Beyond the Hype: How AI Makes a Meaningful Difference for Investors
While our focus as a wealth manager will always be on each client’s individual outcome, we explore different technologies to make your personal experience more valuable. This is why we use services that incorporate Artificial Intelligence and Machine Learning.
Here we look at two examples of how machine learning makes a difference to how your money is managed.
Supervised learning: to evaluate investment suitability
As part of your client account opening journey, we make sure that investments are suitable for you. To ensure you are not overexposed to too much risk, we collect relevant information. This includes your other investments and debts, history of investing, income, expenditure etc.
Over time, the size of this data per client and, therefore, the complexity of evaluating it has grown beyond what advisers can calculate quickly. Twenty years ago, this was an easier problem to solve because life, and our regulatory frameworks, were less complex.
In Netwealth’s case, we record more than 30 parameters when evaluating the suitability of each portfolio. So, how complex does this make each evaluation? Well, let’s simplify it and consider that all the answers are “yes” or “no”. Using traditional, deterministic programming techniques we’d have to write code to deal with 230 (over a billion) different scenarios. This is not practical, especially when we consider that every additional parameter doubles the number of scenarios!
This is where machine learning comes in. Rather than expressly define all possible outcomes for every possible set of inputs, we can train a model to learn how to classify the data itself. For this process we have defined five possible classifications:
- Definitely approved
- Likely to be approved
- Not sure
- Unlikely to be approved
- Definitely not approved
What is known as a Multiclass Decision Forest algorithm allocates new data into one of these classifications based on everything it’s been taught before. Any borderline cases get flagged to our experienced advisers, who may alter the classification. This alteration data is then fed back into the model, which triggers it to change its internal algorithm to match the newly learned information. The result is a process that gets better over time and can accurately classify account opening suitability at scale.
A good analogy is teaching an infant to sort coloured blocks into three buckets labelled “Red”, “Green” and “Blue”. Any blocks that directly match one of the three primary colours are easy. However, a purple block will have them confused between red and blue. They might guess. If the teacher disagrees, they would correct the infant and move the block. If they agreed, the teacher would reinforce that correct decision. The infant then adjusts its internal “algorithms” to match the new data. Given a new coloured block that is similar to the purple one, but not exactly the same, the infant is now very likely to be able to work out the most likely classification.
Unsupervised learning: to ensure best-in-class security
Security is core to everything we do at Netwealth. As you can probably appreciate, protecting some of our most critical resources can be very complex. We cannot possibly envisage every conceivable scenario that might warrant intervention. So, how do we maintain the balance between accessibility and security, without needing dedicated teams to manage it?
This is where what is known as Anomaly Detection algorithms become very useful. These essentially train themselves over time to recognise what is “normal” behaviour. New behaviours are statistically tested to determine whether or not they are unusual. Depending on where the behaviour occurs and the severity of the deviation from ‘normal’ the behaviour may be either reported, mitigated or blocked.
For example, a Netwealth employee logging in to an internal system at an unusual time of night might result in an anomaly report being generated. However, an employee attempting to log in from a different country might be blocked immediately.
Being part of Microsoft Azure’s cloud infrastructure also allows us to take advantage of more global anomaly detection – particularly for DDoS (Distributed Denial of Service) attacks. Always-on traffic monitoring, and real-time mitigation of common network-level attacks, give Netwealth the same defences used by Microsoft’s own online services. The entire scale of Azure’s global network can be used to distribute and mitigate attack traffic across regions.
Again, advanced technology complements what we do, but doesn’t define it.
Finding meaning beyond the hype
These are just a couple of examples where the combined powers of cloud computing, artificial intelligence and machine learning can be put to good use for the benefit of our clients. We recognise there is a great deal of hype around these topics and their capabilities but, used wisely, we believe they provide significant advantages that weren’t available even just a few years ago.
A common worry is the “computer says no” scenario, whereby decisions are entirely at the discretion of computers. Netwealth could be better thought of as a modern car. Yes, we employ thousands of computers to help “keep us on the road”, but there will always be a human at the wheel.
Please remember that when investing your capital is at risk.