Westpac tightens AI spending controls and moves routine work to cheaper models

Australian bank Westpac Banking Corp has tightened control over its artificial intelligence spending. According to Bloomberg, the company now tracks employees' AI token consumption across the organization and routes simple tasks to cheaper models.

What exactly Westpac is doing

The bank is rolling out stricter monitoring of how employees use generative AI models in their day-to-day work. That means tracking token consumption. Tokens are the units that measure how much text a language model processes, and they determine what each request costs. Westpac sends tasks that don't need the full power of flagship models to cheaper, lighter models, and reserves the expensive ones for complex use cases.

• Westpac Banking Corp is one of Australia's largest banks
• The bank tracks employees' AI token usage across the company
• Simple tasks move to cheaper models instead of flagship ones

Why companies have started counting tokens

The cost of using large language models grows with the scale of deployment. The more employees have access to AI tools, and the more often they use them, the bigger the bill from the model provider. At banks and other large corporations, where AI services have thousands of users, that spending becomes a noticeable budget line. This explains the growing interest in so-called FinOps for AI: the practice of spreading workloads across different models on a simple principle — complex tasks go to expensive models, routine ones to cheap models — rather than pushing every request through the most powerful and expensive system.

Westpac's approach is typical of large financial institutions. They want the benefits of generative AI while avoiding runaway spending on cloud AI services, especially in a regulated industry where every cost line faces extra scrutiny.

What this means

As generative AI becomes part of everyday work at large companies, managing model costs is turning into a discipline of its own. Businesses are learning not just to deploy AI, but to calculate what every request costs and optimize that spending the same way they manage any other operating expense.