← Cases
United StatesFinance and insurance2018 – ongoing

Bank of America

Erica virtual financial assistant

Bank of America's in-app assistant Erica passed 3 billion client interactions in seven years

WorkingCompany's own claimOfficial ↗

Problem

Bank of America's retail clients had many routine questions and tasks, such as checking balances, spending and upcoming payments. The bank wanted clients to handle them inside its mobile app without waiting for a person.

Approach

In 2018 the bank added Erica, a virtual assistant built in-house, to its mobile app. It answers questions, carries out tasks and sends information such as spending trends and payment reminders. Erica relies mainly on deterministic natural language processing; the bank has made more than 75,000 updates since launch and grew its coverage from about 200–250 question types to more than 700.

Results

3B+cumulative client interactions since the 2018 launch (August 2025)Company's own claim[1][2]
~50Musers helped since launchCompany's own claim[1][2]
58M+average client interactions per monthCompany's own claim[1]
98%+users who find the information they need, according to the bankCompany's own claim[1][2]

Takeaway

The bank grew Erica through steady improvements over seven years and made it central to the app. Usage and success figures are the bank's own.

Categories

Business function
Customer service
Evidence
Company's own claim

Sources

Source summaries

[1]Bank of America20 Aug 2025

The bank reports that Erica passed 3 billion client interactions, has helped nearly 50 million users since 2018 and averages more than 58 million interactions a month. It says more than 98% of users find what they need, that Erica has delivered over 1.7 billion proactive notifications and that it has been updated over 75,000 times. It also reports that over 90% of staff use an employee version.

[2]Banking Dive25 Aug 2025

Banking Dive interviews the bank's head of digital platforms and a Forrester analyst about Erica's 3 billion client interactions. It reports that Erica handled 200–250 question types at launch and more than 700 now, that about half or more of interactions start with a suggestion from Erica, and that it relies mainly on deterministic language processing, with generative AI tested only internally. It also cites J.D. Power and Forrester rankings of the bank's app.

Full details are in the original sources.

Added 25 Oct 2026 · Updated 25 Oct 2026
Open case page ↗