RBC partners with Cohere to develop AI platform for bank employees

Royal Bank of Canada RY-T and artificial-intelligence company Cohere Inc. are building a bank-wide generative AI platform for financial services, a step that the country’s largest lender says is the first for a Canadian bank.

Canada’s biggest banks have increasingly been adopting AI to help employees do their jobs more efficiently. But generative AI has largely been restricted to more niche operations or testing programs.

On Thursday, RBC and Toronto-based Cohere launched an exclusive partnership to co-develop and deploy across the bank’s operations a customized version of the company’s new generative AI platform.

The platform, called North for Banking, will help employees complete tasks and find information, including searching for answers and solutions specific to customers’ needs. Generative AI allows employees to submit the context of the individual’s or the client’s circumstances in the assessment, which enables the platform to solve problems and provide advice.

“Where there’s more value is when you can make it more relevant to the workflow of our own employees who manage RBC sensitive data for our clients,” said Dr. Foteini Agrafioti, RBC Borealis senior vice-president and chief science officer, in an interview.

“Up to now, we haven’t felt comfortable bringing RBC sensitive data closer to these technologies until this point. The commitment that we’re making is to work together to deploy North for Banking in a secure way so we can access RBC data through generative AI.”

RBC has been using generative AI for more niche applications internally to test the technology. The bank is using generative AI applications in its customer service centres to help staff answer client questions faster and with more detail, as well as its capital markets unit to assist research analysts.

In October, the bank was ranked in the top three in the world for artificial intelligence maturity among 50 global financial institutions in the Evident AI Index. Canada’s five largest lenders were on the list, with Toronto-Dominion Bank ranking in ninth place.

Cohere, founded in 2019, builds large language models (LLMs) that power chatbots and generative AI platforms, which analyze text and media to create responses. Businesses can tailor Cohere’s North for Banking to their own needs, such as assessing procedure and policy documents, answering questions and automating processes.

The startup has attracted a wave of funding over the past year. The federal government committed $240-million for Cohere in December to build a new multibillion-dollar data centre in Canada that to power AI models.

In July, Cohere raised US$500-million to better compete with OpenAI, Anthropic, Google and other major competitors. The financing round valued the startup at US$5.5-billion.

Some of the largest banks globally have also been experimenting with generative AI. Last year, U.S.-based JPMorgan Chase started rolling out its own version of OpenAI’s ChatGPT. BBVA – Spain’s second largest bank – became one of OpenAI’s biggest financial services customers.

But regulators have expressed concerns over potential privacy issues. In 2023, Canada’s privacy commissioner launched an investigation into ChatGPT in response to a complaint claiming the collection, use and disclosure of personal information was done without consent. In December, Italy’s data protection agency said it fined OpenAI €15-million after investigating its use of personal data.

In the case of RBC’s work with Cohere, all the data will be powered and stored internally within the bank’s systems.

The process of building LLMs is costly and requires a significant amount of computing power from graphic processing units, or GPUs, which trains AI systems with large amounts of data. While Cohere has computing resources, RBC will be using its own system to run the internal platform.

In recent years, RBC has built one of the largest GPU farms among Canadian businesses, Ms. Agrafioti said.

GPUs allow for AI systems to be trained on very large amount of data efficiently. RBC has been investing in building out an enterprise GPU cluster over the past several years within its data centres.

Ms. Agrafioti said that much of the risk with data and privacy breaches stems from generative AI systems predominantly running on cloud platforms or shared services.

“Bringing all of that infrastructure onto our premises in our data centres is what addresses that risk off the bat in a definitive way,” she said. “It’s a very complex equation but you’re immediately removing that factor out of it, and all of a sudden that opens a ton of opportunity.”