Bank of America Corp. first launched its artificial-intelligence driven chatbot, Erica, nearly a decade ago in 2016. Several iterations and a wealth of patents later, the platform handles about 2 million customer interactions each day, the equivalent of what 11,000 employees could do.
If that sounds impressive, the flipside is the cost: the company has spent nearly $120 billion on technology over roughly the same period, and last year’s $12 billion tech budget included $4 billion for development, including improving Erica and building new apps, on top of the $8 billion required to maintain existing systems. These are huge sums and investors in many big banks have long asked what returns they’re getting for this cash. It’s good that some answers are starting to emerge — but they’re somewhat limited and there are two important warnings in this story.
First, costs are high in part because companies must be extremely cautious in deploying new tools, especially generative AI, because mistakes can be ruinous for trust and waste the investment. Second, AI promises to turbocharge competition problems because it looks likely to put even greater distance between the largest lenders that can spend the most and the rest of the pack.
Bank of America is a case in point: Its yearly tech budget is greater than the entire cost base of more than half the lenders in the KBW Banks index. JPMorgan Chase & Co.’s $18 billion annual tech spend is greater than total expenses at all but five other banks in the index.
Details on what gains BofA has got for its money were the most interesting parts of last week’s investor day, its first since 2011. The bank’s consumer arm has cut staff to 55,000 this year from 101,000 in 2011, entirely due to better technology, it said. Since 2018, it’s also slashed fraud losses across the bank by half, it added.
AI has been a big part of this. BofA has built everything itself rather than using Silicon Valley firms, which has made it one of the biggest owners of intellectual property in finance alongside Capital One Financial Inc. These two account for 65% of all AI-related patents owned by banks, according to analysts at Wells Fargo & Co.
But while more firms are admitting to what they spend on tech, they’re still giving little away about the actual return on investment for AI — and what data there is looks disappointing. Fewer than half of 280 finance executives surveyed by Boston Consulting Group this year could quantify returns on AI investment at all. Of those that could, one-third pegged their payback at less than 5% so far, while another quarter put it at between 5% and 10%.
Part of the problem is that there’s no off-the-shelf product to pick up and slot in, as there was with Microsoft’s Excel spreadsheets, for example. Even those that choose to work with a major GenAI company — like Morgan Stanley did with OpenAI — still need to invest a lot of time and money to turn a large language model into a useful tool, whether that’s a public-facing chatbot or an internal assistant for research or sales ideas.
Even before a firm gets that far, it needs to have spent time and money on its data to make it useful for any kind of AI project – that means cleaning, sorting and labelling it all. Morgan Stanley spent several years doing this even before it started to think about working with AI. Bank of America spent $3 billion between 2014 and 2019 on making its own data useable. Banks have been doing this for other regulatory and business reasons, but it highlights the costs of just getting to the start line for an AI project.
JPMorgan is spending about $2 billion a year on AI projects, and it disclosed last year that these are leading to cost savings of almost $2 billion a year, much of which is fraud related. But that doesn’t mean it’s making a 100% return on investment — a lot of other data and tech spend got JPMorgan to the point where AI could even start to be useful. Big tech budgets are helping big banks leap ahead.
Even banks that can invest such mind-boggling sums on building and improving software still need to spend heavily on testing to the point of destruction before they can roll products out. Brian Moynihan, BofA chief executive officer, made the point about its AI platform simply last week: “It has to be perfect.”
“If people lose trust in that answer [from Erica], 11,000 people have to be put on the phones and in the branches tomorrow. Tomorrow,” he said, with emphasis.
This isn’t just about banks, whose duties towards clients are heavily regulated. The basic dynamic Moynihan described is true for any company, no matter whether the users of its AI are individual customers, other companies or its own staff. The eventual rewards from AI in efficiency and maybe personalization of service may show immense promise, but the time and money required to reach those are also great and mostly paid upfront. And there’s no guarantee of success.
The more that AI delivers on its promises, the more companies that are already the biggest and richest will land those rewards. One day, maybe sooner than we think, that will create a competition problem that politicians and regulators should start thinking about how to address.
A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts.
Bloomberg News provided this article. For more articles like this please visit
bloomberg.com.
Read more articles by Paul J. Davies
