Custom AI Application Customer Propensity Analysis Framework
A conglomerate in the MEA region wanted to find ways to cross-sell credit cards to its customers. The conglomerate, which offers services in the retail, entertainment, financial, and real estate sector, wanted to target customers who displayed high spending capacity and would be inclined towards purchasing credit cards. They wanted to generate data-driven insights on their customer base, to effectively target the right segments.
Traditionally, financial institutions have relied heavily on factors such as contractual term or customer satisfaction to influence the purchase of new products. Although this strategy has proven effective in the past, it has not necessarily translated into the purchase of multiple products from the same company. A recent survey highlights that while 75% of respondents held credit cards, only 33% held credit cards from their primary bank. There are two reasons for this market gap – the primary bank’s credit card offers an inferior value proposition, or they do not sufficiently market their credit cards. To tap into this market opportunity, financial institutions must re-route their strategy from traditional cross-selling techniques and assess customer perceptions and attitudes through granular analyses. They must attempt to identify customers’ motivations and tap into their purchase drivers to effectively promote their products.
Over the past decade, several financial institutions have been considering cross-selling as a viable option, as re-targeting existing customers has a 60%–70% chance of conversion, as opposed to acquiring a new customer, which has a mere 5%–20% chance of conversion. Effective cross-selling relies on a ‘customer-centric’ view instead of a ‘product-centric’ one, where a customer-centric view allows businesses to structure their engagement model on their customer base and personalize service offerings based on buyer personas. Financial institutions now create actionable buyer personas to deepen their relationships with existing customers and retain them for longer periods.
Cross-selling is a long-standing technique employed by marketers to offer additional products to customers during their purchase, encouraging high customer spends. This must not be at the cost of undermining customer trust but should be based on an ethical relationship between two parties. An essential cornerstone of cross-selling is knowing how the existing products in the company’s portfolio complement each other.
The primary benefits of cross-selling for financial institutions include
· Enhancing customer experience
· Acquiring new customers and retaining existing customers
· Creating a positive perception of the business
· Encouraging customers to invest in multiple products and thereby establishing brand loyalty
· Improving customer lifetime value (CLV)
Cross-selling is a practice prevalent across the financial industry. Amidst the proliferation of customer data and data mining processes, dynamic customer profiles can be created, allowing companies to understand clusters that would benefit from cross-selling. With the exponential growth in data, financial institutions can combine internal data sources such as transaction records and availed services with external data from third-party agencies to enrich their buyer personas and analyze customer journeys, spend capacity, drivers of purchase behavior, and so on.
Cross-selling especially benefits conglomerates, where the business’ customer base comprises a wider and different audience than traditionally targeted. Research suggests that cross-selling accounts for almost 20% of the value companies derive from revenue synergies. The more the products complement each other, the greater the potential to create newer product bundles, thereby allowing access to a previously inaccessible customer segment.
In today’s competitive market, it is essential to maximize the value of each customer, and for this, businesses require a well-organized cross-selling solution.
If financial institutions want to truly tap into existing market opportunities and reap profits, they must look beyond basic demographic data and understand the behavioral and attitudinal traits of their customer segments.
Cross-selling in the traditional sense relies little on analytics and more on the assumptions of sales staff. It is driven by experience, which is only reliable to an extent, as these efforts take place in silos. Another common technique employed in cross-selling is the inspection of historical data, i.e., past purchases. Further, financial institutions also promote additional products by offering periodic deals to their customers. While these practices have been beneficial in the past, they do not often translate into best practices in this age, especially for companies looking to scale.
Analytics is a key tool helping financial institutions unlock data-driven insights. As cross-selling has received increasing attention over the last few years, analytics has played a stronger and more integral role in helping businesses succeed in their cross-selling endeavors through timely insights. For instance, if a customer has purchased a ticket for their travel to a foreign country, their transaction history can be leveraged to suggest the purchase of a travel credit card. With efficient data mining and the continuous generation of real-time insights, businesses can contact customers at the right time and possibly convert insights into sales.
Predictive analytics is a highly beneficial analytics technique that aids cross-selling. With data on purchase volumes, frequencies of purchase, behavior, and demographics, it can help businesses make accurate forecasts for product purchases. For example, by closely analyzing a client’s purchase habits, businesses can forecast the spending behavior of the customer for the near future and create personalized offers. With predictive analytics, decision-making is bolstered through actionable insights.
