American Express: Machine learning for customer churn prediction and more effective customer retention
The financial services industry is especially challenged in customer retention. American Express has used machine learning to predict churn for its own customers, and have transformed that capability as a product for its merchants.
Importance of customer churn prediction in the financial services industry
In a business environment of increasing competition and digital disruption, customer retention is becoming increasingly important for services companies. However, there is a gap in addressing the retention challenge, as 85% of customers report that companies could do more to retain them. [i]
Effectively predicting and minimizing churn can lead to significant cost savings. However, current customer retention processes tend to be more reactive than preventative.
Customer retention presents significant challenges in the financial services industry. Companies are facing competition within existing established players, and also competition from the booming Fin Tech sector. With increased options, customers are also getting more savvy. In essence, the cost to acquire customers have significantly increased, along with the risk of losing customers.
American Express, as a financial services provider and a network provider, is challenged in multiple dimensions of customer retention. Amex’s customers are both card holders and merchants.[ii] On the consumer end, the credit card industry is notorious for churn[iii], and JPMorgan Chase’s launch of Chase Sapphire Reserve further pressures Amex. On the other end, merchants are increasingly seeking providers’ help on identifying customer purchasing behaviors. For Amex, effectively managing churn is not only crucial to its consumer business, but can also be a key competitive advantage for the network business.
Amex’s approach to address customer churn through predictive analytics
Amex has already started preparing itself to use machine learning algorithms for various parts of its business. In 2010, AMEX upgraded from traditional database technology to a Hadoop infrastructure to work better with machine learning algorithms.[iv] They implemented data platforms and NoSQL databases to enable large-scale machine learning applications.[v] Amex employed machine learning techniques for a wide range of use cases, most notably in fraud detection.
Amex has also gained traction in the customer churn prediction use case. Through its vast amount of historical transactions, Amex has created a machine learning model to forecast potential churn. It uses 115 variables that define customer behaviors, and it believes that it can identify 24% of Australian accounts that will close within the next four months.[vi]
In the merchants business, American Express launched Amex Advance in November, 2017. American Express Advance is a predictive analytics platform for business clients, which aims to provide customized services for merchants to understand their customers’ behaviors.[vii]
Overall, Amex is continuing to harness the predictive analytics power for its own businesses, and transforming the predictive capabilities into products for its merchants.
Recommendations for next steps
As Amex continues the journey of predictive analytics, data availability and quality are crucial factors. Amex needs to focus on both the quantity and quality of collection of payments data. To build a full picture of customer purchasing behavior, it needs to venture into new areas. Its corporate investments arm has already invested in digital payments and digital concierge startups to further build data around customer behavior. Amex needs to integrate these different pieces of data with its existing database. It can also try to collect richer data around customer purchases, by asking consumers to volunteer more information about the purchase. For example, Google and Facebook often ask questions about a recent behavior to gather more data.
In addition, human judgment is required to truly understand the causation of risk elements and address the root causes. Amex needs to hire high caliber data scientists who will work cross-functionally with the business side to make sense of the “at-risk” rankings for customer churn. At the same time, research has shown that there is a difference between targeting “high-risk” customers and customers who are likely to react well to retention campaigns. [viii] With this distinction, the company can effectively segment high risk customers based on (1) the root cause of churn risk, and (2) the probability of reacting well to retention campaigns.
As other financial services firms also invest in machine learning and advanced analytics, how can American Express differentiate themselves? Should American Express continue to build its predictive analytics in house, or should it leverage third party solutions that are specialized in the area? With increased scrutiny on data privacy, and EU’s GDPR coming into effect, what are the limitations American Express can face in terms of building the full picture of the customer behavioral patterns?
