Investment Management Company
Unlocking up to $42 billion of immediate growth for an investment management company.
For a large investment management company with over $215 billion in assets under management, there exists an opportunity to capture more of
the global market from their existing channels. Our client currently has 10% ($15.5 billion Q4) of the $156.3 billion Q4 global market share.
The company works with sales teams that have historically focused
on selling popular “category-killers” products. This has led to the under-utilization of several products that have excellent performance compared
to the competition.
To take advantage of this growth opportunity to sell a wider range
of products, the company is interested in offering a more diversified basket of products to their current investment banking or financial services customers to increase sales.
The following are the two Key Questions of interest:
- What is the list of products that financial advisors are interested in?
- For any specific product, which specific financial advisor should it be offered to?
To answer these questions Theory+Practice created a Recommendation Model to predict the quantity of each category of funds which will appear in a financial advisor’s next quarter portfolio (next basket) given the sequence of previous purchases (past baskets). This prediction will enable our client
to generate a competitive offering to the financial advisors’ next basket.
We will also be able to generate a list of financial advisors that are likely to be interested in a given product for the investment management company.
Our Recommendation Model leverages the fundamental concepts of;
Next Basket Recommendation (NBR) and Recurrent Neural Networks (RNN).
NBR considers the problem of recommending a set of items into the next basket that users will purchase as a whole, based on the baskets of items that users have purchased. And, RNN takes into account the temporal aspect of financial advisor+product interactions in the recommendation process. This approach will predict not only the funds that are likely
to appear in the financial advisor’s next basket but also predict the quantity of purchase. By combining both NBR and RNN in our design
we were able to answer both of our core questions simultaneously.
To train the Recommendation Model, we mindfully chose to use global market sales, so as to create a more robust and generalizable model.
We did this by utilizing global sales as the input the model learns instead
of our client’s individual sales. The model was trained using a sequence of eighth quarters of global sales and predicts a list of products that are more likely to be purchased by each financial advisor in the future (ninth) quarter.
Our solution enables our client’s regional managers to look into the future instead of looking at the past to decide what products to put in front
of their customers.
Theory+Practice’s Recommendation Model will equip the company’s regional managers to target their most valuable customers with the most appropriate products. Our secondary models, additionally, will enable
our client’s regional managers to determine the most valuable interventions to be applied and identify the moments that would deliver the highest possible impact.
By triangulating a set of assumptions (such as market share, product crowding, and competitive awareness), a set of conditions (such as financial advisor incentivization and market knowledge utilization), and a set of targeted intervention strategies (such as switching and new product suitability discoveries), our predictive ROI service revealed a growth opportunity of $42 billion at the highest and $2.12 billion at the lowest
as the achievable added value, within a single quarter.
Deep segmentation & data modeling leading to better client acquisition & less member drop out.
Using a machine learning and advanced analytics approach, a leading Credit Union with over $2B in assets under management has an opportunity to better understand their members in order to drive revenue and increase retention with better service and increased utilization of products With a single effort and in a short timeline Theory+Practice enabled this opportunity by:
- Identifying unique member segments in relation to business KPIs such as loyalty, engagement, and liquidity risk (amongst others)
- Forecast and explain member behavior based on past actions specifically to predict the risk of members’ dropout and becoming inactive
By understanding members through these two lenses, they now have the potential to increase revenue through targeted marketing efforts to attract new high life-time value members and enable a hyper-personalized approach to deepen the relationship with existing members. This, in turn, can increase membership retention, engagement, and loyalty.
To answer these questions Theory+Practice split the project into three steps:
- Created Minimum Viable Data (MVD) by selecting the most relevant and high quality datasets from over 200 datasets and historical records, followed by the generation of micro-signals capturing relevant behavior of members in terms of loyalty, engagement and risk.
- Developed Unsupervised Deep Segmentation that utilized 19 different datasets to create over 300 distinct signals associated with member relationship, engagement, loyalty, liquidity, financial literacy as well as base demographics. An unsupervised segmentation model was then used to divide members into 6 groups with well-defined characteristics.
- Created a Propensity Model that forecasted the likelihood of members leaving the Credit Union one month and one quarter in advance with 96% & 88% accuracy respectively. Beyond model accuracy, information was provided about the driving behaviors behind the risk of dropping out to enable appropriate interventions.
The deep segmentation model, combined with the risk propensity model, shed light on distinct member groups with different wants, needs, and preferences.
The following are just a few of the patterns that we identified:
- 1.6k new members who are likely to increase their engagement in the next months and more that are likely to become wealth management clients.
- Members that targeted messages for certain products would accelerate their adoption and increase up-sell and cross-sell opportunities.
- Members in each subsegment who are less likely to drop out in the next months and twice as likely to increase their number of active accounts in the next months, as well as members who are twice as likely to drop out than other members in these subgroups.
- Identified a smaller cluster of very loyal members who are also wealth management clients and have a high probability of using a new product in the following months.
We also created new indicators to predict member behavior and identified the most important drivers behind the risk prediction that span the engagement and loyalty of the members as well as the transaction patterns and their account balances. Additionally, these variables capture differences across the population of members as well as the changes in the average of a member’s own behavior.
These observations created an opportunity for intelligent interventions that can increase higher retention and increase the life-time value of customers. A member relationship manager can leverage this information to engage proactively instead of reactively with clients at risk. The manager would be able to identify the appropriate intervention at the time; for example, an offer to open a new account, support for a payment plan, or incentives to increase the utilization of products.
These insights and new indicators provided a clear and direct way for the Credit Union to better serve their members, drive organization KPIs including profit while ultimately improving member satisfaction and loyalty.