Artificial Intelligence (AI) and Machine Learning (ML) are already playing a significant role
in the efficient delivery of some services in key industry sectors. Among the front runners
are the consumer retail sector and financial services.
Evolving Enterprise commissioned leading industry analysts, Juniper Research, to assess the innovations in these two sectors and to see what lessons other industries can learn from their ground-breaking experience.
The introduction of AI in retail and financial services
While financial services and retail are highly different markets with different challenges and drivers, one consistent similarity is the need to digitally transform and innovate. For financial services, the rise of digital channels has meant that financial players have had to transform the way they interact with users, which has required fundamental change throughout organisations. In the retail market, the rise of eCommerce has meant that retailers now face an increasingly omnichannel world, where they must engage in digital strategies, or they will succumb to competition from eCommerce vendors. These challenges have meant that stakeholders in both markets have sought out technologies which can
accelerate their digital strategies and deliver fundamental improvements in their ways of working. As one similarity between the markets is the widespread availability of data, AI has emerged as essential to efforts in both sectors.
The rise of AI
Before analysing its deployment in fintech and retail, it is important to establish a definition for AI, as the area suffers from confused definitions.
Juniper Research defines Artificial Intelligence as a computer program that uses a combination of digital building blocks, such as mathematics, algorithms, and data, to solve complex problems normally performed by humans.
AI is not generally used for complicated tasks; it is merely scaling tasks at a rate that cannot be matched by humans. By utilising AI to undertake previously manual tasks, human intervention can be reduced to near zero, while increasing the total volume of tasks undertaken. AI use across fintech and retail generally requires certain common capabilities, which are important to recognise when analysing use cases in these sectors. In this context, the definition of AI does not distinguish between different types of AI, such as deep learning and supervised learning. Over time, more advanced AI techniques will be used, but this is limited at present in these segments.
The use of AI in fintech
Ever since the advent of computer use by financial institutions (FIs) in the early 1950s, financial services has been a sector that generates an enormous amount of data. This has meant that FIs have consistently adopted analytics tools to understand the wealth of data they generate. In this context, it is unsurprising that financial services is one of the first sectors to be significantly disrupted by AI deployment.
However, for AI to reach scale in financial services the ‘black box’ challenge must be resolved, which means that whenever AI is used, there is a fundamental problem with implementing it in important use cases. Current rules-based systems follow a traceable approach, where decision processes can be recreated. For AI systems, this is not as straightforward. Algorithms are so complex that it is difficult to establish their reasons for reaching a certain decision. In processes where transparency is a must, such as regulatory compliance or lending, this is a challenge. Vendors must focus on boosting the ‘explainability’ of AI systems in order to combat this challenge.
The most progress in terms of AI deployment in financial services has been made in the following segments:
• Regulatory technology (Regtech)
For the purposes of our comparison with retail, we will focus on the two highest value areas, Insurtech and Regtech.
Regtech: AI for transaction monitoring
Transaction monitoring is a key task in the prevention of money laundering, which is a significant cost to the economy. Transaction monitoring systems (TMS) are an integral part of fraud prevention and compliance with anti-money laundering (AML), yet transaction monitoring is among the most difficult of compliance actions for large FIs to complete.
The estimated amount of money laundered globally in one year is 2-5% of global GDP, or at least $800 billion (possibly as much as $2 trillion).
It is in this context that the magnitude of the AML task facing FIs becomes apparent. Market globalisation has meant a significant increase in the number of customers to check, and TMS are incapable of handling this enormous task in an effective manner. The volume of alerts generated by current systems is such that 90% of them are ignored, with 8 out of 10 of the remaining 10% turning out to be system-generated false positives.
It is clear that current rules-based approaches are not sufficient to address the complexity of the task. False positives in the know your customer (KYC) and AML processes are a drain on time and resources. The aim for companies must be to achieve maximum process efficiencies while keeping down the costs associated with regulatory compliance. Consequently, AI is being progressively introduced as a solution to improve TMS.
AI is uniquely suited to enhancing TMS. TMS as machine learning techniques mean that an AI system will learn how to interpret data more effectively over time, increasing the process’ efficiency. An AI system in this area can be trained effectively by feeding it data about previously confirmed instances of money laundering. This will mean that AI will produce fewer false positives than a traditional TMS, reducing compliance team’s workloads dramatically. There can also be an improvement in the data which can be used in transaction monitoring. By leveraging natural language processing (NLP), AI can understand notes and other more diverse forms of data, which can enhance transaction monitoring processes.
Automation of the repetitive processes associated with AML tasks (onboarding, transaction monitoring and fraud) will enable companies to identify the relevant information more quickly, achieve consistency and increase the responsiveness to issues identified.
