Can machine learning improve client outcomes?

  • Describe how machine learning can be used by financial services
  • Identify some of the challenges of machine learning in the sector
  • Explain how to think about machine learning in financial services

Delivering appropriate advice

As a society, I believe we have a responsibility to enable appropriate financial advice to be delivered to a wider group of people. Due to the costs involved – because it is a regulated and complex space – in order to do that effectively, the underlying advice model must be able to be scaled.

Article continues after advert

To do that, you have two options.

You can either automate the advice process, giving people questions to answer and, via algorithms, enable them to self-serve. Or, the other route is by helping advisers become more efficient by automating many of their processes and tasks, freeing up time so they can deliver advice to more people at a lower price point.

Both of these things are happening currently, to varying degrees of maturity and success through ‘if this, then that’ flowchart style approaches. 

However, if one were to apply ML, an analysis of historical outcomes for a group of investors could be used to train a ML algorithm. It could be taught to devise investment strategies for another group with comparable characteristics. 

While their expected results are not guaranteed, it might lead to a fairly decent core model that might work for certain similar types of investors. Such a model might be used to benefit people in future, who would otherwise be excluded due to costs.

Critics could rightly argue over the bluntness of such a tool, but in my view solutions of this nature should never be pitched as a direct replacement for bespoke personalised advice. These techniques, in common with robo-advice, in general represent an alternative approach to support a wider group of people making financial decisions. 

We would also nearly always advocate the option to introduce a degree of human intervention at some point in the process, such as in traditional robo-advice, when the ‘if this, then that’ protocol hits a proverbial wall.

Where I see a prime opportunity within the financial planning process would be around tax optimisation. For example, solving a specific technical challenge in a quicker and more efficient way using ML. But I see this very much working as one tool among many rather than a complete revolutionary overhaul. 

Avoid one-size-fits-all approach

Players across financial services are constantly seeking inspiration from other business sectors in terms of how they conduct their operations. This might be through advertising or branding, or aspects of customer experience through to their digital offerings. 

Yet one warning I am always keen to issue, especially when it comes to how to digitise, is akin to the concept of cargo cult programming, which refers to a style of design characterised by the blind adoption of patterns or technologies that are appropriate in one context but not perhaps in ours.