Why doesn’t the Netflix model work in investment services?
The success of the business model of companies such as Netflix, Amazon, Spotify also forces even those who deal with investment services to ask a serious question: can we offer the same level of personalization?
According to McKinsey, by 2030 the 80% of wealth managers will offer a service like Netflix’s hyper-personalized, subscription-based model. That seems like a good idea: in fact, Netflix claims that the 75% of what clients watch comes from the very personalized recommendations they receive.
So, the new buzzword in the financial world is “hyper-personalization”. Like Netflix’s style. But talking about it, is one thing, and implementing it, when dealing with investments and other highly complex financial services, is another.
The success of Netflix or Amazon lies in the ability to learn from customer buying behavior through Machine Learning algorithms, in other words the boundless world of Recommendation Systems. Typically, their centerpiece is a Collaborative Filtering or Content-based Filtering algorithm, or a hybrid system.
Both methodologies make automatic predictions about customer’s preferences by exploiting similarities, regarding purchase preferences or interest/services, towards data products shown by the customers’ community (hence the term “collaborative”).
Let’s make an example: the idea is that if customer A has purchased the same mutual fund as customer B – let’s say a short-term government bond fund -, and if customer B has in the past also purchased an equity fund on the tech sector, then it is likely that you can successfully propose the tech fund to customer A as well.
Ok, but where is the value for the wealth manager?
The government bond fund could be a defensive investment for customer B in a broader long-term oriented portfolio with a certain objective and risk acceptance. And perhaps customer A has different objectives, different timeframes, and a totally different risk attitude. Indeed, in terms of compliance and product targeting, the tech fund might be off target for client A and dangerous for the advisor from a regulatory perspective (MIFID or IDD).
Ergo, a system that works quite well with a book or a movie, would have serious problems with investment funds, or other typical wealth management products.
Getting out of the impasse
If you want to create a Recommendation System for wealth managers, the idea of leveraging what happened to similar clients and products remains valid, but it needs to be articulated much better, it needs to go much deeper.
Consistently with the fact that the financial industry is highly regulated, services and products are complex, and people do not fully understand them – indeed the bulk of purchases are marketed by a financial advisor or other similar professional figure whose role is crucial. Therefore, it is necessary to consider the client’s financial DNA, which includes many factors, such as the financial culture, the economic/wealth aspects, the behavioral traits, and the MIFID or IDD regulatory profile, to mention just a few.
In addition, it should be kept in mind that a given product is typically part of a broader product portfolio, which includes various investments, insurance protection and financing too. Therefore, a holistic approach is needed, looking at 360°, considering the customers’ characteristics and the financial products/services associated with them.
The Virtual B solution
In short, more articulated models must be used, that incorporate a greater level of financial know-how. In Virtual B we have solved this problem with a couple of tricks in the modeling strategy:
- embed a core of financial know-how inside the Machine Learning model from the beginning, in a way that the algorithm is routed well in its learning path – this is Bayesian Machine Learning, which allows you to embed “best practice” information in a statistically correct way;
- consider a well-rounded client (and product) profile, including regulatory information from the MIFID or IDD survey, and accurately estimate the client’s financial needs – all of them, whether investment, insurance protection, or financing needs.
It deals, wanting to assign a method label, with a “knowledge-based recommendation system”.
This methodology ensures that the recommendation provided by the system is focused on the individual client’s actual needs and objectives, in a timely manner. It can be used by a financial advisor or insurance agent, typically, but if desired, the recommendation can also be used through a direct channel.
This means fulfilling the MIFID/IDD regulatory paradigm “Know Your Client, Know Your Product”, but most importantly it means implementing the hyper-personalization paradigm.
Do you want to learn more?
If this topic interests you, contact us at the link below and learn how to increase your data’s value.