On September 1st, Eventus Systems and VoxSmart put together a joint statement comment letter regarding FINRA’s June 2020 report, “Artificial Intelligence in the Securities Industry.” The broader acceptance and use of artificial intelligence (“AI”) technologies continues to advance across the financial services industry and has become much more accessible to broker-dealers through both internal development and experienced service providers.
While many firms are familiar with some uses of machine learning (“ML”) or natural language processing (“NLP”) in trading and surveillance platforms, their applications are often intermittent and reactive. And, as firms implement new technologies to help scale and improve surveillance, the increased use of machine intelligence presents many operational and regulatory challenges which, if not addressed, could hinder further innovation.
A more proactive adoption of these technologies is needed so firms benefit from the efficiencies artificial intelligence can bring to their organization. In parallel, continued guidance from the Regulator is critical to ensure firms successfully navigate the complex legislative areas such as recordkeeping and data governance, among others, without restricting the innovation and engagement with AI assisted tools. When exploring the application of AI technology to digital communications surveillance, let’s consider FINRA Rule 3170 which addresses voice and video recordings.
FINRA Rule 3170
FINRA describe the term “tape recording” as including any electronic or digital recording that meets the requirements of Rule 3170 however, the rule also refers specifically to “telephone” conversations, which may generate some confusion around the scope required by firms. The terminology arguably falls short of many modern communication tools and there is a call for updated guidance on the required voice, mobile and video recording requirements.
While phone conversations are required to be recorded, voice channels in general have always been under-monitored, with any controls being manual rather than automated. In the past, a barrier to this has been ineffective transcription technology which has created more noise than informative insights.
While the monitoring of audio channels is not covered by rule 3170, a lack of sufficient and consistent surveillance coverage leaves firms open to extreme levels of risk on the channels they are already recording under the FINRA regulation. Advancing NLP and ML technologies, offers firms the ability to scale and improve surveillance by covering many accepted digital channels within reasonable time frames.
Voice Transcription tools will help firms meet FINRA rule 3170
The effectiveness of a surveillance program heavily relies on the quality and quantity of data available to the surveillance system. While there are many complexities surrounding voice and video surveillance, firms should not de-prioritize these channels and the controls in place for their use.
Machine Learning methods which answer these surveillance complexities will be extremely useful for firms wanting to take a more proactive approach to their market surveillance. Voice transcription in particular is one area in which Artificial Intelligence can aid compliance with FINRA rule 3170.
Intelligent speech-to-text engines that apply natural language processing and machine learning approaches are particularly powerful when used in conjunction with other rule-based methods.
At VoxSmart we use AI to enhance the accuracy of our voice transcription engine. We take a targeted approach that considers both the context of the communication and the content of the conversation to generate high quality transcriptions which deliver the correct meaning of the audio call, not just a literal translation of the words used.
The financial sector uses a lot of industry terminology, slang and trader jargon which, to many generic transcription engines, is very difficult to convert into meaningful text. We have built a proprietary financial language model which is able to recognize financial terminology (using natural language processing), learn where it is used or should be used within a conversation (using machine learning) and generate accurate transcriptions in near-real time.
Our recommendations to FINRA
The joint comment letter highlights that as firms employ advanced surveillance and monitoring techniques using AI, the market would benefit from updated regulation and guidance relative to books and records requirements to consider. For example:
- Guidance or expansion on the umbrella of required voice and video recordings
- Updated expectations for capture, storage and surveillance of an expanding set of digital channels (text, voice, video and productivity tools such as Slack, Teams or others)
Continuing guidance as digital channels and other collaboration technologies evolve is critical to widespread use and efficiency of modern surveillance tools. We believe that a rapid, controlled, and uniform adoption of AI for trade and communications surveillance will enable supervisors and compliance officers to manage risk in a highly complex, interconnected, and evolving environment.
Both Eventus and VoxSmart strongly support FINRA’s initiatives concerning the adoption of these new technologies into the regulatory framework governing broker-dealers. To read about how the use of AI tech in trade surveillance applies to FINRA Rule 2010, please visit here.
If you have any questions about the joint comment letter or the application of AI in the securities industry, please don’t hesitate to reach out to us at: firstname.lastname@example.org
We have gathered six months of capital markets communication data and produced a high-level report highlighting some of the key insights and trends we have witnessed unfold since the start of the global pandemic. download now.