Your call is important to AI

We’ve all been there. You’ve called your bank for some reason and while on hold they ask you to enter your ID or credit card number to speed up the process. Then the agent answers and asks for the same number.

Capitec has just launched their AI agent called Pulse especially to deal with customer questions. If you’re contacting the bank from within the app, Capitec says call handling times are reduced by 18%. Importantly, customers are not dealing with the AI, rather the AI is helping the support agent. UK based neobank Monzo claim that over 40% of self service requests are resolved by AI with no human intervention. So that’s definitely some progress.

So far AI seems to have been used more for customer experience, call centres and replying to emails. What we’re not seeing is much real time usage to decide on risk profiles or help us invest better.

When the bot gets fired

Of course spending the money and diving right in doesn’t ensure success. Klarna, a Swedish Buy Now Pay Later (BNPL) firm, replaced 700 customer service staff with an OpenAI chatbot. The bot handled 2.3million conversations in 35 languages saving $40million. Then customer satisfaction tanked, complaints spiked. The CEO admitted they went too far, and they fired the AI and reverted to hiring humans for customer support.

Another area that AI is being used is fraud but here it all seems to be about responding, rather than preventing. In May 2020 FNB launched their Manila system, apparently named after the “The Thrilla in Manila”. It’s designed to keep an eye on customer accounts, monitoring some 50 data sources and if an account is flagged it’s able to write natural language reports in seconds rather than hours. Sounds great, but the money has bolted. It’s reactive not preventive.

Your selfies are your credit score

Last year a collaboration called Telco Data Score was announced between MTN and TransUnion. The system draws on call data records from mobile network operators. This includes over 400 telco data variables, such as voice and data usage patterns, airtime top-up frequency and amounts, device information, location data, and SIM card tenure. The data is then used to provide a credit score and enable formal lending rather than using a mashonisa. Singapoe based Credolab is using the number of selfies you’re taking, games you’ve installed and your typing speed to deliver a credit score. Creepy but very useful for the unbanked who can now avoid loan sharks.

The black box bites back

Of course this is not without risk. Apple had a problem with their credit card offering via Goldman Sachs. The Goldman AI credit algorithm was alleged to have given women lower credit limits than men with similar financial profiles. Goldman Sachs was investigated by New York State regulators for gender bias. The investigation found no evidence of unlawful discrimination, but the reputational damage was done and ultimately they exited consumer credit entirely.

The issue is that black box, which nobody can really look into. In a recent episode of the Odd Lots podcast Bill Demchak, CEO of PNC Financial Services, commented that they were not using AI to help make decisions on loan applications. He reasoned that if the AI rejected the loan they wouldn’t be able to tell the client why. Yes, a black box that nobody, not even the creator, can stare into. As soon as the banks start using AI for more than just improved support, the problems multiply.

Eighteen billion dollars and counting

But in America things are always bigger. Their largest bank, JPMorgan Chase, is spending $18 billion a year on tech and a lot of that is going on AI. To put $18 billion into perspective, they could buy Nedbank and Shoprite outright and still have money for lunch.

For JPMorgan Chase, part of that budget is being spent on 450 AI ‘proof of concept’ ideas and they want to have 1,000 of these ideas by the end of the year. Some 200,000 employees have access to their internal LLM suite. Their “Coach AI” tool apparently improved client response times by 95% during market volatility and investment bankers now create five-page pitch decks in 30 seconds, work that used to take junior analysts hours.

Another neobank, Nubank says they’re building an AI private banking experience for everybody at a cost of $0.90 per customer. Colour me sceptical. Isn’t the point of a private banker the idea of being able to request outlandish things they can resolve? Can an AI really do that? We may find out soon enough, as Nubank has a stake in recently rebranded local Gotyme Bank. So maybe they bring this cheap AI private bank to South Africa?

Brace yourself

However, regardless of the risks and the awkward mistakes, AI is only going to get more space in banking. McKinsey estimates AI could add around $300 billion in annual value to banking globally, and all the banks will want a slice of that.

But do expect more AI agents from banks, some of it will be great. Equally, expect to find yourself yelling into your phone / keyboard as an AI bot goes rogue and straight up lies to you.

Simon Brown

AI in the Wild is a regular column from Simon Brown.

AI is everywhere and only getting better. Record capital raises and valuations, and competing LLMs are all fun, but meaningless to our every day lives. This column will focus on how is it impacting us in the real world.

5th | 15th | 25th every month