What proprietary AI has Bank of America built in-house?
Bank of America builds its AI on platforms it owns and routes the hardest reasoning to outside models. Erica, launched in 2018, was built in-house on open-source language models and trained on client interactions; as of March 2025 it used no generative AI or large language models, and its team of data scientists had made more than 50,000 background improvements. By August 2026 Erica had passed 3.6 billion interactions, and several patents by inventor Cameron Wadley sit at its foundation, including 'System for Predictive Usage of Resources', used in Erica's Forecast my Balance. The bank rebuilt Erica in 2025 as a second-generation, model-agnostic platform; generative AI reached EricaAssist, used by more than 18,000 contact-center agents, in July 2026, while a generative 'Erica 2.0' for clients is in development with no timetable. Around Erica sit AskGPS, a generative assistant for about 3,000 Global Payments Solutions staff (2025, extended with an Intelligence Hub on September 28, 2026), a marketing simulator built from 30 in-house AI models, and a large language model that improves OCR. In December 2025 the bank was running more than 270 AI and machine-learning models in production. An orchestration layer called Orchestra sends simple classification to approved open-weight models on the bank's own GPUs and harder reasoning to proprietary models whose suppliers it has not named.[1][2][3][4][5][6][7][8][9]
Data as of Oct 22, 2026 · BofA profile updated Oct 19, 2026 · 27 sources cited on this page · 140 banks profiled on this site · Markdown
Which AI assistants, models and tools has Bank of America built in-house?
Systems Bank of America describes as its own, with the base technology executives have named. Scale figures are as of the source date.
| Platform or model | What it does | Built on | Since | Status |
|---|---|---|---|---|
| Erica (clients) | Virtual assistant in the app; solves 98% of inquiries without further human help; 60% of interactions are proactive outreach; 24.6 million users and 200 million interactions in Q2 2026; more than 1.9 billion proactive insights since launch | In-house natural-language processing and machine learning on open-source models, trained on client interactions; no generative AI or LLMs as of March 2025 | June 2018 | In production; generative 'Erica 2.0' in development[8][1][2] |
| EricaAssist | Always-on widget for contact-center agents: generative AI now summarizes each call in real time and surfaces guidance in seconds, such as which of about a thousand reasons caused a card decline; more than 18,000 agents | The Erica platform plus generative AI; models not named | October 2020; generative AI July 2026 | In production[1][6] |
| Erica in CashPro | Erica inside the corporate CashPro app; live chat volume down 42% | The Erica platform | August 2023 | In production[8] |
| AskGPS and AskGPS Intelligence Hub | Generative assistant over thousands of vetted internal documents for about 3,000 Global Payments Solutions employees; more than 30,000 staff queries answered by December 2025; the Hub adds Intelligent Treasury Management Reviews, Digital Account Schematics and Enhanced Relationship Insights | The bank's generative AI application and proprietary payments data | 2025; Hub September 28, 2026 | In production; Hub rolling out in phases[8][7] |
| Marketing simulator | Finds the best return on a marketing dollar across message, product, channel and geography; campaign conversion rates running 60% higher | 30 AI models developed by the CTIO's teams | Reported December 2025 | In production[8] |
| LLM-enhanced OCR | A large language model deployed to make optical character recognition 'much more accurate' | Large language model, not named | Reported December 2025 | In production[8] |
| Merrill SENSE | Large-language-model search engine for marketing teams; among the AI capabilities that won innovation awards the bank announced in October 2025 | Large language model, not named | October 2025 | In production[10] |
| Model inventory | More than 270 AI and machine-learning models in production across the bank; the AI and machine-learning patent portfolio has more than doubled since 2022 | Mostly in-house | December 2025 | In production[8] |
Which Bank of America teams and people build its proprietary AI?
