Banking’s Next Best Action: AI-Based Customer Focus

How do banks leverage the volume and depth of customer data available to truly deliver personalized experiences at scale? Customer relations cannot be improved with big data alone. Instead, convert data into AI-driven insights and find the next best actions for them. Yet, what should be considered when implementing the «next best action» approach?

Since the advent of the digital revolution, the banking industry has shown considerable improvement in understanding the needs of customers and adapting to them. The creation of data and the ability to extract insights from the data have proven to be powerful tools in this regard.

With increasing digitalization and the resurgence of artificial intelligence (AI), banks can strengthen their customer relationships by implementing a next best action (NBA) approach. By leveraging the availability of cheap computing power and sophisticated algorithms, NBA can empower banks to perform such complex tasks as risk mitigation, prevention of churn and fraud, service provisioning and optimization of interaction time. As banks adopt the NBA approach, data and the ability to generate insights from it become an indispensable skill which can help them very efficiently to:

• Personalize product offerings: Predictive analytics can anticipate a customer’s needs and make offers that are suitable and time relevant.
• Engage in meaningful conversation: Contextual intelligence and awareness can make routine client interactions more meaningful, effective and personal.
• Increase loyalty and revenue: More impactful conversations can increase the likelihood of retaining customers and doing more business with them.
• Increase operational efficiencies: Relevant information identification helps narrow your search criteria to the most relevant attributes and reduces noise. This in turn improves operational efficiency.

NBA also promotes a data-driven culture which leads to a more analytical mindset within the organization.

Data is the key

Data forms the cornerstone of any quantitative analysis including NBA. Data can be broadly classified into two categories: structured and unstructured. Structured data is well organizedand consist of a pre-defined model that makes it an ideal candidate for applying AI techniques such as machine learning or natural language processing (NLP). Banks usually do a commendable job in maintaining and handling their structured data and it is usually their de facto data source when applying analytical techniques. Unstructured data is basi- cally the compliment of structured data. It is data stored in different forms like text files, e-mails, reports, videos and images instead of a pre-defined data model. Unstructured data offer banks the ability to look beyond their conventional data source and find that extra edge. The problem of managing structured big data is well documented and practiced. I will therefore focus on managing unstructured big data.

Concerns in managing unstructured big data

Managing unstructured data comes with its own share of perils. Unstructured data can be managed in smaller quantities. But when the quantities increase to a size where management and manipulation present logistical challenges, it becomes a problem.

• Volume: Many firms may not be able to keep up with rapidly increasing volumes of unstructured data. This presents a challenge in not only storing this data, but also securing it.
• Relevance: Just because they are in an unstructured format, it does not mean that the data is relevant to the problem at hand. The inability to gauge relevance increases complexity when managing large data volumes.
• Usability: The lack of well-defined processes makes it extremely difficult to extract value from unstructured data. This includes ways to organize, store, locate and extract unstructured data.

Addressing the concerns

A mix of business and technical acumen is required to address the concerns posed by unstructured data.

• Storage solutions: Technologies like compression, deduplication and tiering should be adopted to reduce space and lower costs associated with handling large volumes of data.
• Timely insights generation: It is imperative to generate timely insights to solve business problems. Speed and accuracy are a top priority when it comes to data usability. To achieve this, organizations can explore newer breeds of ETL (extract, transform, load process) and other analytical tools that take significantly less time to analyze and present information. Some of the tools being used are «IBM SPSS Miner», «Teradata» and «IBM Watson Analytics».
• Supportive technology stack: The technology stack should complement the tools selected. Some of the things to keep in mind when choosing a technology stack are: installation, maintenance, scalability, avail- ability and durability.
• Well-defined business objectives: A good approach to solve the relevance issue is to understand the business objectives and the type of data required to achieve them beforehand. For example, analyzing e-mails for compliance is a different goal than analyzing network traffic for measuring network performance.
• Data consistency: Since unstructured data may come from disparate channels and could be of different types, it is important that data records agree with each other. It is a complex process that requires a combination of policy changes and technology.
• Data security and compliance: With more data comes more responsibility. Security is a big concern since some of the data stored may contain sensitive information. Data handling should comply with industry standards specific to that data type.
• Converting unstructured data to structured: If unstructured data cannot be converted to structured data, it is redundant. Automation systems are extensively used to classify unstructured data, to add metadata to them and to remove redundancies. This will eventually transform unstructured data into structured data and make them usable.

Next Best Action (NBA) Approach

I have looked at concerns and solutions when handling large data volumes. In this section, I'll examine where all the parts fit in the overall strategy and describe the NBA approach.

• Build a holistic customer profile: Use all the customer information available to build an integrated view of the customer. This would not only offer a better view of the customer, but also be an input for the next best offer algorithm. Things that can be included are: account balance and transactional history; logs (e.g. calls/ e-mails) of interaction with bank officials; demographic information; a list of all products and services offered; as well as current and previous products and services. Since building such a profile may involve sensitive customer information, utmost precaution must be taken in managing this data.
• Building the algorithm: The algorithm can be built according to various degrees of complexity. It could span from a few business rules to complicated multi-layered predictive models. The decision to choose an algorithm depends on the following (non-exhaustive) factors:

1. The amount and quality of data available
2. The skill sets available
3. The ability to deploy the models at scale
4. The operational & maintenance support

• Execution: If the bank cannot execute or deliver their offer to customers, even the most pertinent data or accurate model cannot do very much. To properly execute, the banks must think about the technical, operational and organizational implications. Banks should strive for a coherent experience in tandem with all concerned (internal/external) departments for the customers.Example: A customer gets a product offer via e-mail and contacts the call center to find out more. The call center should have access to the customer’s profile to answer the customer’s queries and, where applicable, cross sell.
• Evaluation: Periodic evaluation of the algorithm and overall process should be undertaken. This not only helps to improve the overall process but to better understand the customer. These factors should be kept in mind when evaluating:

1. The number or type of offers made and their acceptance rate
2. Different customer segments and mediums
3. Impact of cross-selling on long term customers

In summary, a next best action approach based on natural language processing and machine learning techniques can open new opportunities and improve relations with existing customers. However, care must be taken in executing this strategy for two reasons. First, it is a perpetual process that requires continuous improvements and updates. Second, as sensitive data is typically involved, appropriate data man- agement policies and standards should be incorporated.