3 Big Myths About Machine Learning We Need to Bust
2018 has been the year that the financial services industry welcomed machine learning (ML) and artificial intelligence (AI) with open arms. However, there is much hearsay on the topic of machine learning, so what should anyone believe? Let’s start with three big myths, as explained by Dave Webber, director of concept management, data strategy and […]
2018 has been the year that the financial services industry welcomed machine learning (ML) and artificial intelligence (AI) with open arms. However, there is much hearsay on the topic of machine learning, so what should anyone believe? Let’s start with three big myths, as explained by Dave Webber, director of concept management, data strategy and innovation, TransUnion.
While reports last year found that the financial industry was seriously lagging in its adoption – only 37% of organisations said they expected to use AI functionality within the next 18 months – there has been a step-change this year. A study by Adobe and Econsultancy found that the majority (61%) are either already using AI, or plan to adopt the technology within the next 12 months.
Whilst these numbers are positive, it still highlights how over a third of financial services businesses still aren’t even considering using ML or AI. This needs to change.
Both AI and ML have the ability to completely alter and change the way we do business, streamlining processes, reducing costs and improving efficiency. You only need to look at the benefits other sectors, which adopted the technology earlier, are experiencing right now to be convinced. For example, retail businesses are using ML to provide convenient, responsive and personalised services for customers based on their online browsing behaviour, while in healthcare, computer-assisted diagnoses are being used to uncover diseases quicker and more accurately than the human eye.
The benefits of AI and ML are numerous, but in the financial industry, there are still a number of common misconceptions and myths that are slowing down adoption of the technology. It’s time we dispelled them.
1. Machine learning is too risky for financial services
Trust takes on an entirely different and elevated meaning in the finance sector compared to an industry like retail, in which a trial mentality is more acceptable. Fraud protection specialists, for example, may therefore place more trust in traditional systems that they have always used successfully to flag fraudulent applications.
But as fraudsters grow ever more sophisticated with their techniques, a greater array of data and tools are required to detect suspicious or criminal activity, and the demands on the human workforce to spot things will be too high. Through ML, we’re instead able to pull together a vast amount of contextual data to determine whether an application is legitimate.
Over time, automated systems can become more adept at spotting irregular data, resulting in more efficient fraud detection. Believing ML is too much of a risk in financial services is risky in itself. There is simply too much at stake when it comes to detecting and fighting fraud. By letting ML make decisions based on accurate and robust data, you can spend more time and attention on what the parameters of these decisions should be.
2. Machine learning is not ready to make better decisions
Financial institutions are understandably hesitant to adopt unproven technology, but in actual fact, we’ve been using ML and AI-based technology in our daily lives for years. Whether it be Amazon’s product recommendations, or asking Google or Siri for the quickest route home, we already trust ML implicitly – often without even realising.
The benefits of this realisation are also not limited to just fighting fraud; ML can also be a great asset in credit risk and affordability modelling. Through analysing huge amounts of existing data, the technology can help lenders back up their decisions to both consumers and regulators. It can also be used to flag applications where a customer is likely to default, as sample sets can be contrasted with actual decisions of the time to modify a business’s risk profile.
Far from exploring uncharted territory, ML focuses on using data to allow both man and machine to make better decisions together. Here at TransUnion, we conducted our own year-long machine learning trial which highlighted the possible benefits in the credit, fraud and insurance industries. In one case, the level of default in a portfolio of 60,000 credit cards was significantly reduced, whilst overall bad debt dropped by 10%.
3. Machine learning is difficult and time-consuming to implement
Owing to its growing status as the next generation of essential predictive tools, it is justifiable to view ML as an arcane technology that only a select few people can understand. But in reality it is relatively simple to adopt, and is primarily focused on supplementing traditional methods by working alongside existing systems.
As its name may suggest, ML is a largely autonomous process; it develops its algorithms over time as more and more data is used. This results in the quick and easy creation of more bespoke, dynamic and constantly developing models.
Clearly, far from being a time-consuming burden, the way ML can assist and supplement existing jobs will lead to dramatic increases in productivity and efficiency.
2018 has been a landmark year so far for the technology, but for the adoption rate to keep climbing, we need to bust these myths. Society already trusts ML implicitly, and the firms that are able to accept this fact and embrace it won’t just have the potential for substantial increases in productivity and efficiency, but they will ensure that they remain relevant in the future.