However, my view is that there needs to be a narrow band of applicability for those automated processes, so it is important to keep these aspects fairly simple, transparent and easy to understand. These could then be supported and built on by more sophisticated requirements with some human intervention. What I do not believe is that we ought to rush to build very complex robotic advice processes to replicate large portions of the advice journey because we will very quickly end up with things we do not understand.
From a regulatory standpoint, if an adviser is unable to fully explain the inner workings of the algorithms or programs on which their client’s decisions have been made, that presents a problem. Going back to my cat example – if a machine learning program cannot fundamentally tell you why it thinks that picture is of a cat or not of a cat, then an adviser hauled before the regulator is not going to be able to tell them why they did or did not recommend that product. This is a problem, and one that we will likely explore in more detail in a future article.
With regard to the quality of data, we often refer to the concept of ‘garbage in, garbage out’. This has now moved far beyond sheer quality of data input, missing fields, typos and inconsistent fields, for example, and into the sphere of mandated restrictions in the form of data protection.
The EU-driven rules around General Data Protection Regulation, which came into effect in 2018 (they were published in 2016 but came into legal effect in the UK in May 2018), present a particular tension for our sector and the advancement of, specifically, machine learning.
Machine learning is based on data, and the more data you have to help build your algorithm, the better its quality is likely to be. Therefore, the optimum scenario is to be able to collect as much data as you possibly can and hold on to it for as long as you possibly can.
In financial services, we have the additional consideration that the nature of the data being collected is highly personal and potentially sensitive. This warrants tighter regulation around its use, around data privacy and therefore anyone using that data.
Building AI or machine learning models, for example, would have to adhere to particular restrictions and rules, in turn possibly limiting the way in which they are allowed to use the data and therefore also limiting its efficacy.
