AI promises to grow revenues, cut costs and reduce customer churn, but for that to happen banks need the right data architecture, access to useful and purposeful data and be able to explain to regulators the thinking inside the black box, writes Prema Varadhan
It’s still fairly early, but already it feels as if 2019 will be the year of artificial intelligence (AI). The conversation is really ramping up between technology providers, banks and global bodies such as the World Economic Forum are also joining in. Exciting times.
In January, there were 11 public sessions on AI at Davos – more than for any other topic. This is not just because everyone is keen to jump on the AI bandwagon. Rather, the field is so new and raises such deeply complex issues that it needs to be addressed in depth to ensure that AI delivers the right solutions, particularly in financial services.
Need for transparency
Perhaps first and foremost is the need to ensure that AI models are transparent and explainable. Financial institutions using AI must be able to demonstrate to regulators that a decision obtained from a black box is fair and just.
For example, a bank must be able to explain why a customer was turned down for a loan or why a group of customers was classified in a certain way. Today it can do so because it has set rules that are followed by staff, against which the decision can be checked. Down the line, AI may make those decisions following a model that will have learnt independently over time, making it far harder to explain. AI can lack transparency and traceability – anathema to regulators. While lots of work is being done in this area, a uniform solution has yet to be found.
The right kind of data
Another focus of research and debate is data. For AI to work at its best, with the minimum chance of bias, it must be powered by huge amounts of high quality data. All this data must be clean, as well as useful and purposeful to the desired outcome.
For banks, this means having the right data architecture so they can store data in the right format and have it readily available. That may sound simple but is incredibly hard for incumbent banks to achieve. Neo banks, however, are far more data savvy. And underlying this is the need for data at volume to ensure the algorithms and models are as good as they can be.
Greater social good
At a global forum on AI I attended last autumn, the issue of sufficient volumes of data was recognised as a major hurdle. The solution requires the co-operation of a large number of banks, sharing data for the greater good. This is particularly true in areas such as fraud mitigation and anti-money laundering.
While at first it might seem unimaginable for banks to share their data for fear of giving away commercially sensitive information and breaching data-privacy rules, there is a way of making it work. Banks can share anonymised data in order to collate information on the vast scale that AI systems require to spot patterns to identify more fraud and money-laundering activities.
The benefits of data sharing are so compelling that one regulator at the forum offered to create a sandbox environment to test anonymous data sharing. This should be welcomed – not least because the fraudsters are also looking at how to use AI technology for their criminal ends. It’s a case of the industry working together – financial institutions and regulators – to stay at least one step ahead.
But beating criminals isn’t the only motivation for banks to use AI. Temenos, at its Centre of Excellence, is looking into how AI can be used to help deliver new revenue streams. This includes delivering compelling customer experiences, greater personalization and smoother on-boarding to differentiate the bank’s offers as well as increase revenue generation from improved cross-selling and up-selling. There is also scope to reduce costs, better identify profitable versus unprofitable customers, build customer loyalty and cut churn.
All this is vital because as banking continues to digitise, making so much of the work homogenous in terms of costs, basic products and services, the need for players to differentiate themselves grows. By using AI, a bank can customise its products to the level of servicing a market of one. AI opens up the very real prospect of individual offers, tailored by algorithms to suit each customer, making each customer experience uniquely relevant. It will allow the bank to know and understand a customer’s needs so that it can market products and services appropriately. Already, our technology can spot when a customer is at a high risk of closing the account, helping to stem customer churn.
Back office benefits
Banks will also be able to take full advantage of AI to make the back office more efficient. For example, bot digital workers will be able to take on some phone activities such as reviewing and resolving payments-in-exception. Initially using a limited list of criteria to check against, the bot digital worker will quickly learn new patterns, develop new rules and cut the number of exceptions – improving the customer experience and reducing costs.
Similarly in debt collection, AI models will be able to spot customers at risk before they default so the bank can offer preventative solutions and ultimately slim down losses.
AI has the power to make banking proactive rather than reactive as it is today. To enable that, banks will need to have the right data architecture and the industry will need to work out a way of co-operating for the greater social good. But perhaps the biggest hurdle – one we are some way off overcoming – will be making the AI models transparent so regulators are sufficiently comfortable to allow the technology to be rolled out across all functions and departments. When that happens we will have intelligent banking.
Prema Varadhan is chief architect at Temenos. This article was originally published in AI Business.