Will Apple’s and Google’s on-device AI be useful in banking?

This week, Apple announced that it’s partnered with OpenAI and will incorporate generative AI in its next generation of smartphones. Google made a similar announcement last month that it’s embedding its Gemini generative AI into its Android smartphone operating system.

It’s early days and it’s not known whether Apple and Google will open this technology up to be customized or adapted by others. Apple and Google did not respond to requests for comment. But experts say the new technology could provide opportunities for banks, fintechs and their customers.

“I think we’re going to be surprised at how fast our mobile phones will morph from a pocket computer to an intelligent assistant,” said Alenka Grealish, principal analyst at Celent. “The implications for banking are profound because generative AI is clearly driving us toward a new wave of [user-interface-less] banking.”

“Apple and Google are clearly in a race to see who can put the most ‘smart’ in our smartphones, knowing that more AI features will keep consumers and app developers happy,” said Jim Perry, senior strategist at Market Insights. “These advances are inviting us to reimagine how we use our phones every day.”

Apple is adding generative AI capabilities to iOS 18, iPadOS 18 and MacOS Sequoia. It’s upgrading its voice assistant, Siri, to communicate over text messages. When Siri is stumped by a question, it will ask the user if they want to pose the question to ChatGPT. Apple is also adding summarization, rewriting and proofreading to iPhone-native apps like Notes, Mail and Pages. Within the Notes and Phone apps, users will be able to record a conversation and get a transcription and an AI summary of the transcription.

“It’s long overdue for Apple to be part of the generative AI game,” said Zor Gorelov, CEO of Kasisto. “I think Siri needed to be improved,” he said.

Google’s Gemini on Android will also be a generative-AI-driven assistant. It will let users drag and drop generated images into Gmail, Google Messages and other places, or tap “Ask this video” to find specific information in a YouTube video. Gemini Advanced will let people use “Ask this PDF” to quickly get answers without having to actually read a document. Gemini Nano, Google’s name for its on-device AI, will work with TalkBack to help people with blindness or low vision get clearer descriptions of what’s happening in an image. Nano will also try to identify scams.

Fraud detection

Google is currently testing Nano’s ability to provide real-time alerts during a call if it detects conversation patterns commonly associated with scams. 

“For example, you would receive an alert if a ‘bank representative’ asks you to urgently transfer funds, make a payment with a gift card or requests personal information like card PINs or passwords, which are uncommon bank requests,” the company stated in a blog. “This protection all happens on-device, so your conversation stays private to you.”

This feature could help banks without any effort on their part, if consumers opt in and if it works well.

“Whether or not people will want AI to be listening to their phone calls to detect whether there’s a spam call they don’t want to take, I don’t know,” Gorelov observed. “If they can detect fraud and eliminate spam, I welcome that. But if they use that to better target ads, that would be really, really annoying.”

Google and Apple both said in their announcements that they’re building data privacy into their on-device AI.

Personalization

As mobile-based intelligent assistants get better at understanding consumers’ prompts and crafting responses, both banks and consumers will become comfortable using them in new ways, Grealish said.

“It’s totally realistic to envision soon the ability to talk to your mobile wallet and ask it to pick the best card for, for example, a car rental based on reward points earned and liability insurance coverage,” Grealish said. In a related vein, generative AI is bringing about improvements in language translation, which opens up mobile banking to support customers in almost any language.

Futurist and Breaking Banks host Brett King has long envisioned a personalized AI assistant that would work across many devices. He first wrote about this concept in his 2015 book, “Augmented.” This is a few years out still.

“For the meantime, you are going to have to use the Apple AI system or the Google AI system or the Amazon AI system,” he said in an interview Wednesday. 

The biggest likely impact for banks of AI getting built into smartphones is that this could help  mobile wallet providers become more competitive, King said. Worldpay’s Global Payments Report 2024 estimated that by 2027, 61% of all payments will be made through mobile wallets. In 2023, 50% of payments were made this way.

“If you want to further take business away from banks, how will you do it?” King said, speaking of fintech mobile wallet providers. “You’ll integrate artificial intelligence into the wallet architecture, you’ll provide an AI-based money management capability, and you’ll start a learning engine that allows you to do embedded finance and credit and savings activity smarter because you’ve got this AI money coach that is trained on your behavior and your financial needs.” 

But there’s also opportunity for traditional banks in this scenario, King said. 

“If you’ve got a bank account that’s abstracted by an AI, banks become the dumb pipes,” he said. “But there’s another element to this: when you are asking an AI as your agent to do financial activities, do you really trust that the AI is reliable, it’s not going to have a hallucination in the middle of your bank account transfer? So integration of bank architecture into this AI-based cloud stack is one way that these players will be able to guarantee trustworthiness of their platforms.” 

To take advantage of this opportunity, banks will have to integrate into these third party wallet systems. 

Small language models

The on-device AI upgrades from Apple and Google are part of a trend toward using small language models, Gorelov said. Small language models use fewer parameters than large language models.

“These models are smaller, they’re optimized for certain use cases, and you can put them on a device,” Gorelov said. “They’re easy to train, easier to operate and enable some very powerful use cases.” For instance, they can be useful in email and image generation as well as personalization, he said.

“Overall bringing generative AI to devices is very promising from a privacy perspective and from a use case perspective,” Gorelov said. “And small language models are powerful enough to enable some really powerful use cases.”

The new technology will not be available to everyone, however. Consumers use many different versions of iPhones and Android devices with different levels of capability. 

“It seems we’re rapidly approaching a time when the digital divide is going to be measured by the version of the phone we carry and whether or not they have the capacity to handle the new AI functionality,” Perry said. “So for banks, the usefulness of these new tools will depend, in part, on the compatibility of the devices in use. The more that Apple and Google bring AI enhancements to their phones within the limits of their existing hardware, the more likely they’ll get the majority of consumers to adapt their behaviors.”