[i] Ascarza, Eva. “In Pursuit of Enhanced Customer Retention Management: Review, Key Issues, and Future Directions” https://www.hbs.edu/faculty/Publication%20Files/ascarza_et_al_cns_17_e08d63cf-0b65-4526-9d23-b0b09dcee9b9_538a6ea6-a480-4841-b9f0-a87be24989ba.pdf
[ii] Forbes. “Inside American Express’ Big Data Journey” https://www.forbes.com/sites/ciocentral/2016/04/27/inside-american-express-big-data-journey/#443621e73d89
[iii] Gerdeman, Dina. “A Smarter Way to Reduce Customer Defections”. https://hbswk.hbs.edu/item/a-smarter-way-to-reduce-customer-defections
[iv] Bernard Marr & Co. “American Express: How Big Data And Machine Learning Benefits Consumers And Merchants” https://www.bernardmarr.com/default.asp?contentID=1263
[v] RT Insights. https://www.rtinsights.com/american-express-recommendation-engine/
[vi] HBS Digital Initiative. “American Express : Using data analytics to redefine traditional banking” https://d3.harvard.edu/platform-digit/submission/american-express-using-data-analytics-to-redefine-traditional-banking/
[vii] MarTechSeries. “American Express Launches Amex Advance Personalization Services” https://martechseries.com/predictive-ai/ai-platforms-machine-learning/american-express-launches-amex-advance-personalization-services/
[viii] Ascarza, Eva. “Retention Futility: Targeting High-Risk Customers Might Be Ineffective” https://www.hbs.edu/faculty/Publication%20Files/ascarza_jmr_18_783d54d4-e548-41ed-b1d7-8a180f1ae85a.pdf
Student comments on American Express: Machine learning for customer churn prediction and more effective customer retention
Interesting article! While there is a lot of discussion about customer retention and forward-focused analysis, I wonder how Amex is also thinking about using available data to better target future customers. If Amex can attract higher quality customers in the future (as measured by length of time the customer remains a cardholder), they will likely need to expend less resources on retention efforts down the road. It would also be interesting to better understand which card(s) Amex customers tend to switch to, such that Amex can gain a clearer sense of where their value proposition falls short for various demographics.
Your commentary on predicting customer churn in the financial services industry is very thoughtful and intriguing. In addition to predicting customer atrophy from financial services providers such as American Express, it seems that similar technology might also be applicable to employee retention. This could serve your question about differentiating Amex through improving its internal human capital (both in disruptive technologies, such as the data scientists referenced above, and other functions like customer service, product development, etc.)
Regarding data and privacy, it seems like the effect of regulations like the EU’s GDPR might be ameliorated by the fact that Amex would be assessing its existing customers, versus prospecting for new customers.
Overall, as more competition arises in the credit card industry AMEX must take steps to be proactive in maintaining their customer base. In terms of differentiation, many banks(such as JP Morgan Chase) have additional levers to maintain customer loyalty. In considering options for moving forward, I think AMEX must be extremely cautious in looking to ask consumers to reveal more data. Many lawsuits have been filed against corporations claiming breach of privacy and government trends indicate that the political atmosphere will lead to further consumer protections. One feature of its product offering that AMEX can potentially leverage is its points system. Customers who are more price sensitive or points savvy, may be willing to complete surveys or opt in to share more specific purchase data if it results in gaining more points that can be used towards travel and other activities. This would ultimately help reduce the risk of claims against breach of privacy and reputational risk if executed properly.
Using machine learning to predict churn is an interesting proposition across many industries beyond financial services. I’d agree that with vasts amount of data, you can find the typical overall predictors of churn with a high degree of accuracy (especially in a transnational type environment). A concern I have is that different customers or businesses (in a B2B context) churn for very different reasons and typically will be hard to capture in the data – it would require data inputs by AmEx or other vendors that are highly manual (because you don’t know if a customer churned because of budget issues, consolidation, etc., which likely won’t be included in the data). So while you may be able to predict churn overall, it is critical to understand the reason for churn to inform retention campaigns. This is where i believe machine learning may fall short given inadequate data capture on the front end (you need those churned customers to provide additional information).
Awesome article – thanks for sharing! In answering your first question, I think machine learning will be important to stay competitive in the payments space. Churn is a key indicator of success in this industry, and as such, AmEx should continue to build out their capabilities to master this in house.
Additionally, your post made me think more deeply about how AmEx can use machine learning to also: (1) produce new products to existing customers (i.e., assist in new product development) and (2) predict how profitable new merchants and customers will be when signing them up for new cards.