Cross-selling is a delicate art that requires precise, insightful planning driven by data-backed decisions. These decisions must stem from a rigorous modeling approach that projects the potential profit margins for the business while maximizing opportunities. Profitability analysis is a component of enterprise planning (ERP) that allows business leaders to predict the profitability of the proposed product and drills down to details such as customer characteristics and purchase patterns. For instance, if a business introduces a new credit card promotion, a profitability analysis will help assess its ROI for the near future, thereby helping make informed decisions.
A propensity model is a mathematical program that can predict a buyer’s probability or propensity to purchase an item in the future. By narrowing in on a specific group of customers, companies can achieve better ROI and reduce their marketing spend.
Essentially, this model helps companies answer the questions – ‘Who should we market to?’, ‘How should we reach out to the targeted group?’, and ‘What products should we offer them?’.
Propensity models allow for effective customer segmentation. They evaluate the likelihood of a customer purchasing a product, based on past purchases, the timeline of purchase, chronological order of purchase, and combination of products purchased. This data helps analyze customer activities to curate effective cross-selling techniques. Building a propensity model is especially important when you want to run a marketing campaign, with limited resources, for a select subset of customers
A propensity model gives businesses customer insights that can bolster their cross-selling campaigns. With this data, businesses can achieve a greater degree of analytical maturity as information is no longer generated in silos. A few of the essential capabilities of a propensity model include the following:
1. Enabling Behavioral Segmentation
Businesses can segment customers into different profiles and distinguish the high-spending customers from the less active ones. This enables businesses to target marketing communication specific to the customer segment and use insights to improve products and services. Businesses can even create specific cross-selling bundles, to the preference of the identified customer segments.
2. Utilizing Streaming Analytics
A propensity model offers real-time insights to channel marketing efforts. The model utilizes demographic, behavioral, and historical data to predict customers’ propensity to respond to a certain product or promotion. These insights can help businesses create efficient and personalized outreach strategies and improve customer relationships. This also helps increase conversion rates and reduce rejection rates, while optimizing marketing spends.
3. Recommending Customer Offer Affinity
To capture customers in a competitive environment, businesses must continuously engage with their customers. A propensity model recommends the ‘next best offer’ to continuously create a cycle where customers are satisfied. It offers a close view into the best product bundles for a specific customer segment that could increase their overall CLTV. This helps businesses connect the right products to the right customers at the right time.
4. Predicting Customer Churn Rates
The model helps companies identify customers at risk by studying customer inactivity and disengagement over a given period. Within the data set, these aspects are inspected through the frequency of account actions and so on. The model gives insights into customer churn rates at a relationship level and helps companies take necessary steps to retain their customers.
5. Capturing CLTV
The model predicts customers’ future relationships with the business. From past behavioral, demographic, transactional, and marketing data, the propensity model predicts future customer purchase rates and the monetary value to be gained from customers. These insights can enable executives to allocate investments while ensuring that resources are efficiently utilized to retain and satisfy customers.
6. Assessing Marketing Expenditure
With knowledge on customer preferences, businesses can deploy decision trees to the model to obtain rich data on the target group. This helps brands identify the most effective marketing channels and reduce marketing expenses wherever necessary. Such insights can aid businesses in creating marketing campaigns that can effectively hold customers’ interest.
Let us briefly look at the steps required to create an effective customer propensity application:
- Explorative data analysis is conducted on structured and unstructured data. The identified variables and their predictive power in terms of correlation are closely analyzed.
- Before a customer propensity framework is built, the behavioral and attitudinal attributes of the market are examined. The possible profitability is assessed based on the data from different streams of the business. This data includes frequency of purchase, type of purchase, and so on.
- Extensive features selection and the creation of new parameters are undertaken to capture key trends and the behavior of customer segments, in order to measure profitability.
- Feeding this data into a propensity framework allows for a closer view of the demographic. The framework segments customers based on common characteristics such as purchase patterns, engagement, psycho-graphics, purchase behavior, and such. The model closely examines the data at hand to predict the propensity of customers who would purchase the given product.
- Historical data from previous engagements and promotions is fed into the model to gain a filtered and refined customer profile.
