The Banking & Payments Show: Will banks provide good chatbots?

On today’s episode, we discuss ChatGPT, generative AI, and AI’s role in banking. In our “Headlines” segment, we examine if 2023 is really going to be the year of the chatbot in banking. In “Story by Numbers,” we reconcile two sets of data to find out how valuable chatbots really are and forecast how many people will consider using a bank chatbot this year. And in “For Argument’s Sake,” we debate new ideas that you may not have thought of when it comes to AI in banking and generative AI. Tune in to the conversation between our host Rob Rubin, analyst Eleni Digalaki, and Victor Chatenay, strategy and innovation manager at NatWest.

Subscribe to “The Banking & Payments Show” podcast on Apple Podcasts, Spotify, Podbean or wherever you listen to podcasts. Follow us on Instagram

Episode Transcript:

Rob Rubin:

Hello and welcome to the Banking and Payment Show, A Behind the Numbers podcast from eMarketer. Today is February 21st. I'm Rob Rubin, GM of Financial Services here at Insider Intelligence, and your host. If you enjoy this podcast, please give us the five-star rating and subscribe.

The title of today's episode is, "Will Banks Provide Good Chatbots?" and I invited Head of Financial Services Research at Insider Intelligence, Eleni Digalaki, and Victor Chatenay, Strategy and Innovation Manager at NatWest and an Insider Intelligence alum, to join me.

Hi guys, how are you doing today?

Eleni Digalaki:

Hi Rob, thanks for having us.

Victor Chatenay:

Hi Rob. Doing great over here in sunny London.

Rob Rubin:

Yeah, well, welcome back. I'm glad to have you here.

Victor Chatenay:

Oh, happy to be here.

Rob Rubin:

We have a lot to talk about so I really want to get right into the headlines.

The two headlines today both draw similar conclusions. The first article's from my friend Ron Shevlin from Forbes. Thanks to ChatGPT, 2023 is the year of the chatbot in banking. In an article that we published on January 23rd, ChatGPT aims to bring AI to the next level in banking. Both articles point to the benefits of the technology, the ability to understand with context what customers are asking, filling support gaps to free up time-starved service teams, and increasing employee productivity.

Eleni, I'm going to start with you. What are we going to see in terms of generative AI applications from banks in 2023?

Eleni Digalaki:

Yeah, Rob, that's a good question and I think there's a lot of focus on chatbots right now because of ChatGPT, but there are also a lot of concerns around accuracy and unlike conversational AI, generative AI sort of makes things up and does that very convincingly. So it can tell you things that are not accurate essentially, and probably today, banks shouldn't use it for direct customer communication primarily because of this reason.

Rob Rubin:

So they're not going to use it. Are we going to see them use anything this year?

Eleni Digalaki:

I think generative AI, yes, and there are many other use cases that are mature enough to be implemented today. So for example, when it comes to customer service, I see it more as an internal chatbot, so helping customer service assistance become more efficient, going through general databases and finding information for them, sourcing answers for FAQs or information about products and services, for example. Technology can do that in seconds, so it's extremely fast. So it would save assistance a lot of time and also help them sort of focus on other more complex tasks.

Rob Rubin:

You know, Victor, Ron Shevlin in the Forbes article said that 2023 is going to be the year of the chatbot in banking, and I'm pretty sure Eleni is saying maybe not 2023. So, I'm wondering what you think about that.

Victor Chatenay:

I think I agree that I don't think banks are falling over each other trying to implement this technology as fast as possible. I think it's more of a cautious stroll down the lane rather than kind of racing to the finish line.

I think these headlines are obviously very exciting, but there's a key step I feel the media and coverage of ChatGPT is not addressing here, and it's that transition between people just going on ChatGPT for free to play around with and actually having an AI-powered chatbot offered by your bank. There's quite a big step that's missing here. And in particular, I think we were talking about this earlier on Rob, there's specific tech and regulatory challenges with a bank being able to implement this kind of AI capability.

Rob Rubin:

What are some of the tech challenges to this?

Victor Chatenay:

Yeah. So something that I think people forget is obviously behind every AI algorithm, to train them, you need a lot of data. That's what will make sure that it's successful and smart.

Rob Rubin:

Well, banks have tons of data.

Victor Chatenay:

Banks do have a treasure trove of data, but we suck at managing it, right? That's kind of an open secret.

Rob Rubin:

Really?