As named by the bank and the trade press. Bank of America does not publish a separate AI research lab or an AI headcount.
| Team or person | What they build | Role | Source date |
|---|---|---|---|
| Technology organization | The 30 models behind the marketing simulator; the Orchestra routing layer | Hari Gopalkrishnan, chief technology and information officer | December 2025; October 2026[8][9] |
| Consumer technology | Erica, EricaAssist and the planned Erica 2.0 | Tom Ellis, CIO and head of consumer technology | August 2026[1] |
| Erica team | A team of data scientists that reviews Erica's behavior and tunes its answers to bank policy | Jorge Camargo, managing director, mobile app, online banking and Erica AI executive | March and August 2025[2][11] |
| Consumer client experience | Chose the first generative tasks for EricaAssist from contact-center feedback | Ashley Ross, head of consumer client experience and business transformation | August 2026[1] |
| Global Payments Solutions data and AI | AskGPS and the Intelligence Hub | Jarrett Bruhn, head of Data and Artificial Intelligence for GPS | September 2026[7] |
| Inventors | Patents incorporated into Erica, including predictive balance forecasting | Cameron Wadley (70 patent applications, holds more than 40) and Katherine Dintenfass | 2020; June 2026[4][3] |
Which Bank of America patents are tied to Erica?
American Banker reported in September 2020 that Katherine Dintenfass and Cameron Wadley obtained a patent called 'System for Predictive Usage of Resources' that year, partially used in Erica's Forecast my Balance capability. Its logic covers confirmed upcoming bill payments, seasonal payments from transaction history, anticipated discretionary spending and savings, and Wadley said many of the patented inventions have been incorporated into Erica.[4]
Banking Dive reported in June 2026 that several of Wadley's patents are at the foundation of Erica, which itself came out of an internal innovation challenge. Wadley explained the reason to patent: 'We don't want to do something based on feedback from clients that we roll out there and it delights them, and then all of a sudden we have to take it away because we didn't protect it with some sort of a patent.' The bank then held about 8,400 granted patents and applications, about 1,600 of them AI-focused.[3]
What does Bank of America say about building versus buying AI?
Erica is the template for building. 'We based it on the best AI available at the time,' Tom Ellis said of the 2018 launch. 'We used open-source models to do that. And we've trained it over time on the basis of client interactions.' Jorge Camargo says the value comes from continuing investment in tuning rather than a one-time build.[1][11]
Generative AI goes to employees first. Ellis said contact-center agents get new capabilities before clients 'because that way we've got a human in the loop', and Camargo said in March 2025 that for client-facing Erica 'there can be no hallucinations, no kinds of errors that would be impactful to our clients'.[1][2]
Bought models sit behind a bank-built router. Gopalkrishnan said in October 2026 that Orchestra sends simple classification to approved open-weight models on the bank's own GPUs and harder reasoning to proprietary models, which also controls token costs; coding assistance is bought, with 18,000 developers on GitHub Copilot reported in December 2025.[9][8]
What has Bank of America not disclosed about its proprietary AI?
The models behind EricaAssist's generative features and the proprietary models Orchestra routes to are not named, and Erica 2.0 has no announced timetable or feature list.[6][9][1]
The 270-plus production models are not itemized, and the patents tied to Erica are known only from press interviews; the bank has not published patent numbers for Erica or a count of AI engineers.[8][4][3]
What does the BofA profile record on this topic?
The rows below are the entries of the Bank of America AI strategy profile that bear on this page, as of Oct 19, 2026; the profile holds the full record of 46 dated moves, 18 use cases and 70 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Erica | Consumer virtual assistant; 2 million interactions a day. | In production | [12] |
| EricaAssist and employee tools | Human-assisted AI agent for customer-service staff; Merrill advisor meeting journey; 400,000 prompts a day across the workforce. | In production | [6][13] |
| CashPro Forecasting and Data Intelligence | AI cash forecasting for corporate clients; 3,000 companies, 250,000 hours saved in 2025. | In production | [14] |
| CashPro Chat for corporate clients | AI-driven chat used by nearly 70% of corporate clients for account information, transaction tracking and service resolution, with a 43% containment rate in Q3 2025; the Q2 2026 earnings release counts 40,100 CashPro Chat interactions supported by Erica technology. | In production | [15][16] |
