CIBC is rolling out its CIBC AI platform for employees starting this month.
It will be available across the bank, with adoption expected to take place over several months across different business units, CIBC’s executive vice-president of infrastructure, architecture and modernization, David Gillespie, said in an interview.
Banks have used machine learning since the 1990s and the public launch of ChatGPT in 2022 prompted increasing use of natural language AI. CIBC built its Knowledge Central chatbot last year to help frontline staff and contact centre agents find information in internal policies and procedures to process client requests.
“That platform was then extended for our CIBC AI, so we’ve reused that platform, which means the next build isn’t super expensive,” Gillespie said.
One of CIBC AI’s applications is reading unstructured documents. For example, a senior analyst may spend around 15 hours reading 20 documents and writing an analytical document about a client, Gillespie said. Current AI prototypes can do it in minutes.
Several thousand employees are already using CIBC AI through a pilot program and all employees must take AI training to have access to the technology, Gillespie said. Every executive at the bank has already completed a one-day AI training program in 2024.
The training helps employees understand what it means to have AI to support their day-to-day roles, find opportunities to apply the technology and implement it safely.
To mitigate risk, CIBC reviews every AI use case. Part of the bank’s AI governance is ensuring that AI outputs are appropriate to limit hallucinations.
For example, when AI processes a document, it assesses the quality of its own output, looking at the confidence interval or the likelihood that each parameter is true. If it fails any of the confidence intervals, the system hands it over to a human. Even if it passes, the AI is still subject to the same quality controls as work by humans.
“The business controls infrastructure wrapping around that is very important for things that are happening [in] real time. A human has to assess and review and apply judgment,” Gillespie said. “If you’re having to write an email to send to someone, a human needs to review that before they click send.”
CIBC has already had its first regulator “look at” their AI platform to see how the bank deployed AI and how it assessed use cases. The regulator is not Canadian, but Gillespie won’t say which country it’s from.
“Regulators continue to write new requirements and it’s a constant journey to stay aligned with them,” he said.
Singapore firm releases new underwriting platform
Singapore-based insurance technology provider Neutrinos launched its new “underwriting autonomous suite” in March, the company announced.
The underwriting platform can extract data from documents and identify missing fields to reduce errors that delay underwriting. It also uses AI to score risk and support underwriting decisions.
Neutrinos claims the platform can cut underwriting costs by up to half and speed up policy issuance.
Its website lists insurers Manulife Vietnam and Oklahoma City-based Woligo as customers.