Today’s campaign process (Figure 2) is focused on generating insights for the bank’s marketing department and then planning campaigns around these insights. Most of the insights are generated using statistical regression models for descriptive and predictive analytics. There is hardly any use of artificial neural networks (ANNs). Risk of non-acceptance and feedback cycle time are quite high. Marketers have to wait for at least several weeks, if not months, to get feedback on the effectiveness of campaigns.
In future, data will be processed as event flows and streams to be able to generate real time alerts with low latency (Figure 2). Auto-encoders using ANNs can generate alerts from streaming data. Banks stuck in the low trust zone will be busy providing alerts and self-service capabilities, albeit smart and useful. Progressive banks will attempt to climb the pyramid of trust using capabilities such as “Do-It-For-Me (DIFM) transactions” (automated mundane tasks). They will leverage ML to generate proactive insights and offers by analyzing data at rest and in motion. As they build more trust with their customers, they will offer proactive contextual advice. As they consolidate their trusted customer relationships, they could finally end up offering “DIFM advice” automated actions to realize mutually agreed financial goals. AI-ML-powered customer engagement will compel prospects to switch banks and reduce the need for banks to conduct mass campaigns.
Most ML products offer tools for visualizing data and modeling using statistical as well as ANN-based models. Automated feature engineering (ability to predict which of the input variables have a high degree of correlation to output automatically) capabilities are also critical as they reduce the need for manual intervention from expensive and specialist data science resources. ML models are trained and tested offline using batch data. However, once the F1 scores (measure that balances between precision and recall) and accuracy reach acceptable levels, ML models in production can generate insights, advice, and automatic actions customer by customer in real time while the bank is engaged with the customer. Optimizing metrics such as F1 and accuracy is to help minimize business costs and impact (e.g. avoid classifying a customer likely to churn as someone loyal).
Other factors that are critical in ML adoption are the ability to reduce over fitting of input data and the ability to explain predictions. Based on outlier patterns and errors in insights, offers, and advice generated, the models can be retrained manually. Several of the fintech vendors provide such capabilities today. In future, unsupervised learning could produce “self-learning ML models”.
Benefits of ML led digital insights
- Timely generation of insights and advice for customers directly, rather than internal bank consumption and targeting of appropriate customers later
- Improved customer experience (customer satisfaction, net promoter scores) by 20–30%6
- Increase in customer engagement (frequency, time spent) through digital channels by 20–60%6
- Increase in new accounts opened and average account balances by 10–20%6 • Higher customer loyalty as well as associated reduction in customer churn risks
- Reduction in operational costs and redeployment of customer service personnel in more niche, value-added activities to support high value customers
Banks have earned valuable customer trust as safe keepers of money and reliable fulfillers of financial transactions, but they have yet to push their trust bar higher. The majority are still fixated on customer churn and mass acquisition campaigns. It is time they redirect their energies and budgets towards engaging, personalized, contextual, and proactive insights and advice powered by AI and ML technologies. As they scale the pyramid of trust, they will automatically experience a marked surge in prospects wanting to sign up!
¹IDC forecast – Worldwide spending on Cognitive and Artificial Intelligence systems https://bit.ly/2N7w4Pf
²Raising the CX bar – How to Close the Trust Gap in Retail Banking, Celent, Dec 2018 https://bit.ly/2V1VBOK
³Meet NOMI https://bit.ly/2B6zGew
4CommBank using data to drive “Next Best Conversation” strategy https://bit.ly/2XVcUm0
5Introducing The Hub https://bit.ly/2UAvEW6 6Personetics https://personetics.com
- Alenka Grealish, Senior Analyst at Celent’s Research and Advisory Group
- Eran Livneh, VP - Marketing at Personetics