| AskGPS Intelligence Hub and Payments Insights | CashPro's Payments Insights analyses payment efficiency, cross-border flows and working-capital performance for corporate clients; the AskGPS Intelligence Hub gives Global Payments Solutions staff Intelligent Treasury Management Reviews and Digital Account Schematics built on the bank's AskGPS generative-AI assistant. | Rolling out | [17] |
| Behavioural biometrics (BioCatch) | American Banker (March 2026) names Bank of America among banks using BioCatch, whose technology reads how a user logs in, the device, the angle the phone is held at, keystrokes and past behaviour to decide whether the user is a real customer, a fraudster or a bot; Bank of America has not described its deployment. | In production | [18] |
| AskGPS knowledge assistant | Global Payments Solutions assistant trained on 3,200 internal documents, speaks 29 languages and answers questions for staff serving 40,000-plus business clients. | In production | [19] |
| Capital Markets Insights and the TED score | An AI-driven algorithm on market variables produces a score of the macro backdrop for investment-grade issuance on a given day, shown in the CashPro App alongside same-day and monthly issuance data. | In production | [20] |
| Orchestra model-routing layer | Sends simple classification tasks to approved open-weight models on the bank's own GPUs and harder reasoning to proprietary models, which the CTIO says also controls token costs. | In production | [9] |
| EricaAssist at the contact centre | Desktop widget used by about 18,000 customer service employees; generative-AI additions tell the agent why the customer is calling and recommend next steps, and average call time falls by almost a minute. | In production | [21][22] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Oct 6, 2026 | Evident AI Index: seventh, up from tenth | American Banker reports Evident's 2026 ranking, led by JPMorganChase, Capital One and RBC, puts the bank seventh. | [23] |
| Oct 1, 2026 | CTIO at Fortune AIQ Summit: Orchestra routes work to open-weight or proprietary models | Gopalkrishnan says AI is often rejected in favour of deterministic rules, every project is reviewed against 16 risk pillars, simple classification runs on approved open-weight models on the bank's own GPUs, and agents stay assistive with humans in the loop. He gives Erica's total as 3.6 billion interactions. | [9] |
| Sep 29, 2026 | Fortune AIQ 75 places the bank 19th; EricaAssist at 18,000 service staff | Fortune ranks the bank 19th behind JPMorgan Chase (1st) and BNY (15th) among US banks listed and reports EricaAssist deployed to more than 18,000 customer service representatives this summer. | [22] |
| Sep 28, 2026 | AskGPS Intelligence Hub and CashPro Payments Insights launch | Global Payments Solutions rolls out Payments Insights, a CashPro feature analysing payment efficiency, cross-border flows and working-capital performance, alongside the AskGPS Intelligence Hub for GPS staff — Intelligent Treasury Management Reviews and Digital Account Schematics that build on the bank's existing AskGPS generative-AI assistant. | [17] |
| Jul 21, 2026 | EricaAssist gets generative AI | Employee-facing agent summarises the call reason and recommends next steps; average call time falls by nearly a minute. | [6] |
| Jun 12, 2026 | About 8,400 patents and applications, 1,600 AI-focused | Cameron Wadley, a wealth-management technology executive with 70 applications filed, says several of his patents underpin Erica and now run in all eight lines of business. | [3] |
| Mar 17, 2026 | 'Build once' strategy explained: second-generation Erica platform in 2025 | Fortune reports that an AI council with the head of strategy ran about 15 proofs of concept outside the Erica platform, found teams 'reinventing the wheel', and the bank rebuilt Erica as a second-generation, model-agnostic platform in 2025; the first generation started on BERT and OpenNLP. | [5] |
| Mar 10, 2026 | 30 billion digital client interactions in 2025 | 20.6 million users interacted with Erica nearly 700 million times in 2025; cumulative Erica interactions passed 3.2 billion. | [24] |
| Jan 22, 2026 | Erica 'does the work of 11,000 people' | Investor-day disclosure: 20 million regular users, 2 million interactions a day. | [12] |
| Nov 5, 2025 | Investor day: 270 models, fraud losses halved, 18,000 coders on agents | Moynihan says the bank runs 270 AI and machine-learning models; the CTIO reports 18,000 developers on coding agents and a 20% productivity gain; Holly O'Neill says AI-enabled processes cut fraud losses in half since 2018. Technology spend is about $100 billion over a decade, $4 billion a year on new initiatives and $1 billion on cybersecurity. | [25][26][27] |
| Oct 28, 2025 | Datos Insights names CashPro Data Intelligence best AI innovation | The suite of CashPro Search, Forecasting and Insights wins the 2025 Datos Impact Award for AI; Merrill SENSE, a large-language-model search engine for marketing teams, wins a second award. | [10] |
| Oct 27, 2025 | CashPro Chat usage at a record | Client use of the AI-driven chat up 21% year on year, with nearly 70% of corporate clients using it; 43% containment rate in Q3 2025. | [15] |
Does Bank of America build its own AI models?
