Bank of America's chatbot, Erica, answers 2M messages per day

Bank of America's chatbot, Erica, answers 2M messages per day
Screenshot from the BoA's Erica webpage

I'm working away on a keynote presentation for an upcoming event and I came across this press release from Bank of America ("BoA") I bookmarked back in April.

It's one of those summary releases that bank public relations teams tend to release to showcase digital credentials. I know how this works having been the Chief Digital Officer of Northern Europe's largest financial powerhouse. If you've got good metrics to share, and they're not commercially sensitive, then why not? It helps the industry, too – because public announcements like these are very useful benchmarks.

So let's take a look. It is fabulous reading for anyone in Conversational AI.

2 Billion Interactions

Yes, you read that right. Since 2018, BoA's virtual assistant, Erica, has had 2 billion interactions with customers. Interestingly, it took 4 years to reach 1 billion interactions (see press release) and only 18 months to reach the second billion.

Erica, by the way, should be written Erica®. It's a registered trademark – and this, dear reader, is superb ambition: Whilst many European companies are still messing around with their generic-named 'virtual assistants', the BoA team hasn't just named their bot, they've trademarked the name!

Depending on which global top ten bank list you're looking, BoA is usually to be found around the #5, #6 or #7 spot. They have a huge amount of retail customers: 69 million consumer and small business customers, according to their overview.

So they're understandably not messing around when it comes to automation. Whilst other much smaller players are taking a slightly different direction (see Frost Bank's "AI means Always Intelligent" human-first approach), bigger players such as BoA have to address how they manage customer interactions at huge scale, with sustainable cost models – which is why Conversational AI is such a compelling option.

The release reports that 42M of BoA's customers have used Erica since it began. In simplistic percentage terms then, 81% of BoA's retail and SME customers have used the chatbot since it launched.

Screenshot from the BoA press release

Proactive Insights

BoA's Erica isn't just about receiving incoming requests. Instead, it's also regularly offering customers proactive insights. This is would put them right up at Stage 5 (and potentially Stage 6) on my Conversational AI Maturity Index. Far too many organisations are still sitting and waiting for customers to speak to them, rather than proactively reaching out to offer insights and perspectives.

Here's an overview of some of the 30+ proactive insights Erica offers, along with some metrics:

  • Monitoring and managing recurring subscriptions – 2.6 million per month
  • Helping clients understand spending behaviors – 2.2 million per month
  • Keeping clients informed about deposits and refunds – 2.1 million per month

Some banks I know are doing a million chat messages a year. In total. So this gives you an idea of the scale of what BoA is doing with Erica.

Assistance

When customers do use the chatbot, here are some of the more popular use cases:

  • Requesting an account number or routing number – 1.7 million per month
  • Finding transactions – 1.5 million per month
  • Assistance with money transfers and bill pay – 900,000 per month
Screenshot from the BoA website showing screenshots

Containment

A popular metric for most Conversational AI installations is 'containment' – or a measure of how much you're able to 'contain' the customer's enquiry to automatic channels. You do this by (hopefully) answering the query fully, or by answering and then orchestrating a fix – for example, if the customer is requesting a replacement debit card, the chatbot should be able to understand this, check details and then connect to the relevant system to request the replacement – all without having to escalate to a human. That's the key metric.

BoA report that 98% of customers get the answers they need from the chatbot in around 44 seconds. I can't say whether this is a 'containment' metric as I don't have additional context - but it does sound very good.


Find out more at the BoA Erica website here.

Nice work team BoA!