Testing controls using voice recordings and conversation notes is often very time consuming. Cognitive voice-to-text transcription services as well as natural language processing can improve the efficiency of these controls.
Challenges
Working with growing and complex data sets such as voice recordings and texts in multiple languages has become part of our daily business. In most financial services firms, working with controls based on such data sets is often very time consuming and inefficient.
Many examples exist. The obvious ones are MiFID and, more generally, investor protection controls related to recordings of client dialogues or notes from client meetings. Other examples of controls and processes based on these types of document include KYC, cross-border and travel controls, monitoring employee absences and training mechanisms to improve sales efficiency.
There are several aspects that create inefficiencies. First, it is expensive and time consuming to read, process and find specific material – such as statements of intent – within hundreds or even thousands of these data sets.
Second, traditional key word searches, which are still often used, return high amounts of false positive results due to context-free hits. Again, these false positives mean costly and time-consuming reading.
Inefficiencies can also arise as a result of the sample sets being small and sometimes ineffective.
Finally, in many cases, these inefficiencies span all three levels of defence.
Raising the risk bar with cognitive technologies
In recent years cognitive technologies have shown their potential to automate core business processes across the financial services. More recently, cognitive technologies have been used in risk and compliance matters. The two main drivers are higher control quality and coverage, and increased efficiencies. Control quality and coverage has increased because the testing sample is larger and because all process steps are tested using an unbiased model. Automating time-consuming tasks has also resulted in greater efficiency.
Cognitive technologies will improve investor protection since a large part of the control subject is human language and the conversation has several media breaks, i.e. where the two cognitive technologies – natural language processing and voice-to-text – combine.
State of the art voice-to-text capabilities make it possible to transcribe recorded conversations, such as phone calls with multiple speakers across different languages, and to store them as searchable and machine-processable text. The quality of this transcription has now reached a level that makes it ready to use in many use cases.
Natural language processing identifies relevant concepts and entities in human language. The extraction of relationships between the entities makes it possible to validate relevant control questions. This can either be applied to text transcribed from voice recordings or to existing documented notes.
Natural language processing and voice-to-text hold huge potential and will increase the efficiency and effectiveness, as well as quality and coverage, of investor protection controls.
Why now is the right time for you
Controls are multiplying and complexity is growing, and financial service firms are seeking to improve their internal control mechanisms.
Cognitive technology has proven its value for marketing and customer management, with large amounts of conversational data being analysed and used to derive actions. Why not also leverage it for controls? It is clear that by not doing so, benefits will be missed.
- Manual work does not scale. You may have well-established control procedures but you will need to change these eventually, as manual work only addresses a small fraction of data – and that is not effective. The opportunity lies in combining industry knowledge with technology to increase scale and coverage.
- Leveraging technology means learning to deal with patterns. Finding out about behaviours gives you power. It also means starting to structure your actions with compliance in mind and changing the game as you move forwards.
The public perceive control management as part of a financial services firm’s core responsibilities. New generation clients want you to leverage what you know. And to do your best to build an organisation that is self-aware and always improving.
While you need to provide the capabilities inhouse, once you establish them, your expertise can become a service (EaaS). This can be done by establishing a centre of excellence for compliance and controls that helps other (financial and non-financial) institutions. Another approach could be the establishment of an API market and control code framework.
Shape your own future – it pays to be proactive about finding your own path to improvement.
Key takeaways
Humans still have qualities that machines cannot entirely reproduce. Yet, in a closed game with set options, as Garry Kasparov said, machines make fewer mistakes than humans. It is not that they solve the equation, but they are faster at becoming experts in large amounts of data when trained well. Before there were cameras, you had to believe the stories, drawings and reports of others. Now we have a lot of pictures, i.e. data. When we apply this to areas that require more technical rigour, why would you do a hand sketch based on a small number of photos, rather than analysing all of them for potential flaws? Disruptors do not always come from the same industry as you, but why not learn from other use cases and borrow their tools?