Victor Chatenay:

You can see that with how banks struggle, for example, with building a smooth customer experience across multiple channels or across business units. If we were actually good at using customer data, I'd feel richer than the Googles and Amazons of this world, but we don't live in that world and I think even, no offense to the banking sector out there, but if we were to look at the slight improvements we've done in data sharing over the past few years, at least in the UK and Europe, it's because of open banking, which the Regulator has essentially forced us to get better at sharing data.

So we're just not there, and I think if we take the example of ChatGPT, we would need to feed OpenAI with our very sensitive customer data to train ChatGPT to be able to address our pain points. And that's going to be a very slow and painful process.

I just saw on the news, obviously Microsoft's Bing is going to be integrating this chatbot and that makes perfect sense because Microsoft has the tech capabilities, it knows the power of data and how to use it, but when OpenAI or the AI startups are going to approach financial institutions, training their AI to understand bank customers, that's going to be a lot harder because we just don't have the tech capabilities, and I think it's going to take a lot of handholding from the startups to the banks.

Rob Rubin:

It's going to be a scenario where a bank comes out with a great version of a chatbot and it becomes sort of a hit where then the banks are going to be all sort of tripping over themselves to replicate it.

Victor Chatenay:

I feel like if that was the case, we would've already seen that. I mean, for example, Bank of America's Erica is something that they're very proud of and kind of shows what you can do with a chatbot, but it's not so much putting this in the context of yes, artificial intelligence and chatbots are very important and banks are looking at it, but these headlines kind of suggest that we've reached a new threshold in AI development and we're going to see, over the next couple of years, a lot of changes in companies' AI in a way that hasn't been done before and it could be a lot more impactful. And I'm just of the opinion that as far as the banking sector is concerned, we're just not there yet. And we are first going to see that adoption happen at the big tech companies and startups, perhaps even of industries such as e-commerce.

But as far as financial institutions are concerned, we're way behind, as we always are to be honest.

Eleni Digalaki:

I think that we're sort of limiting ourselves a little bit by focusing on the chatbots alone because there are other use cases like generating content for marketing campaigns that it can be used today by bankers quite efficiently and save a lot of time.

But also, going back to what Victor mentioned about data, and that's a very good point, the advancements in generative AI also mean advancements in something called synthetic data generation. And that is the idea that you can take real data and sort of augment it and create bigger and richer versions of original datasets, and this can really help with things like product development and sort of creating synthetic personas of customers or understanding your audience better. And so with the applications there, it could also be very meaningful. And since I touched on product development, it can also help with things like coding and debugging, so it can help those teams also become more efficient.

And just to add as a final note, that despite the many limitations it has today, let's say around accuracy or fairness, the transformation will come more suddenly than most people realize I think, so that tech will develop much faster than other technology cycles like the smartphone for example. Already if you look at what we expect from ChatGPT-4 compared to ChatGPT-3, the improvement in the advancement is just mind-blowing in terms of how many more parameters it has, how much more accurate it will be, and how much more quick.

Victor Chatenay:

I guess my point on that is just about what we mean by transformational and where banks will be comfortable applying these AI capabilities. So things like marketing campaigns and product development, while impacted by this, is that transformational? I mean, compared to, say, automating customer interaction, helping with credit risk and liquidity management? These are the hugely manual cost-intensive processes that banks have, and because of the black box issue, and as you say, the AIs can often create information, we're just not there in terms of allowing them to drive efficiencies there.

So that's why I think it is limited to things like marketing campaigns, which all in all, is not exactly transformational in my opinion.

Rob Rubin:

Right. I think I agree and I think it's clear that generative AI is going to have a tremendous impact on banks, but it seems like banks' approaches are going to be more measured and perhaps the chatbot app is going to be the one that might be most visible, but it might not be the one that is the first to be the most successful.

I want to dig into it a little bit more in our "Story by Numbers" segment.

And in "Story by Numbers," I've selected two sets of numbers for us to spend time reconciling, and the first set of numbers are, and remember this, it's 13 and 15%, so let's just say 14%, and this is the percentage of UK and US mobile bank customers that have found chatbots extremely valuable. So, not very much.

The second set of numbers is 38% and 39%, we'll just say 38%. This is what Insider Intelligence is forecasting for the percentage of UK and US populations that will use a bank chatbot in 2023, which to me, is quite a lot. And I know that the bases for these numbers are very different, but the story that few consumers find chatbots extremely valuable and yet a pretty high percentage of the population is going to use one this year is what I want to explore because it seems like a decent percentage of the population's going to be using bank chatbots, but not many customers find them valuable.

