Recommendation solutions are sometimes used by wealth management firms and banks to automatically suggest the product that a client may be interested in. They aim to increase client satisfaction and avoid wasting both the client’s and the relationship manager’s time with inappropriate offers.
The use of these typically rule-based solutions have however had limited success. Now AI and machine learning are promising to revolutionize the domain with deeper insight being gained from data to better match products with client affinity.
However, despite the appeal, some wealth managers have been reluctant to adopt such approaches. There may be reasons for this that are specific to the firm’s strategy, but one that often gets mentioned is the amount of data that is required – sometimes referred to as the ‘law of small numbers’. Here Avaloq is in a good position to help.
Before delving into this question though, let us consider the type of mathematics and data mining that a machine learning tool is trying to do when recommending an investment.
A typical scenario
Let us imagine that you are a client advisor or private banker at a well-known wealth manager. In 15 minutes, you have a meeting with a client. You have to make a decision fast.
Your investment team has given you a list of funds that are recommended for your clients. To keep it very simple, let’s say the recommendations are to invest in either US, Swiss or UK equities. However, your client normally invests in Asia, occasionally in Europe, including Switzerland, but never in the US. On the other hand, your expertise as an advisor is in Swiss equities.
What would you do?
Given that all we have is the above information, a simple answer is to sell Swiss equities. The problem is that real-life is not so simple - and time is not on our side. The recommendation list itself may be very long, the client’s preferences are complex and changeable, and the success of a relationship manager is dependent on many factors. In addition, there are other restrictions and administrative processes that will have to take place in parallel, such as compliance checks and adherence to the investor’s risk profile and strategic objectives.
A complex problem, but not unsurmountable with the right technology and access to data. Machine learning based recommenders can look for correlations between multiple factors related to the assets list, the relationship managers themselves and the clients. They then identify the most likely match between the three data sets. Such algorithms can observe connections and patterns that humans may not be able to perceive and adapt over time as preferences and circumstances change.
Do you have enough data?
This of course assumes that there is plenty of data to analyze. There are a reportedly 47 million millionaires in the world today and nearly 500 million people who are considered the ‘mass affluent’ (Credit Suisse, 2019). So, on the surface, there is plenty of scope for data mining in wealth management data.
However, while there are millions of millionaires in the world, the number of clients at an average wealth manager may not be enough to identify statistically significant patterns at a granular enough level. At school, we have learnt that a poll of 1000 randomly selected people might be enough to confirm, or at least give us confidence, in a specific observation or hypothesis. But split down by geographic location, tier, age, sex and other preferences or demographics, and this becomes far from convincing. Imagine this data in a multi-dimensional matrix: how many clients and how much client history for each cell in that matrix do you need before you would feel comfortable with each of the deductions.
The answer is in the hundreds of thousands of clients – if not millions. This amount of data with the relevant depth and breadth of transaction data is unlikely to be available to most wealth management firms. Unlike the retail giants, such as Google or Amazon, many are by nature boutiques focused on a relatively small number of clients. Often regionally split, they may cover only a few thousand clients in each area.
Indeed, while some wealth mangers may have a strong competence in one sector, they may have limitations in others. A typical situation is that a firm’s advice has developed around serving a specific segment, for example, a similarly ageing, but wealthy set of business people. The firm may have acquired some of the new generation of clients, for example successful footballers or female entrepreneurs. It is unlikely that data collected inhouse will be enough to predict anything about these market segments on its own.
There are, of course, mathematical techniques to help cope with scarce data: extrapolation models, the fitting of curves, using similar parallel data sets etc. But these do have limitations in such circumstances. They normally involve making a set of assumptions about the nature of the data, such as ignoring non-linearity or that things always remain the same, which are not so convincing in practice.
The only answer that solves this issue for most wealth management firms – who are not the size of UBS, BOA or Morgan Stanley - is to find a way to share data insight, so that the gaps in understanding can be completed. Such sharing is known in data science circles as distributed or federated learning. The trick is to do it in such a way that no sensitive data is shared, but that the benefit of the statistical inference is acceptable to the front-line business unit.
Applying distributed learning
Avaloq’s data science team has worked on such an approach. An approach that not only safeguards client data and the intellectual property of the wealth management firm, but also allows them to reap the benefits of predictive analytics. It can help make decisions about what to recommend, where to target activity and which meetings to plan.
Avaloq Insight does the complicated, but calculatable, part for you:
Saves you time and increases the likelihood that your client is going to be happy with your investment proposal.
Ensures that the proposal makes sense for clients because it matches their preferences with like-minded people who have acquired the same assets.
Takes into account where the wealth manager has expertise and success and so is likely to be convincing.
Understands investment constraints and restrictions and it is aware of the house view.
Mark has over 20 year's of experience working with financial technology across the banking, asset management and capital market sectors. He regular writes content for Avaloq's global community of banks and wealth managers, covering technology trends and innovations, to provide insight and provoke new ideas.
Learn more about:
The strains and prospects of a data science implementation project
How wealth managers create tangible business benefits with data science
3 practical wealth management use cases of data-driven solutions