Yes, for its core assistant and much of its analytics. Erica was built in-house on open-source language models and trained on client interactions, rebuilt in 2025 as a second-generation, model-agnostic platform, and the bank ran more than 270 AI and machine-learning models in production in December 2025. For harder reasoning, its Orchestra layer calls proprietary third-party models whose suppliers it has not named.
Is Erica built in-house, and does it use generative AI?
Erica was built in-house in 2018 on natural-language processing and machine learning using open-source models. The Financial Brand reported in March 2025 that Erica 'does not use generative AI or large language models'. Generative AI was added first to EricaAssist, the contact-center version, in July 2026; a generative Erica 2.0 for clients is in development with no set timetable.
What is AskGPS?
AskGPS is Bank of America's internal generative AI assistant for Global Payments Solutions, introduced in 2025. It draws on thousands of vetted internal documents, supports nearly 3,000 employees and had answered more than 30,000 staff queries by December 2025. On September 28, 2026 the bank launched the AskGPS Intelligence Hub, adding treasury management reviews, account schematics and relationship insights built on client and account data.
Which patents are behind Erica?
The bank has not published a list, but its inventors have named some. 'System for Predictive Usage of Resources', obtained in 2020 by Katherine Dintenfass and Cameron Wadley, is partially used in Erica's Forecast my Balance, and Banking Dive reported in June 2026 that several of Wadley's patents are at Erica's foundation. Bank of America then held about 8,400 granted patents and applications, about 1,600 of them AI-focused.
Which outside AI models does Bank of America use?
Erica started on open-source models. In October 2026 the CTIO said the Orchestra layer sends simple classification to approved open-weight models on the bank's own GPUs and harder reasoning to proprietary models; the suppliers are not named. Developers use GitHub Copilot, and the bank has tested Anthropic's Mythos in the Glasswing program.
- BofA Augments Internal EricaAssist with GenAI, Eyes Erica Virtual Assistant Next
- Why Bank of America's Erica Virtual Assistant has a Human Touch
- No 'designated geniuses': How BofA democratizes innovation
- Bank of America's patent machine keeps rolling
- Inside Bank of America's 'build once' AI strategy
- Bank of America Enhances EricaAssist with Generative AI
- Bank of America Expands "Ask Global Payments Solutions" with New Intelligence Capabilities
- What Banks Can Learn from BofA's Multi-Billion Dollar AI Bet
- AI in regulated industries: BofA and S&P on the huge gains to be had, and the risks of rushing in
- BofA AI and Digital Capabilities Win Four Best Innovation Awards
- How Bank of America's Erica raised the stakes for virtual assistants
- How Bank of America's Erica does the work of 11,000 people
- Bank of America's Moynihan touts 'good economics' of AI
- BofA's AI Solution CashPro Forecasting Helps Clients Navigate Year of Volatility
- BofA's AI-Driven CashPro Chat and Transaction Search Capabilities Reach Record Usage by Corporate Clients
- Bank of America Reports Second Quarter 2026 Financial Results (Form 8-K, Exhibit 99.1)
- BofA Debuts Intelligence Tools for CashPro Clients and GPS Staff
- Agentic AI shopping bots are coming. Banks need to be ready
- Bank of America fires up generative AI payments assistant
- BofA Unveils AI-Driven Capital Markets Insights on CashPro App
- Bank of America upgrades customer service employee tool
- How bottom-line-focused financial firms are finding ROI in AI
- JPMorganChase still tops AI ranking as big-bank lead widens
- BofA AI and Digital Innovations Fuel 30 Billion Client Interactions
- BofA's Moynihan pledges growth, 'no excuses'
- BofA co-presidents pledge rigor on investor day targets
- Bank of America runs 270 AI models across operations
When BofA moves on AI, you'll read it here first.
when one of the banks we cover moves on AI, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning