In our last blog, we introduced where ML & AI is being used across the financial sector. In today's release, we will delve more deeply into the retail banking domain. Over the course of this blog we will discuss the opportunities and challenges that exist for ML & AI across the following use cases:
- Fraud Detection
- Vulnerable Customer Identification
- Customer Engagement
- Intelligent Document Ingestion
Covid-19 has disrupted the sector. However, ML & AI can support rapid recovery & resurgence for firms.
Retail banking has a central role in underpinning the world's economic backbone and subsequent ongoing recovery after the global financial crisis of 2008. As well as the more recent disruptions caused to society, as a consequence of the Covid-19 pandemic. With interest rates at record lows and having remained so for the last ~2 years, revenues for institutions offering either retail or business banking services have experienced a sharp decline. McKinsey suggests that this impact could be in the region of 16 to 44% across Western Europe alone.
As a result, firms are exploring ways in which they can reduce their costs to serve customers in order to more effectively manage their balance sheets. This is particularly prevalent in the retail and business banking domains where margins are much finer than those in the investment banking space. Furthermore, retail banking has been operating amidst a seismic digital transformation for the last ~10 years. As an example, up until the first half of 2021, FinTech investments by private equity and venture capitalists were at their highest point ever. With FinTech businesses securing more external investment in the first half of 2021 when compared to the full calendar year of 2020 in the U.K.
Digitisation of channels and customer journeys means ML & AI can be used more effectively.
Whilst on the flip side, many retail and business banking providers have announced a raft of branch closures. With many citing a growing increase in digital systems of engagement with customers which was further accelerated by Covid-19.Subsequently, ML and AI afford attractive opportunities to further accelerate the digitization of the retail banking sector, in order lower its cost to serve customers and increase balance sheet management effectiveness. Let’s expand on these use cases starting firstly with Fraud Detection.
The threat of fraud facing banks, payments firms and their customers has grown dramatically in recent years.
In their 2020 report, UK Finance conservatively forecasted that unauthorised financial fraud losses across payment cards, remote banking and cheques totalled £824.8 million. Fast forward to the first half of 2021 and a revised report indicated that there had been a 30% increase in year on year fraud cases.
Indeed, whilst systems of engagement are evolving across the retail banking estate, so are the scamming techniques of fraudsters. Increasingly deceptive techniques via phone, text message, email and social media are becoming increasingly commonplace. With an increase in the prevalence of criminals placing their dark traits on authorised push payment (APP) fraud. Whereby the customer is tricked into authorising a payment to an account controlled by a criminal.
Subsequently, where banks are deemed to have been held responsible for not having identified the spurious transaction patterns, then customers must ultimately be reimbursed by the bank. However, in the event that the bank has activated its relevant checks and balances, yet the customer still proceeds with the transaction, then the customer is held liable for the transaction and will most likely not get any form of compensation. This ultimately leads to disenfranchisement with the firm's product and an increased likelihood that the customer will take their future banking needs elsewhere.
Ultimately, this is bad news for the Bank and the customer, whichever the outcome!
There are multiple opportunities for firms across the fraud domain. To name but a few:
1) Conventional data collection is usually siloed and FOCUSED operational data. There is massive value in collecting much more detailed and comprehensive payment, web, mobile application and contact centre data, for analytics and subsequent execution of rules and machine learning.
2) Machine learning is at a stage now where vast volumes of streaming and batch type data can be analysed in near real time to detect and interject in scams, payment fraud and money mewling activities. Performing pattern detection on known tactics and behaviours of scammers.
3) Customer interaction tends to be focused on outbound human contact. This works if contact is completed the first time but causes many issues if customers have to call back and wait in queues.
A chatbot based approach where customer interaction is instigated and worked through the review process would significantly improve the detection rates and the customer experience. This could also drive significant cost savings in contact centres.
In the United Kingdom, both the Prudential Regulatory Authority (PRA) & BofE have increased their focus on ensuring banks and the wider financial services sector address big challenges around vulnerable customers. The FCA describes a vulnerable customer as “someone who, due to their personal circumstances, is especially susceptible to harm - particularly when a firm is not acting with appropriate levels of care.”
The examples and scenarios of when a customer could be deemed vulnerable are far-reaching. In short, all customers could become vulnerable. However, there is an increased risk of a customer becoming vulnerable based on a set of common characteristics. The FCA define these as:
- Health – health conditions or illnesses that affect the ability to carry out day-to-day tasks
- Life events – life events such as bereavement, job loss or relationship breakdown
- Resilience – low ability to withstand financial or emotional shocks
- Capability – low knowledge of financial matters or low confidence in managing money
Research conducted by the FCA found that the number of financial services customers that could be classed as 'vulnerable' increased by 15% over the initial months of the Covid-19 pandemic in the UK. More recently, it has been estimated that around 27.7 million individuals could be classified as vulnerable. As such, it is clear to see that this is a big problem which data, analytics, ML & AI could help to stem the tide of.
Customers who are vulnerable or are moving towards this status give indicators whether explicit and implicit. These indicators can be found in;
- Transaction data, where changes move towards core discretionary and non-discretionary spend
- Changes in credit reference data indicating worsening financial and affordability status
- Contact data content and sentiment which could indicate life events, health, education needs and many other indicators not obvious from structured data
To drive these insights, ML & AI services such as natural language processing (NLP), pattern recognition and segmentation can be used. These may drive automatic vulnerable indicators or potential vulnerable indicators for review by customer care teams driving tailored interactions for customers who need more support.
