Customer complaint data reflect the relationship between financial providers and their customers. Regulators, such as the Financial Conduct Authority, rely on published complaint data as a metric of firm performance. Additionally, industry transformation with increasing competition from Fintech and large technology players as well as growing regulatory scrutiny of the sector means that dispute prevention becomes paramount for financial institutions to maintain their competitiveness.
This post considers how financial institutions can harness Natural Language Processing (NLP) technologies to prevent and mitigate disputes with their customers.
Unlocking the value of enterprise data
‘Data' is categorised as structured or unstructured. Structured data, including transactional and financial data, are generally ‘easily decipherable’ by AI algorithms and integrated into analytics. However, structured data is limited by its inflexibility, and can only be applied to specified purposes.
Conversely, unstructured data accounts for 80% of all enterprise data. Typically, unstructured data, such as call-transcripts and emails, have long been uncultivated due to data processing limitations.
The advancement of NLP, which enables machines to interpret natural language in spoken and written form, offers the opportunity to cultivate unstructured data, with more accurate conversions between speech and text. Integrating unstructured data into data analytics can position institutions to anticipate business risks with greater precision and efficiency.
Understanding Customer Feedback
Customer satisfaction hinges on dispute prevention. The handling of customer complaints is often the first point of contact, which be a double-edged sword. A positive experience can resolve negative feedback. A less-optimal experience, however, may escalate into litigation.
NLP can be applied to unstructured data (such as feedback forms and social media posts) to identify the root causes of customer dissatisfaction.
As a test case, the Bank of Italy has integrated sentiment analysis to understand customer feedback on Twitter, by feeding tweets about five European banks to predict customer preferences. If a tweet recalls poor service experiences, the customer service team could approach the customer proactively upon notice by the algorithm. Cumulatively, semantic analysis can pinpoint weaknesses, including skill gaps among service agents, and identify rectifications for future training purposes.
If a customer complaint escalates into a potential dispute, effective deployment of NLP in a business' risk management system can help to ascertain the urgency of such a complaint and promptly direct the complaint to suitable personnel.
The FOS reported that 40% of the 18,000 complaints received annually involve payment fraud. As illustrated by a case study, existing review processes rely heavily on manual keyword filtering by a team of hundreds of employees. Manual distinctions are, however costly and often difficult. For instance, consider two types of complaints: 1) ‘My bank is opening up credit card applications I didn’t ask for’ and 2) ‘my credit card company keeps calling my house, but I haven’t missed a payment’.
Both mention the terms ‘bank’ and ‘credit card’. However, the first complaint pertains to the bank's service, while the latter concerns the disturbance from a debt collector. fter processing the transcribed transcripts to remove typos and jargons the use of NLP in pattern matching reduced the need for manual review by 80%, helping the bank to identify the exact source of the fraud.
The early resolution of customer complaints offer opportunities to strengthen customer relationships by turning bad experiences into positive ones. The adage ‘Customer is King’ is becoming ever more prevalent. The application of NLP to unstructured data has significant potential to overcome cost and time intensive manual review of customer feedback and complaints, allowing for the prompt prevention of escalating disputes with customers.