CIBC is carefully weighing the financial implications of expanding its artificial intelligence tools across its workforce. The new version, dubbed CAI 2.0, is a major leap forward in capability compared to the original version of the platform, which costs about $3 per user each month. While the advanced features of the new tool are a game-changer, they also come with a much steeper price.
CAI 2.0 is built on agentic AI, a next-level technology that enables it to handle complex workflows. This includes everything from generating well-structured spreadsheets to building polished presentation decks and fully formatted business reports. The benefits are clear, but the trade-off is the rise in costs. Monthly expenses vary greatly, from as low as $4 to as high as $250, based on how intensely an employee uses the system. Jaime Tatis, who leads data and AI initiatives at the bank, pointed out that the power of CAI 2.0 makes it impractical to let all 50,000 staff members use it freely at once.
The high cost of expansion
The first iteration of CIBC’s AI solution, CAI 1.0, was developed in-house and currently sees daily use by around 20,000 employees. The bank does not log the exact questions users enter into the system, but it does keep a close eye on how often the tool is used and the skill levels of its users. This monitoring has allowed CIBC to identify roughly 1,000 internal AI experts, known as champions, who provide valuable feedback and help shape future improvements. The data collected on usage patterns, token expenses, and employee proficiency is being used to refine the rollout strategy for the more advanced CAI 2.0 version. These champions are also tasked with training others and ensuring the efficient and appropriate use of the platform.
Rolling out CAI 2.0 with a strategy in mind
The launch of CAI 2.0 started small, with just 50 users. Since then, it has gradually grown to include about 250 individuals from key departments like business banking, capital markets, legal, procurement, and risk. The bank’s executive leadership is also testing the new platform, with CEO Harry Culham among those involved. Unlike CAI 1.0, which did not have spending restrictions, the newer version will feature usage limits and charge-back systems for each business department. These measures are crucial for managing rising costs. The bank is also looking into open-source models, such as those provided by Cohere, to help reduce expenses and cut dependency on U.S.-based vendors like Anthropic. However, open-source models still require the company to handle security and liability, making them more complex to manage at scale.
Reducing regional dependencies and managing global risks
Tatis highlighted that relying on Canadian AI models could be a way for the bank to avoid being overly dependent on a single region. This concern has been amplified by global developments, such as recent U.S. government restrictions that limited access to Anthropic’s Claude Mythos 5 model in June. To mitigate vendor-related risks, CIBC is creating a tailored software framework that connects advanced AI models from companies like OpenAI and Anthropic with the bank’s internal data while maintaining strong security and access controls. The most costly models, such as Anthropic’s Opus, will be available only to a select group of 150 to 200 employees to ensure expenses stay manageable. Tatis believes that the development of open-source AI models in regions like Europe and China could offer the bank additional flexibility and backup options should geopolitical tensions impact access to U.S.-based models.