Digital forces and PSD2 have revolutionised payment services, encouraging customers to make more frequent purchases via online retail channels and through contactless methods. As a result, banks have exciting opportunities to embrace AI, – but they require robust, clean big data and advanced analytics to compete with new market entrants. Temenos’ Steen Jensen, MD Europe, reviews the benefits and steps forward.
Making sense of payments
Disruptive forces in the world of regulation, business and technology has led to a need for banks to be more efficient than ever. Regulation such as PSD2 and open banking has given fintechs the freedom to access established banking systems, and create new products. The ease by which we buy goods and services: by credit card, direct debit, e-mobile, e-wallets – even our watches – is creating a robust cashless society that is getting bigger by the day. As a result, banks want, and need, to find new ways to process and make sense of this payment data; the richest source of data available to a bank. If cloud technology can help us support increasing volumes and efficiency gains, then what tools do banks have to make sense of it? Enter Artificial Intelligence.
Improving the customer experience
Right product, right time
With big data at their fingertips, financial institutions can analyse behaviour and effectively target customers with offers. For example, an Asian consumer bank used Machine Learning to analyse key datasets and discovered unsuspected similarities. As a result, it has defined15,000 microsegments in its customer base, built a next-product-to-buy model, and tripled the likelihood that a customer will buy.
Convenience for customers
Machine learning can enable quick and painless peer-to- peer payments through integrations. PayPal is already utilizing this technology by integrating its ‘bot’ with Facebook Messenger, Siri, and Slack. This allows users to make direct p2p payments without leaving their preferred app or platform. To pay a friend for a meal within Slack, simply type “/PayPal send $25 to @Sarah”.
In future, payments could be facilitated using SMS, without the need to go through platforms such as Slack at all. Using chatbots, texts could easily pay other people, review the status of group payments, check mobile wallet balances, and remind others that they owe you money. Mypoolin already offers a mobile payments chatbot that enables users to send or receive payments via text messaging. All that is required is a command such as ‘transfer [amount] [phone number]’ for the recipient to receive the money instantly once the payment is processed.
In April 2018, Wells Fargo began piloting an AI-driven chatbot through the Facebook Messenger platform with several hundred employees. It communicates with users to provide account information and helps customers reset their passwords. As a result, AI turned a lengthy trawl through the company website into a simple chat, saving time and effort in the process.
We know that a payments hub in the cloud offers huge efficiency gains, however Artificial Intelligence (AI) can also be used to further enhance operations. Reducing processing times is a key application of AI. With the volume of transactions and the number of customers set to increase massively, traditional customer care is not an option. Natural Language Processing (NLP) allows computers to process and accurately understand human speech, learning as they go. NLP is expected to power the next generation of customer service. As processing power increases and bots are able to understand human language in more flexible ways, the payments industry will shift to a bot-powered frontline in customer service, which has benefits around many everyday payment processes.
Using NLP, customers may receive a new credit card and set up a recurring online subscription using just their voice. The bot is able to inform the user when new products are available or when the card is set to expire. All this without any manual intervention.
The UK’s Co-operative Bank has used a branch of AI known as robotic process automation (RPA) to cut wire-payments processing from an average of 10 minutes to just 10 seconds. Similarly, it has slashed an audit trail on problem transactions from seven hours to 10 minutes. Even in a low-cost environment this technology is useful. ICICI bank in India started with half a dozen RPAs and is now running 500. On average, the bank reports a total 60% time saving.
Using machine learning, payment providers can establish robust baselines of what is normal for each customer. When a transaction conforms, the risk can be considered lower and therefore allowed to proceed. If the transaction does not confirm then further authentication is required. This goes one step beyond biometric authentification which can still be compromised and therefore requires dual-level checks.
Better lifestyle management for customers
From encouraging us to take more steps via our Fitbit or reminders about a cab we’ve booked: lifestyle tools do a lot for us. However, AI and machine learning in particular, is set to take this to another level, particularly from a payments data perspective. Citi Ventures is already active in this space with its Clarity Money app. This motivates customers to participate in third-party services that can improve financial health. For example, it can see how much you pay in interest on one credit card, and could show you one with a much lower rate.
AI in payments can even help us manage our own physical and mental wellbeing, and that of others. For example, I may choose to know when a sibling undergoing treatment for alcoholism makes a restricted purchase, or makes a non-restricted purchase at an unusual time or location.
Moving forward: data considerations
When it comes to maximising the AI in payments opportunity, the adage ‘garbage in, garbage out’ holds stronger than ever. The importance of robust data cannot be understated. Take this analogy from Andre Burrell, Director of Compliance at Microsoft:
‘If you have a child and you are telling him not to touch the hot gas stove, when they eventually touch it, that child will take that one experience and generalise it to life. So if they see a hot pipe, and they sense the heat, they know not to touch it.”
“To train a machine to do something like that, you have to have lots of data, which must be relatively good and clean, so that the machine can take the features that enable you to identify what is a hot stove. Now if you then present the machine with an electric stove, so now that it looks different, you have to retrain the machine to allow it to identify with these different features.”
What can your bank do?
Start small, by cleaning up product data, and ensuring each has accurate metadata to set it apart not only in text, but voice or visual search. Then, consider your infrastructure. Banks must have centralised systems to enable instant access to this data. For machine learning in particular, it’s vital to set up the right data ports to obtain seamless access, asking customers up front to understand which data points they will interact with.
Finally, consider investing in a productized Data Lake. This will overcome the complexity high failure rates associated with a bespoke build. A Data Lake, offered by Temenos, will store and process all the data required to power AI applications using data the bank already holds, and at a lower TCO. Moreover, the set of data engineering tools will allow data scientists to blend and enrich the source data with a whole host of other useful data sources including unstructured data. Banks have the flexibility to deploy this as a standalone solution or embedded with core banking products, both on premise and in the cloud.