AI in insuretech:
- Conversational platforms – AI-enabled conversational platforms allow insurers to use NLP to communicate with their customers in a similarly personalised way as human-to-human interaction, while capturing the relevant information. Machine learning allows conversational platforms to train algorithms and adapt, thereby continuously improving underwriting processes. Some of the use cases for this technology are:
- Chatbot sales agents – for less complex insurance types, such as travel, home or motor, chatbots can potentially act as the sales mechanism. A chatbot can retrieve only the relevant information to provide an automated quote, saving time across the process. Retrieving information in a conversational way can result in less consumer frustration and fewer errors, higher quality data, as well as a more personalised process.
- Policy queries – chatbots are well suited to deal with queries relating to policy changes or terms and conditions, which can function without large customer service teams.
- Claims reporting – particularly in motor insurance claims, details can be collected via a standard set of questions. This process naturally lends itself to chatbot solutions, facilitating accident reporting, initiating the claims process and providing a smoother service to policy holders. It may also help with fraud detection, through analysing question responses.
The use of AI in retail
The emergence of eCommerce giants such as Amazon and eBay from the mid 1990s has created a long-term challenge for traditional retailers. Many have met this challenge by attempting to offer a compelling omnichannel experience, where the physical retail and online channels merge through the use of solutions such as click and collect. While this has dramatically improved the customer experience, it has not prevented a reckoning for many traditional retailers, with multiple large-scale insolvencies, including Toys R Us, Sears and HMV. Therefore, retailers must further adapt their business models to protect their futures in the face of advanced competition. In order to do so, there is increasing momentum behind leveraging the tools that eCommerce players have used to such great effect, particularly AI.
High value use cases
To understand the potential of AI in retail, it is important to examine the highest value use cases.
I. Visual search
One of the biggest ways in which AI is being used in retail to deliver personalised services in visual search. AI systems are uniquely qualified to help narrow down choices, personalise product catalogues or to find similar products using visual search. Retailers can leverage computer vision (CV) technologies to enable systems to understand product pictures, identifying key characteristics, such as colour, material and style. This data is then combined with metadata from product catalogues, enabling the systems to automatically find a product that has similar characteristics.
One such example of visual search is in the clothing retailer UNIQLO, which uses ViSenze’s Search by Image function to merge the offline and online retail channels. Shoppers take a picture of an item, which is then analysed to find matching products online. Visual search systems are also useful in optimising a retailer’s own operations. By using a visual tagging system, retailers can automatically tag their items with the most appropriate metadata, allowing more effective organisation of items on a website. It also enables features such as search for similar, which will find comparable items based on metadata.
II. Customer analytics and marketing
AI is also uniquely useful in allowing retailers to analyse their customers. As shopping habits have changed, the value of knowing the customer has not diminished, in fact, if anything the value of customer information is far higher. However, shopping is now largely impersonal, which has implications for customer retention
This is also important in the context of changing shopping habits. In North America and Europe, discount retailers have enduring popularity, offering branded goods at low price points. Stagnation in wages and slow economic growth has fuelled this. Retailers that offer goods at high prices need to understand their customers better and target them more effectively, maximising their Average Revenue per User (ARPU).
Knowing their customer allows connected retailers to more effectively handle their relationships with their most important customers. By using analytics systems, retailers can look at a customer’s transaction history and immediately establish patterns which may be related to selling to them. Once these patterns are identified, the business can act on them in the hope of increasing their purchases. This will primarily take the form of specifically targeted marketing, which will attempt to personalise marketing and provide offers that are relevant to the customer, based on their previous purchases or interests. This can increasingly be determined from other sources, such as integration with social media, which has become popular.
In terms of marketing, AI offers great opportunities for designing optimal website layouts and automating A/B testing. Instead of running individual manual tests, testing can be fully automated, with many permutations checked at once. This allows retailers to optimise their marketing strategies on the fly, swiftly making changes that make users more engaged with a retailer’s online content and store
AI clearly has potential in both financial services and retail, across numerous use cases. To date, this potential is more realised in retail than in financial services, as retailers are faster to adapt to new methods of working.
Banks and other financial institutions can be slow to adopt new technologies. There are also fewer regulatory concerns in the retail market. While data protection rules generally mean data must be handled for a specific, justifiable purpose, retailers are used to collecting and handling data for loyalty schemes. This loyalty scheme data can then be analysed by AI, adding an extra layer of value to existing data sets. This is now even more valuable, given the fact that many retailers are omnichannel, so many schemes cover online and offline channels.
In contrast in financial services, there are massively complicated regulatory frameworks which are delaying innovation, meaning that AI will be restricted to a largely advisory role. The creation of more explainable and transparent AI systems will accelerate AI innovation in financial services.