In 2010, Bank of America’s consumer banking division employed 100,000 people and managed $400 billion in deposits.
Today, per the bank’s November 2025 Investor Day, roughly 53,000 employees manage over $900 billion.
The bank rebuilt its operating model around AI with 270+ models in production, $13 billion annual tech spend, and a virtual assistant that has handled 3.2 billion client interactions since 2018.
More than 90% of 213,000 employees across all business lines now use AI tools daily at the Bank of America (BofA). That’s the highest adoption rate of any bank we’ve covered.
In today’s AI at the Top, we will learn about:
BofA’s architectural decision
Internal AI tools and use cases
Lessons from BofA you can apply at work
How BofA approaches its AI Strategy
The platform that later became a foundation for internal AI tools
Most enterprises build AI tool-first. A chatbot here, a copilot there. Each team solves its own problem.
But BofA invested in one foundational AI platform, Erica, and reused its NLP engine across consumer banking, wealth management, commercial banking, employee support, and global payments.
When asked about why focus on a foundational platform like Erica instead of shipping multiple tools for each department, CTIO Hari Gopalkrishnan told Fortune:
“The biggest challenge is to have institutional patience to recognise the value you get by slowing down a bit upfront, accelerates you a lot going forward.”
This patience is now compounding.
The NLP engine built for consumer Erica in 2018 now powers multiple internal AI tools.
Governance with separate risk tiers
BofA doesn’t govern all AI the same way.
Client-facing tools like Erica use proprietary AI. These are rule-based systems where outputs are predictable and controlled. The bank can’t afford hallucinations when a customer asks about their mortgage.
But but… internal tools like coding assistants, research summarisation, meeting prep, etc., use genAI with human-in-the-loop controls. Relatively low stakes and fast iteration.
Also, an AI council co-led by the CTIO evaluates every use case through a 16-parameter framework.
From 45+ proofs of concept, only 15 generative AI use cases made it to production.
The enterprise builds on top of existing tech
Like its competitor, Goldman Sachs, BofA takes a multi-model approach by working with OpenAI, Anthropic, Nvidia, Google, Cohere, and AWS.
Then the enterprise builds on top of existing technologies over acquiring niche models or startups.
BofA employs 60,000 technology workers internally and holds 1,500+ AI/ML patents, a 94% increase since 2022. Its private cloud, consolidated from 67 data centres to 23, saves $2 billion annually.
“We don’t want to be building things that are increasingly foundational and are available as a commodity. We want to leverage the heck out of innovation happening in the industry.”
Enjoying this read? Share the smartest part with your followers. Click to Share.
YOU DECIDE
🌟 Want to be featured in the next issue? Reach out with your best AI use case and we’ll spotlight it.
How BofA uses AI internally
Erica, the compounding consumer assistant
Erica launched in 2018 as banking’s first widely adopted AI assistant.
In seven years, it has handled 3.2 billion client interactions with:
20.6 million active users
58 million interactions per month
700+ intent types (up from 200 at launch)
98% success rate
Average interaction under 48 seconds
50-60% of interactions are now initiated by Erica. Think flagging duplicate charges, reminding about bills, and surfacing spending patterns. This contributed to a 60% reduction in service call volume.
AskGPS, BofA’s internal AI assistant
In September 2025, BofA launched AskGPS, a generative AI assistant for its Global Payments Solutions division.
It’s trained on 3,200+ internal documents and serves employees supporting 40,000+ business clients across 29 languages.
Before AskGPS, answering a complex payments query meant calls across regions and time zones. Now it resolves in seconds.
AskGPS runs on the same platform architecture as Erica. Remember, we talked about using Erica as a foundational platform?
Different business unit, same foundation.
AI in fraud detection
BofA runs 50+ AI-enabled fraud detection models.
These combine supervised learning (known fraud patterns), unsupervised learning (detecting novel schemes), and graph neural networks (mapping connections between seemingly unrelated transactions to catch coordinated fraud rings).
At the November 2025 Investor Day, the leadership mentioned it had cut the fraud loss rates in half.
Beyond these three use cases, BofA has deployed AI tools across every major function:
How BofA Compares to the banks we’ve covered
We’ve now studied three of the largest banks. JPMC, Goldman Sachs and Bank of America.
Each took a different path to enterprise AI:
This tells us there’s no one right way to approach your enterprise’s AI strategy. All three banks we covered in the AI at the Top have unlocked real business efficiencies while saving time and budgets.
Companies choose based on their datasets, governance preferences, talents, and business preferences.
What’s next for Bank of America
While large banks are racing towards autonomous agents, CTIO Gopalkrishnan values practical orchestration over complex agentic AI. The bank views Erica as an agent that predated the buzzword and is adding reasoning capabilities incrementally.
As short-term plans, Erica is expanding to desktop (currently mobile-only), generative AI capabilities layered into Erica for Employees, and the 20% developer productivity gains are being reinvested into new growth programs in 2026.
What Enterprise Leaders Can Learn from Bank of America's AI Strategy
Platform patience compounds. BofA's foundation approach was slower upfront. Now every new tool deploys faster because the infrastructure already exists. AskGPS launched on the existing Erica architecture in a fraction of the time a standalone build would take.
Bifurcate your AI governance. Client-facing = Rule-based systems. Internal = generative, human-in-the-loop. BofA doesn’t treat all AI risks the same. This lets you move fast where the stakes are lower without compromising where they’re high.
Start with the mundane to get 90% adoption. Erica for Employees handles IT tickets and HR queries. Not glamorous. But it made 213,000 people daily AI users before the bank asked them to adopt more complex tools.
Governance accelerates deployment, not slows it. The 16-pillar framework prevents expensive dead ends. Similar to what we saw at Goldman Sachs and Mastercard, embedding compliance into the platform from day one means less friction when it’s time to scale.
Sources:
YOUR TURN