Bank of America's Decade-Long AI Journey with Erica
Bank of America Corporation launched its artificial intelligence-powered chatbot, Erica, nearly ten years ago in 2016. After multiple upgrades and numerous patents, the platform now manages approximately 2 million customer interactions every day. This massive volume equals the work capacity of about 11,000 human employees, demonstrating the scale of automation achieved through artificial intelligence.
The Staggering Cost Behind Banking Technology
While these numbers appear impressive, they come with an enormous price tag. Over roughly the same period, the bank has invested close to $120 billion in technology. Last year's technology budget alone reached $12 billion, with $4 billion allocated for development – including enhancements to Erica and building new applications – on top of $8 billion required to maintain existing systems.
These astronomical figures have naturally raised questions from investors about the returns generated from such substantial investments. Some answers are beginning to surface, though they remain limited and come with important caveats.
Why Banking AI Demands Massive Investment
The high costs partly stem from the extreme caution companies must exercise when deploying new tools, particularly generative AI. Mistakes can severely damage customer trust and render investments worthless. Additionally, AI threatens to intensify competition problems by creating wider gaps between the largest lenders who can afford massive spending and smaller competitors.
Bank of America serves as a prime example: its annual technology budget exceeds the entire cost base of more than half the lenders in the KBW Banks index. Similarly, JPMorgan Chase & Co.'s $18 billion yearly technology spend surpasses total expenses at all but five other banks in the same index.
Measurable Returns from AI Investments
During its first investor day since 2011 held last week, Bank of America revealed significant gains from its technology investments. The bank's consumer division has dramatically reduced staff from 101,000 employees in 2011 to 55,000 this year, attributing this reduction entirely to improved technology. Since 2018, the bank has also cut fraud losses across the organization by half.
Artificial intelligence played a crucial role in achieving these results. Unlike many competitors, Bank of America developed all its technology internally rather than relying on Silicon Valley firms. This approach has positioned the bank alongside Capital One Financial Inc. as one of the largest owners of intellectual property in finance. Together, these two institutions hold 65% of all AI-related patents owned by banks, according to Wells Fargo & Co. analysts.
The Challenge of Quantifying AI Returns
Despite increasing transparency about technology spending, most firms remain hesitant to disclose actual return on investment for AI initiatives. Available data appears somewhat disappointing. A Boston Consulting Group survey this year found that fewer than half of 280 finance executives could quantify returns on their AI investments. Among those who could measure returns, one-third reported payback of less than 5%, while another quarter indicated returns between 5% and 10%.
Part of the challenge stems from the absence of ready-to-use AI products comparable to Microsoft's Excel spreadsheets. Even institutions partnering with major generative AI companies – like Morgan Stanley's collaboration with OpenAI – must invest significant time and money to transform large language models into practical tools, whether customer-facing chatbots or internal assistants for research and sales ideas.
The Hidden Costs of AI Implementation
Before even beginning AI projects, companies must invest heavily in preparing their data. This involves cleaning, organizing, and labeling information to make it suitable for AI applications. Morgan Stanley spent several years on data preparation before considering AI implementation. Bank of America invested $3 billion between 2014 and 2019 just to make its data usable.
JPMorgan provides a clearer picture of AI returns, spending approximately $2 billion annually on AI projects while generating nearly $2 billion in yearly cost savings, largely from fraud reduction. However, this doesn't represent a 100% return on investment, as substantial previous data and technology expenditures were necessary to reach the point where AI could deliver value.
The Imperative of Perfection in Banking AI
Banks investing enormous sums in software development must still allocate significant resources for exhaustive testing before product launches. Bank of America CEO Brian Moynihan emphasized this point regarding their AI platform: "It has to be perfect."
Moynihan explained the critical importance of reliability: "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." This urgency applies not only to heavily regulated banks but to any company using AI, whether serving individual customers, other businesses, or internal staff.
While AI promises substantial rewards in efficiency and service personalization, the required time and financial investments are enormous and mostly upfront, with no guaranteed success. As AI increasingly delivers on its promises, the largest and wealthiest companies stand to gain the most, potentially creating competition concerns that politicians and regulators should address sooner rather than later.