The rise of neobanks like Monzo & Starling in the U.K. and N26 in Germany, has seen an entire shift away from traditional bricks and mortar branches. Whilst established players are seeking to emulate the digital experiences of their Unicorn challengers. Which in turn is having a drastic impact on the number of retail branches and ATM’s that have adorned our high streets for decades. As an example some ~4000 high-street branches have been closed in the U.K alone over the last 6 years.
Whilst the U.K government is considering new legislation that grants the likes of the FCA special dispensation to refuse the ongoing widespread closure of branches and ATM’s. As society becomes more accustomed to handling everyday banking tasks through digital channels, it is very likely that our high streets will continue to see a graceful degradation of retail banking branches and ATM’s over the next decade.
In response to their ongoing efforts to reduce the overheads of running branches and ATM’s, firms are widely deploying the use of chatbots to expedite customer queries for everyday bank tasks. By combining both unstructured, semi-structured and structured data, firms are able to build vast data sets that can train ML & AI driven live chat interactions to support queries such as requesting new debit cards or updating addresses through both mobile and web channels.
In this instance, Natural Language Processing (NLP) is typically used in customer engagement applications, as it allows firms to analyse and extract data from vast amounts of text. As part of this process, ML models can learn from the feedback of previous interactions with customers, by collecting historical data and establishing a classification of customers’ preferences, reactions, patterns and behaviours.
In turn, this enables the firm to enhance its digital channels of engagement and support faster, first time resolutions to customers queries. Ultimately, with the intention of improving customer experience. Indeed, some studies have reported that customers actually prefer to discuss sensitive financial matters like fraud or debt management with conversational ML & AI solutions.
Furthermore, in order to accelerate the customer engagement process, voice biometric solutions are now being commonly deployed to circumvent the use of traditional security questions and validation exercises. Firstly accelerating the process of customers being identified in a timely manner. And secondly, enabling firms to reduce the attack vectors for fraud. As an example, HSBC has enrolled ~2.8m customers to use its Voice ID system and estimated that this added layer of security has prevented ~£250m of UK customers' money from falling into the hands of criminals last year alone.
So the saying goes, if you automate rubbish then you get garbage out on the other end!
With banking institutions all over the world looking to introduce efficiencies across their important business services, automation has played a key role in redefining how firms interact with and service their customers. Robotic process automation (RPA) is one particular domain that is set to play a pivotal role in task execution across the financial services sector over the next few years. This is even more compelling when paired with intelligent document processing capabilities.
Combining robotic automation and artificial intelligence, RPA is the process of automating across applications and systems to perform repetitive tasks that were once performed by humans. It is also referred to as “smart automation” or “intelligent automation” and thus, refers to any software system that can be programmed to perform tasks that previously required the input of human intelligence to be executed successfully.
Given the regulated nature of financial services, there is a need to capture and demonstrate an audit trail of business events and decisions. Whether this is the approval of opening a new current account, rubber-stamping a mortgage application, or performing a series of steps to process a loan request.
These tasks have traditionally been human-intensive, executed by a myriad of teams, with significant hand-offs and reams of paperwork for the customer to complete and return to the firm either in the branch or via post. Evidently, these hand-offs and an over-reliance on manual intervention can mean that documents get lost. Or approvals take long periods of time for processes to be executed end to end. In an increasingly digital by default society, customers are not prepared to wait weeks to have their mortgage or loan application approved. Indeed, they are not prepared to wait weeks to have a new current account approved as well. They want a response in days or even minutes.
Ultimately, by combining RPA and intelligent document processing firms can reimagine business processes. In turn, this can eliminate waste and reduce operational costs across their balance sheet. As an example, research conducted by McKinsey estimates that currently demonstrated technologies can “fully automate” 42% and “mostly automate” a further 19% of finance activities.
To seize this opportunity firms must adopt a strategic, and not tactical, approach to deploying the use of RPA and intelligent document processing. Ultimately, this should be aligned to the firm's desired channels for interacting with customers, the products they sell and the levels of service they wish to deliver. The opportunities in this space are potentially unbounded for firms. With avenues of exploration ranging across the following domains:
1) Automated loan processing
2) Know your customer (KYC) background checks
3) Mortgage document processing
4) Cheque imaging processing
5) Account opening
6) Account closing
7) Intelligent processing of refunds
8) Credit finance processing
The savings potential across these types of opportunities for financial institutions is estimated to be in the region of 20-25% in respect of both processing time and cost.
Over the course of this blog, we have discussed the retail banking sector and highlighted unbounded potential for the deployment of ML & AI across the domain. Having covered fraud detection, vulnerable customer identification, customer engagement, customer sentiment and intelligent document ingestion we can conclude that:
1) ML & AI can support faster resolution of customer queries
2) The combination of ML & AI can enable firms to reduce complexity & cost overheads
3) RPA can be used to streamline processes and expedite business process execution
4) Customer experience and satisfaction can be increased with ML & AI
5) Firms can increase their ability to track, audit and evidence compliance across products with ML & AI
In our next blog, we will delve into the opportunities that the investment banking sector has in respect of deploying ML & AI.
Interested in seeing our latest blogs as soon as they get released? Sign up for our newsletter using the form below, and also follow us on LinkedIn.