Victor Chatenay:

Yeah. I can tell you, Rob, that this sort data makes my job harder. So essentially, you have the bank executives reading these headlines and thinking, "Oh, ChatGPT sounds really impressive. What are we doing about it?" And innovation teams at banks like myself, "How are we going to implement this? Where is the value? "

But that data you just shared, there's obviously a discrepancy between 13% of customers finding the chatbots extremely valuable, but then we're supposed to say that we expect the market and adoption to grow. That's a challenging story to tell to our leadership. So innovation is a cost for banks and if I'm pitching the bank for budget to explore chatbot use cases, I think what the challenge is really to articulate what the benefits are and if I can't make a valid case that the customer demand is there, that's going to be really challenges, especially with high interest rates and inflation and whatever the banks are having to contend with. It's more difficult than ever in the way to kind of go for these innovation ideas that haven't been validated.

Rob Rubin:

I think that the thing that consumers always found valuable with a bank is convenience. In other words, if you provide a service that makes their life more convenient, they like it. If it's not, then they don't like it as much. So it's the convenience of the chatbots. As they become better and faster, I think that consumers are going to like them more. Isn't that the argument you make?

Eleni Digalaki:

I think in terms of the discrepancy between these two numbers, for one, the number that says 13% find it valuable or whatever the number was exactly, doesn't say why they don't find it viable. So it doesn't say, "Is it a good experience but I don't need it," or, "Is it a subpar experience and that's why I don't find it viable?" And I think it's the latter when it comes to most chatbots today.

So now this will change with ChatGPT because it'll set a new bar, for example, around what a conversational response looks like or what a speedy response looks like. But that doesn't necessarily mean that banks should rush to implement that. It just means that the expectations and the demands are elevated and they need to acknowledge that.

When it comes to how many people use it, I'll just add because I think it's useful to know that sometimes you have to use the chatbot, it's not a choice. So it doesn't mean you want to use it, it's just that you have to go through a chatbot to get to a human.

Rob Rubin:

A human, right.

Eleni Digalaki:

Yeah, exactly.

Victor Chatenay:

I completely agree with those nuances, but I think from a bank perspective, it will likely be the case and I think you probably alluded to that earlier, that the technology will be more readily adopted by other industries. Maybe you'll start seeing those chatbots appear more on e-commerce platforms and social media. And when customers start getting used to using those there and see the value, they're going to start asking themselves, "Well, it's making it a lot easier for me to shop online. Why isn't this helping me with my banking services?"

I like to compare a bit to smartphone technology where banks weren't the first app to go on your iPhone, but you were first able to chat with your friends and share pictures and watch YouTube videos on there. And after a certain while, people started saying, "Well, I'd also like to be able to check my bank account and send money from my phone," and pretty quickly that became table stakes.

So I think that's going to happen here where we kind of kick up a backside from other industries that's going to prove that customers want this and start asking these questions. And then adoption will probably happen very quickly at that point.

Rob Rubin:

Where's the line? A chatbot can tell you where an ATM is that's closest to you, but should a chatbot be telling you how to invest retirement money?

Victor Chatenay:

I'd be really curious to see what the SEC would say if you ask them that question. But yeah, so I think from the bank perspective, it is about helping offer more personalized services to your customers and they can understand your product.

Rob Rubin:

But isn't that advice-based, right?

Victor Chatenay:

Right.

Rob Rubin:

That's the line I was trying to draw there is that it's one thing giving you information, but are you building an AI tool to actually give advice?

Victor Chatenay:

Yeah, and I think that seems to me like it can be a real big challenge for heavily regulated financial institutions, whereas that seems to make more sense for a search engine like Google or Bing where you do go for "advice." For example, you could say something like, "Help me shop for the best interest rate on savings accounts on the market," and then the chatbot for your search engine could help you do that. That seems more palpable than asking a bank to do that, which obviously there's a conflict of interest, but also Regulation will have something to say about that.

So yeah, I mean, that's a really good point and certainly would be a challenge in applying this for bank.

Eleni Digalaki:

Yeah, I think going back to the problem of accuracy right now is important as well. It shouldn't give advice today just because it might be wrong, if anything, but I think when it gets better and more accurate, the idea of personalization might be possible in ways that it wasn't before and the vast data that it has and the speed of response and how conversational it is and entertaining as well, and even funny, it might allow it to do it in a way that's not perceived as creepy from customers and it's sort of more natural and feels okay.

But it's a fine line. I agree, it's a very fine line. So there are a lot of issues there around controls, governance and risk management that need to be ironed out as well.

