5 reasons your Machine Learning Algorithm need a serious make-over

Fonetic Team
Friday, 17 May 2019 / Published in machine learning

5 reasons your Machine Learning Algorithm need a serious make-over

Machine Learning (ML) has been the industry buzzword for a few years now. But, is this ‘fashionable’ AI technology really ‘flawed’ in its approach to regulatory compliance? Not to mention when covering best execution for MiFID II, MAR and Dodd Frank.

The RegTech landscape is changing, and fast. Are you ahead of the curve? 

AI superpowers outside of Silicon Valley

So, where are we now with RegTech?

The recent fame of AI-powered technology isn’t all whizzy neural networks and lighting fast analysis. It has given the perception that state-of-the-art surveillance tools can’t achieve high accuracy levels in audio channels such as mobile, instant messaging or fixed line phone calls.

In fact, there is still a widely held view that accurate speech to text is still a pipe dream.

There are, however, tools available that utilise ML technology, which are being used today to spot abusive behaviour and conduct risk/misconduct.

Our blog post on why Machine Learning makes compliance teams more effective, could not be farther from the truth in today’s industry. Machine Learning is revolutionising RegTech but our neurotic techy side in us need to make sure it’s used correctly.

In this post we will discuss what to watch out for when choosing AI-powered solutions and share how effective Machine Learning can give enormous value to compliance professionals and the Front Office.

Is your AI ‘Fashionable’ or ‘Flawed’?

Artificial Intelligence (AI) tools are growing in popularity among sell side and buy side firms. Solutions using AI and Machine Learning techniques are a great way to achieve better quality supervision and surveillance. Although these tools have help[BA2] ed achieve higher accuracy rates and better predictive tools, so far, Financial Institutions have only scratched the surface of what’s really possible.

There are many new solutions on the market today that use ML technology to monitor employee behaviour and deviations from the norm. This technique can indicate possible employee wrong doing. A greater focus on interpreting the data and giving “insightful” output rather than enhancing the quality of the input data has led to these results being next to useless. This is because the transcription quality derived from voice recordings is often an afterthought.

The adage, garbage in, garbage out has never been more true.

A “Best of breed” approach

As a result, the more experienced firms, having been first adopters of these solutions, are now going full circle and returning to a ‘best of breed’ approach for trade surveillance, voice and eComm surveillance with light touch integration to being the approach to ‘holistic’ surveillance.

So, what can we learn from these firms? While we’ve been working for 10 years perfecting their Voice Surveillance for the Trading Floor, we’ve learnt a thing or too.

Without further ado, Here’s our best practise tips for ML below.

What should compliance teams do when adopting Machine Learning in sell side and buy side finance?

1. Avoid multipurpose engines

Multipurpose engines trained on Big Data from these companies are not aimed at the capital markets space. Instead they are made to understand the highly lucrative retail space. This means they have proved unfit for purpose when applied to Investment banking or broker trading.

If too generic data is used, surveillance systems will be unable to accurately detect the complex nature of trading jargon or financial specific phrase structures.

2. Don’t rely on your own activity to set a good example

Some vendors approach the problem of having too little data to train their models by using the customers activity to train the AI engine. This is an insular approach that will only be fit for purpose within the firm. It has failed to utilise the added functionality gained from having the system deployed across multiple Capital Market clients.

3. Chose market specific AI models

Within the trading community there is frequent use of trading language and slang. In many cases this vocabulary can even be specific to an actual asset class. For example, the terms used by FX traders differ vastly to those trading fixed income and other asset classes.

Unless voice/eComm solutions are configured to reflect this reality then they will not offer any value to the business.

4. Monitor voice effectively

In many companies, the firms’ employees are well informed of the status of a firm’s surveillance capabilities. Most of whom even not far removed from compliance team and many even sit with them on an advisory capacity.

The trading population are aware that voice is still relatively unmonitored as opposed to written communications such as email and instant messaging. This makes many of them willing to take risks and continue abusive behaviour.

5. Use Machine Learning to detect code switching

Employees also use code switching, also known as language switching, to avoid being detected by compliance teams. The general purpose of code switching is to alternate between languages or pre-meditated coded language to discuss something untoward. Something they would prefer to keep hidden from surveillance and compliance teams.

In the words of our Chief Technology Officer Isaías Sánchez Cortina: “Mainstream technology has advanced tenfold in the recent years, allowing for better deep understanding to enhance surveillance techniques within Capital Markets. Still, the financial trading scenario is challenging complex long complex communications, with noisy content; in a very specific domain with scarce suitable data for automatic learning; and the need of high-quality results and remarkably high throughput. Solving these constrains is key to give customers specific and focused technology that meets their individual needs.”

Communications Surveillance needs love and attention

Effective controls across trading floor communications need time and dedication to get right. The right approach is a personal approach. To meet the surveillance needs of your company. There is no size that fits all of Investment banking, Asset Managers or Hedge Funds. You can also read more about buy side needs in our blog on why the buyside should be investing in voice surveillance.

As well as covering key differences in language and terminology, the right solution must help comply with MIFID II transaction reporting, MAR and Dodd Frank regulations among others.

For over a decade, Fonetic has been delivering voice technology solutions designed specifically for Capital Markets. Our product uses AI technology and machine learning algorithms where they truly add value to accuracy and proactive surveillance techniques.

Our surveillance tool, Trade Comms Suite has been developed by our own in-house market experts, speech scientists and computational linguists to meet this challenge.

For more information about how our proven voice surveillance solution can help Compliance and the First Line with effective control, you can download our whitepaper here.

Download Whitepaper- Voice Surveillance: The Proven Approach for efficient Control