Retail banking is often an early adopter of Information Technologies. When it comes to data science, and in particular machine learning, the use cases involving banking that come to mind first usually focus on saving operating costs, fighting fraud, and raising forecasting accuracy. E-banking is often perceived as something disconnected from data science, as it is the introduction of internet banking that gets the attention. Moreover, when the connection of e-banking and data science is made, it's often in terms of Know Your Customer and Anti-Money Laundering regulatory compliance.
Customer satisfaction analytics might be the unsung hero of the injection of data analytics into e-banking. Data science and customer analysis should come as no surprise overall given the amount of attention paid to retail merchandise sales and customer complaints about obtrusive data collection. For banks, though, data collection is a requirement, and the dynamic is different. There is a lot of academic interest in determining banking customer satisfaction, too, even if it escapes the public's eye. And given the volume and variety of banking customer data, researching the topic seems to be a matter of finding a bank willing to turn over sanitized data sets.
Satisfaction in Big Data
The volume and variety of data in a bank of any size can be tremendous. It's also a regulatory headache, because of privacy issues on the one hand, and KYC requirements on the other. Being able to manipulate useful data while protecting sensitive information is best done by utilizing big data techniques. The technology is mature enough after almost 20 years. Researchers generally understand satisfaction analytics well enough so that a bank should be able to adapt well-known principles to its own case. But does it work that way in the real world?
Big data analytics can be used to glean insight from the data already held in data lakes. So does customer data analysis performed by means of machine learning. The main difference comes from the fact that big data analytics requires more human intervention to derive answers. Indeed, the whole idea of machine learning is to gain predictions without direct intervention.
The banking sector has already put a lot of focus on using machine learning to improve customer satisfaction. In a 2021 report, Capgemini showed that 15.1% of bankers polled reported that their institutions had already seen customer satisfaction improvements from ML in the 2-3 years prior. A slightly smaller percentage expected to see similar results through 2024. Moreover, improving customer satisfaction was the most commonly reported benefit of the use of machine learning among the executives polled.
Customer data analysis in e-banking will remain an important component of bankers' feedback and forecasting mechanisms for the foreseeable future. Online banking is expected to grow worldwide as well, which raises the importance of good customer satisfaction analytics even further. IMARC sees the global online banking market rising from $3.84 billion in 2021 to $5.18 billion in 2027. The research company does note that the volatility brought on by the COVID pandemic requires greater attention to changing conditions, but that the public health crisis had also been a catalyst for growth in internet banking.
How to find satisfaction patterns in e-banking
Customer service data analysis regarding e-banking customer satisfaction has attracted academic as well as IT-industry and banker attention. Studies from locales as diverse as Ethiopia and Qatar can be found in academic journals. The methods employed can be as simple as a questionnaire on customer perception towards internet banking. The results of such a simple data collection and analysis may provide insight such as the fact that within a study, one of the items is not related with e-banking. These studies can be useful, but a bank's internal data science customer might be expecting something else.
Being able to run customer data analytics by means of machine learning requires a lot of setup - it won't work (or be legal) just to give a data engineer carte blanche access to your bank's data. Instead, data needs to be saved and made accessible to multiple users, often simultaneously, as we described in our article on data lakes.
Preparing banking data for analysis
One does not simply walk into a data lake and swim to new shores of masterful banking business insight. There is a lot of possibly tedious work in the setup, as data is indexed and metatagged. This indexing is vital, however, and metatagging is what makes the data findable by users. Without proper metags, the data lake quickly turns into what's called a data swamp, and data is difficult to find.Consumer data analytics becomes essentially impossible as important data gets lost and accuracy suffers. Furthermore, the machine learning algorithms cannot adequately learn without drawing upon the proper data set. This requirement extends past the setup phase as new data coming later needs to be consistently metatagged in the same manner.
Preparing data for machine learning, even for an internal client within a corporate setting, requires anonymization. In this case, merely putting the data into categories isn't enough. Here, being able to separate relevant business information from data that is protected by regulation is a vital part of the setup. Machine learning can come in handy in this case as well, if the scope of the data supplied is not static.
It is possible that anonymization could remove valuable information. This could happen to the point that customer data analysis becomes impossible or gives a poor answer. Striking a balance between data privacy and business need when both regulation and collected data change, especially at the volume of data at hand for a mid-sized bank, may require data science techniques to make use of them.
"Taking a financial institution's data and transforming it into something usable for machine learning requires a thorough understanding of the demands of regulation as well as the possibilities that ML provides. Then, being able to bring it all to life is a challenge of its own. In a business like e-banking, though, the banks that go with a machine learning solution will gain insight that will be hard to replicate in any other way. Those banks, then will gain not only in data security and KYC rules, but in revenue generation and customer retention."
— Vlad Medvedovsky, CEO at Proxet (ex - Rails Reactor) - a custom software development company.
