A Sibos poll found that more than half of companies lack granular visibility into their AI use.
Many financial firms struggle to measure the return on investment of their AI deployments because they lack visibility into AI resource consumption, according to a panel poll at Sibos 2026 in Miami.
At the heart of the matter are AI tokens, the basic unit of text used by large language models and the primary metric AI model providers use to measure client usage.
Nearly two-thirds (61%) of poll respondents reported having little or no visibility into AI token usage within their organization, even though they understood their overall AI spending. Only 13% of respondents reported having detailed visibility and actively managing costs at the token/model level.
“Some clients and partners that we worked with have told us horror stories where they let everyone go crazy with this stuff and build their own agents only to be left with a $100 million bill after a few months,” said Melissa Tuozzolo, global head of client services at HSBC, during a Sibos 2026 panel in Miami.
Calculating AI ROI
Understanding AI token cost and usage matters, but it’s not the only factor in calculating ROI for AI-powered projects, said co-panelist Isabel Schmidt, executive platform owner, payments enablement at BNY.
“There’s obviously still a lack of clarity in many organizations around the actual individual cost of tokens,” she said. “But we also need to be very careful to not get pulled into that rabbit hole or what a particular token costs because, in the end, AI is not a goal itself. The ultimate question is: How much does that business process cost me?”
In addition to token cost, organizations need to consider related costs associated with data governance and cleaning, upgrading existing systems, and deploying new technology, Tuozzolo added.
The industry is in an era when pretty much everything being built has an AI component, said David White, global head of product and data at LSEG and a panelist. “I encourage my product teams and others to think in terms of this as just another technology to help deliver value to the customer.”
AI Value-Add
BNY’s Schmidt sees AI’s primary value as improving an organization’s efficiency, effectiveness, and capacity
“There has been a lot of dialogue around the efficiency part of what AI can contribute to our businesses and processes since you can automate processes much, much more easily than maybe we could even a year or two ago,” she said. “It also forces us to think more clearly about some of the others.”
In terms of effectiveness, HSBC deployed an AI-powered “air traffic control” system that directs client queries to the proper expert or teams about six months ago. It reduced first-response time for clients in certain markets by 10 hours. Prior to that, “we had used people that manually take in queries from clients and triage them,” said Tuozzolo.
Other AI deployments within the bank have freed up processing capacity and allowed it to be reallocated.
“Historically it’s always been that when our transaction volumes go up, our costs go up from an investigation and service standpoint,” she said. “As our volume goes up, the number of queries also goes up. This is the first year in a few years that we’re actually seeing it go the other way.”
Tuozzolo attributed these gains not only to AI, but also to HSBC’s investment in a cleaner data layer.
Meanwhile, Sumitomo Mitsui Banking Corp. (SMBC) is deploying an AI-powered call center that will process calls at 70% of the older model’s cost.
“We are seeing most of this cost for the IT project consisting of the AI tokens, but we are seeing a positive ROI,” said panelist Kazuya Ikeda, senior executive manager at SMBC.
Nonetheless, he cautioned that not all value-add for clients leads to economic value. “If other peers are doing the same thing, then it’s not necessarily creating relative value for clients,” he said.
Leave a Reply