Rob Rubin:

Right.

Victor Chatenay:

Because I think also a truly intelligent AI chatbot, if you said like, "Okay, have a look at my finances and tell me what I should do," it might just say, "Well, you shouldn't be using this bank, you should be doing this," or, "You should be looking at all these different providers, which is obviously not necessarily something... A bank's own chatbot is not going to want to tell you to do that.

Rob Rubin:

Why do you have all that money in a savings account with earning no interest?

Victor Chatenay:

Yeah.

Rob Rubin:

I think that's going to be interesting to see how banks evolve chatbots, not only to be more conversational, but also how they'll utilize customer data to form their responses.

Let's carry that thought into our final segment, "For Argument's Sake." But before we get there, let's take a moment to hear about our Virtual Summit, which is happening Friday, March 3rd.

Join us on Friday, March 3rd for our next Virtual Summit, Attention! Seizing the Retail Media Opportunity. The half-day virtual event kicks off with an exciting keynote from our Principal Analyst, Andrew Lipsman, on Retail Media's growth, featuring a special guest from Walmart Connect. The summit continues with deep discussions on topics like connected TV advertising, social advertising and first-party data featuring Insider Intelligence analysts and guests from companies like M&T Bank, Ghirardelli Chocolate, e.l.f. Beauty, and more. It all wraps up with an interactive Ask the Analysts Panel and a live production of Behind the Numbers: Reimagining Retail.

Visit insiderintelligence.com/events and register today. The link is in the show notes.

In our previous segment we discussed that consumer's current experience with bank chatbots hasn't been great and that banks are really going to have to try and figure out the right mix of personalization and information to make it worthwhile. In this segment, "For Argument's Sake," I want to discuss Google, Microsoft and ChatGPT, which we've been discussing, but Microsoft announced a new generative AI search tool using ChatGPT. Google acquired Anthropic recently for $300 million, and it seems like the next generation search wars are on. And at the same time, we published this really cool chart that demonstrated how quickly consumers are actually signing up for ChatGPT. It took only five days for a million downloads, which was twice as fast just about as TikTok. We're going to post a link to that chart in our notes.

And the reason that I'm bringing this up, and I want to sort of change it a little bit, is it feels like the crypto gold rush has dried up now that we're in this crypto ice age and now everybody is swinging to generative AI, and as a result, consumers expectations are going to be through the roof, investors are going to flood the zone, so we're going to see tons of fintech startups in this area. And given that consumers have been pretty underwhelmed with chatbots so far, does all of this help or hinder banks?

Eleni Digalaki:

So the examples you gave for Google and Microsoft I think are worth sticking with for a bit. So Google is leader when it comes to generative AI, but they been hesitant to productize it because of all the issues around the technology, which is what allowed OpenAI and Microsoft to sort of get a big headstart.

This then sort of forced Google to make a move and we saw what happened when they did. So Alphabet lost a $100 billion in market value a couple of weeks ago.

Rob Rubin:

$100 billion.

Eleni Digalaki:

Yeah, after its new chatbot shared inaccurate information in a promotional video.

So the reason I'm bringing this up is that this highlights the stakes for banks if they rush to use the technology in ways that it's not yet equipped to do well or that it should be used, like offering advice to customers that we've been talking about. But at the same time, as we discussed earlier, there are meaningful use cases for banks to explore today, and companies that move fast and do it well seem to be rewarded by the market. And as we see in other industries, for example, Buzzfeed's stock jumped like 150% when it announced it would use generative AI to produce content.

Rob Rubin:

Instead of people?

Eleni Digalaki:

Yeah, or with people, making people more efficient and so on and so on. So I think the banks need to figure out what are some meaningful applications of generative AI for them and pursue them rather than jump to deliver on something like a customer-facing chatbot just because consumer expectations are rising, but they certainly should not stay inactive I think because they really risk staying behind. This is an innovation cycle that's moving fast.

Rob Rubin:

Isn't there innovation that could have more of an impact on the banks around using AI in terms of risk, using AI in terms of managing liquidity? Aren't those actually levers that have more impact on a bank's bottom line?

Victor Chatenay:

Yeah, I mean, I agree. To Eleni's point about Buzzfeed announcing they're going to use AI to write articles and its share price went up, I mean, imagine what would happen to a bank share price if they said, "We're going to manage our credit risk and liquidity with an AI chatbot." It will crash tomorrow, right? So that's obviously a big difference between the two.