Characteristics of satisfaction in e-banking
Customer satisfaction in e-banking is affected by several factors. Some of them, such as bank brand reputation and value, are external to the customer's own experience. Others, such as problem handling, are elements of the customer's own experience. Components of satisfaction in e-banking can range from:
- Brand Perception,
- Contact Mechanisms
- Cost Effectiveness,
- Ease of Use,
- Efficiency in fulfillment,
- Perceived Value,
- Problem Solving,
- Security, and
- System uptime.
This list is not exhaustive, but it gives a good idea as to the scope of the factors involved. Some can be broken down further. For example, Ease of Use can include website user interface as well as the burden placed on the user by security measures. Collecting this data may involve directly asking, as in the questionnaire mentioned above. However, big data and machine learning methods lend themselves to utilizing a variety of inputs to measure the factors.
For example, a survey that uncovers customer sentiment regarding the bank's e-banking website can be compared to bounce rates on specific web pages to gain a better understanding of where improvements can be made on the user interface. Some studies conducted with banks in India have looked at the ranges of e-banking services utilized by satisfied and dissatisfied clients.
As seen in the graphic below, a variety of factors can be weighed against each other and alone in this manner.
ML model to predict customer satisfaction
Once customer satisfaction analytics have determined the salient features of the sentiment toward the bank's e-banking profile, the next step would be to begin working on improvements. Being able to predict how customers would react to changes in the e-banking offering is one of the main reasons banks have embraced machine learning so heavily.
This predictive aspect of ML was put under the microscope in an article in the International Journal of Advanced Computer Science and Applications (IJACSA). Ala Aldeen Al-Janabi examined the use of prediction modeling to predict e-banking effectiveness. Al-Janabi utilized regression and neural network models in his study. He was able to eventually reach a 78% accuracy rate when predicting client satisfaction. Along with reaching this rate, Al-Janabi examined the relative rankings of a set of attributes that included six attributes, with two, interactivity and reliability, having the weakest correlation. He found that security, assurance, responsiveness and usefulness were the attributes that had the greatest effect on e-banking client sentiment.
In order to reach a 78% accuracy rate, Al-Janabi utilized a dataset of 250 records, with 200 used for training. The other 50 were used to check the accuracy of the system. He claimed that "a new hybrid approach using [an] artificial neural network was proposed for determining the relationships between research variables.
How to use predicting satisfaction
The competitive advantage gained by a well-set-up machine learning platform for measuring and predicting customer satisfaction is an important part of any financial institution's e-banking implementation. Being able to see what clients hold to be important is an insight that makes worthwhile the setup and maintenance of the data lake or other repository.
Machine learning platforms give bankers the possibility to model the changes they are considering. Implementing a security measure can backfire if not handled properly. Bottlenecks in the e-banking platform that might be detected by other means can be spotted more easily.
Furthermore, new banking offerings can be tested in a virtual environment instead of being implemented in a hit-or-miss fashion. Rather than bringing together customers for focus group sessions in which a vision is shaped and human issues introduce error, machine learning based platforms can give a rather accurate vision. The cost of bringing a focus group together versus running a computer platform is another factor that leans the scale in favor of ML. Furthermore, modeling possible offerings can take days instead of months.
Reducing human error is a commonly-cited factor for adding IT to any problem. In the case of analyzing and predicting customer preferences and outcomes, human error is magnified. Bringing in machine learning will help reduce the error load and can save millions of dollars. More importantly, by bringing together disparate data can help possibly avoid regulatory missteps that can cost the bank even more while destroying its reputation.
Predicting client satisfaction through machine learning also lets bankers and their IT departments experiment more widely with little fear of creating a monster. ML based sentiment prediction will thus induce greater innovation. Doing so will unlock greater profits while raising the bank's rating. In an environment that is incredibly competitive, being able to deliver innovative products is an important part of a bank's image.
Machine learning is a powerful tool for discerning the salient features of customer satisfaction with a bank's e-banking offering. Given the complexity of the task, only a self-adaptive solution can provide a clear picture over time. Bankers have been engaged in using artificial intelligence solutions for years already, and this trend is expected to continue at similar levels for the foreseeable future. Implementing a machine learning based system for customer satisfaction analytics is intricate, however. Bringing in a technology partner with an understanding of e-banking from both the financial institution and ML sides is required. Experience in implementing such solutions should not be ignored. As competition increases beyond its already fevered pitch, improving customer service and increasing innovation in e-banking, while lowering costs, the banks that utilize such partnerships will be those who advance.
Accurate parsing enables Q&A quality — but is it possible? No matter the industry or sector, businesses regularly deal with the question of how to efficiently process large amounts of info-heavy documents. Organization leaders, including CTOs, CDOs, and CPOs, are often looking for solutions to this question.
Build a modern data stack by following best practices from data engineering experts. Learn about data maturity, data stack components, and how to build.