- A series of algorithms are developed to arrive at the best-fit model, which generates the behavioral attributes of potential customers and scores the population based on the propensity model to generate the target list of customers for acquisition.
Let’s now dive deep into a solution that was built by TheMathCompany to predict the customer acquisition rate for a credit card cross-selling opportunity for a global conglomerate.
A retail and entertainment conglomerate in the MEA region wanted to cross-sell credit cards to customers. This required uncovering insights on target groups who were profitable and exhibited the highest propensity to buy the client’s credit cards.
TheMathCompany helped the client identify the most profitable customer segment in their retail and entertainment user base by analyzing customer behavioral patterns. These patterns were used as metrics to identify customers who were not only profitable but also exhibited the maximum propensity to buy the client’s credit card.
A three-phase solution framework was built to evaluate and rank customers across retail and entertainment businesses. This systematic evaluation was done to gauge the maximum profitability and then the propensity of individual customers to subscribe to credit card offers.
Phase 1: Profitability model analysis to segregate consumers from a vast overlapping customer base
Our team classified customers into two groups to derive the requisite data. The first group comprised customers who were already subscribed to the client’s credit card and used it to make transactions at the client-affiliated retail and entertainment outlets. The second group comprised customers who had not subscribed to the client’s credit card but only accounted for transactions in the client-affiliated outlets.
The information from the first group, such as monthly expenditure at the hypermarket, types, and quantity of items purchased, was used as training data to build a segmentation model that ranked the profitability of customers in the second group. Findings from this analysis led to the identification of close to 405K customers. This identification was in line with the metrics that were driving the idea of profitability in the data model.
Phase 2: Customer propensity model analysis
The information derived from the first phase of the solution framework identified only customers who did not yet possess credit cards under the client’s banner. It was not enough to guarantee subscription and the subsequent profitable utilization of the credit cards. To be more specific, this hypothesis was not enough to help the client count on the degree of the propensity of customers to buy the credit card.
To further segregate the database, our team imported data from a campaign that was previously launched by our client to promote credit cards. The information of customers who displayed engagement in this campaign was tallied with the profiles that were filtered from the second group with the help of the customer profitability model analysis results.
The engagement quality varied during the course of the credit card campaign. For instance, from a digital marketing perspective, the initial analysis of the click-through rate of this campaign issued an expected 70% customer conversion rate. However, a high degree of bounce rate of viewers (who could be potential customers) from the product landing page or while filling the subscription form contradicted the initial expectation of conversion. This hinted at a few possibilities – customers either exited midway because they were not satisfied with the offer or they were forced to exit because of technical issues. The varying nature of the engagement introduced heterogeneity in the datasets. This necessitated the utilization of the decision tree model to help to segregate customers based on their engagement levels.
Phase 3: Replicating the retail data process to identify target customers powered by BFSI analytics
Customer segregations conducted at different levels in the previously launched campaign helped identify the specific set of customers who displayed the maximum propensity to purchase the client’s credit card. However, the measurement KPIs that determined the profitability and propensity of customers in the retail business could not be applied to the customer base in the entertainment business. The frequency and time of transactions made by people tend to vary across these businesses. This necessitated two separate data projects across these sectors to identify customers. Streamlining the project course, which included manipulation and modeling of data for the retail sector, helped our team automate most processes while conducting the second project in the entertainment business. This allowed us to minimize the project delivery timeline by almost half.
- The solution framework led to the selection of almost 405,000 customers as prospects for credit card acquisition with an estimated accuracy of about 80%.
- Aided the client in designing an informed campaign that befitted the credit card desirability index of customers.
- Collaborated with the marketing and digital teams to support the deployment of campaigns for acquisition.
- The parameters generated from the analysis are reusable and can easily be scaled across other brands.
The challenge for organizations lies in their inability to gain tangible value for proposed campaigns and connect the right customers with the right products at the right time. A customer propensity model can examine demographics while accounting for market dynamics and the behavioral attributes of customers to create a refined potential target segment. Such a custom AI application can generate a holistic view of the population and extract valuable insights from existing datasets. Also, these intelligent and custom AI applications shed light on the value and the potential returns from the campaign. Hence, it is integral to utilize an application that can provide a holistic view of customer drivers in order to create specific customer clusters and channel efforts that effectively target these potential profit pools.