But I kind of wanted to come back to the crypto point because I think what was interesting about that one is I think it really showed that there's an indirect relationship between VC funding and bank adoption because yes, all that capital from VCs is redirecting away from crypto into AI, and yet not a week goes by where I don't see on the news that the bank is partnering with blockchain software providers to offer crypto custody or tokenization or things. So it seems that this year is really going to be the year where we see institutional adoption of crypto and sort of "the adults" are coming in, and I think it's kind of similar to AI where-

Rob Rubin:

The adults.

Victor Chatenay:

Well, you know VCs, they're looking for that moonshot investment. It's a completely different focus than the banks. So I don't think banks are going to pay much attention to how much money is raised by these AI startups.

Rob Rubin:

It's the opposite, right? A VC can lose nine out of 10 times as long as they win one big one, whereas a bank has to win nine out of 10 times and that last one can't be too terrible.

Victor Chatenay:

Yeah, exactly. But what will happen with that VC funding is it might help with that sort of consumer demand and expectations we were talking about before. So if all of your startups are getting a lot of funding and they've got this capital and they start growing, they start marketing, a bit like crypto startups where you saw this rise in retail crypto adoption, you might see a rise in chatbots appearing in your everyday life as you use services. And once consumers start getting used to those, they will start again saying, "Why can't my bank do this?"

So I think that's where, again, we need to see consumer expectations shift before banks are more willing to pull money into this.

Rob Rubin:

And consumer trust too, right? Don't you think that there's going to be a trust element in sort of believing that the chat is accurate? As stories come around that Microsoft loses $100 billion or Google loses $100 billion in a day, it's going to make consumers worry that the stories about it not being accurate, stories about students submitting papers that looked really perfect except that the book didn't exist nor did the authors.

Eleni Digalaki:

Yeah, I think the trust issue is very important. And again, I think it goes back to figuring out the right use cases, and I don't fully agree with Victor that the market wouldn't reward banks for implementing AI, generative AI specifically. It just would need to be done well, cautiously, and for the right use case, and with a lot of focus around governance and risk management. And again, I don't think a consumer-facing chatbot, yeah, that would definitely tank the value of a bank today.

Victor Chatenay:

Yeah, I mean, it's just that in reading past the headlines and looking at how banks right now are actually applying this technology, it's just not that game-changing. Bank of America had a record number of patents in 2022, 608, a 19% rise on the year prior, and so I kind of started digging deeper into this. Okay, so what are the patterns?

Rob Rubin:

Yeah, what are they?

Victor Chatenay:

And what came up, or at least what they were willing to share, was customization of client interfaces within multiple banking channels. I mean, it's cute, right? It's nice to have, but it's not exactly this transformative, groundbreaking leap forward that we're led to believe ChatGPT is bringing.

So again, because of that, I think the trust issue, because of the regulation and just the legacy tech that banks are dealing with, when you look past the headline, the use cases that they're willing to go with first, it's the marketing campaigns, the client interfaces, et cetera, which doesn't go to the core where automation could drive value.

Eleni Digalaki:

I think that very often transformation happens slow and then it happens very fast. That, I think, is what I'm getting at, that banks should explore it cautiously and with the right frameworks to do so safely, but they shouldn't ignore it just because they feel today that it's not that transformational.

Rob Rubin:

Just to sum up, I think we all agree that 2023 is probably not the year of the bank chatbot. Sorry to my friend Ron on that. But I think that we see that banks should take a cautious approach moving forward to how to leverage this. And I like, Eleni, what you just said. It's transformation is slow and then it goes fast. And the challenge for banks is if you're not in it right now thinking about it, then you might be behind the eight-ball because it's a slow roll and then it's fast and you haven't done the slow stuff yet. So, I think it's time for banks to be paying a lot of attention.

Well, that's all we have for today. Thanks Eleni. Thanks, Victor. I really appreciate it. This was great.

Eleni Digalaki:

Thank you for having us, Rob. Nice to catch up again, Victor.

Victor Chatenay:

Thanks, Rob. Yeah, this is great. Thank you.

Rob Rubin:

Yeah. It was a lot of fun. I want to also thank everyone for listening to the Banking and Payment Show, an eMarketer podcast. Also, thank you to our editor, Todd.

In today's episode, we reference data from our "2022 Emerging Mobile Banking Features Benchmarks in the US and UK," as well as a recent article we published on ChatGPT in banking and the Forbes article that we talked about. Links to the reports and the articles are in the episode notes.

Our next episode is on March 8th and you'll not want to miss it. See you then.