onboarding someone for surveillance requires only 30 seconds, with the subject turning their face in different directions as 600 pictures are taken to build facial recognition.  

“It is certainly far more accurate than we could ever achieve with voice biometrics,” Whiteman says. “Voice biometrics are popular with the retail banks, in terms of call centers or even talking to ATMs, but this is far more—this is in the high 99% accuracy [range] when deployed correctly.”

The surveillance can also be configured around how frequently a client wants to check the face of an employee to make sure they are present. The tradeoff is that more frequent processing of a face would require more computing power.   

Similarly, Whiteman says the accuracy of detecting an object such as a certain brand of mobile phone or a bottle of alcohol, depends on how well the system has been trained to recognize it. For example, if a firm wanted to check whether their traders were drinking vodka, they could specify the brand they wanted to monitor and onboard it onto the system. 

“We can onboard to that level of detail, but not all of it is in the database; that would need to be trained in. And obviously, the more that we train in, the more the database grows,” Whiteman says. 

Waiting for the regulatory kick

Responding to skepticism about the usefulness of video surveillance, Whiteman says banks do not tend to take action “unless they are being kicked” via litigation or regulatory investigation.

“The regulators don’t say, ‘It’s voice or it’s Bloomberg chat, but it’s not Symphony chat.’ They say it’s all communications. If you read Dodd–Frank or Mifid, it’s about pre- and post-trade communications, and you should be capturing all the communications that relate to it. If a bank just rolled out 55,000 Zoom accounts of which 5,000 are traders, surely that constitutes what should be part of compliance. The banks will argue it until they get slapped across the wrists and told they should be recording.”   

Since the launch of Citycom’s solution, the vendor has been updating it with new features, including adding audio sync to the video surveillance. “We can use both [someone’s] words and the computer visual to surface events of interest,” Whiteman says. “So it could be the combination of me saying, ‘I don’t think we should sell anything in this volatile market,’ and me holding up a note that says, ‘Sell Tesla.’”

In March this year, Citycom configured its offering to Zoom, Microsoft Teams, and WebEx. Previously, clients had to send the vendor recorded files from these platforms to process. Whiteman says more than 70% of the banks his firm has spoken to have embraced the use of these video platforms.

At the same time, video surveillance is still at an early stage. VoxSmart’s Rooney says he plans to add more video surveillance but will take a phased approach, first focusing on audio surveillance, and then treating video as an additional source of context. 

He says for compliance investigations, in specific cases, a lot of contextual information is needed, such as market data and news information “These are all additional sources of information, and we see video as going along with that to fill in some of the extra blanks, whereas we think the policies, the alerts, will still be driven mostly by the text and trade data that goes along with it,” Rooney says.

He says his firm is still having conversations with clients about the potential use cases for video data, and looking at what extra information can be extracted. If a trader wanted to engage in fraudulent activity, Rooney says they would not use a recorded channel in the first place. 

“You are trying to detect behavior that you don’t like, rather than outright and bare-faced wrongdoing. Most people are using those communication channels to do things that might be wrong, but they either think of them as [being] tacitly allowed, or they don’t believe they are wrong at all,” Rooney says.