As AI is shifting past the pilot stage, the query arises whether or not banks can flip rising AI consumption into outcomes which are predictable, repeatable and defensible.
The numbers recommend most can’t but. The 2026 State of Tokenomics examine of 472 organisations discovered that three in 4 couldn’t confidently show AI enterprise outcomes to the CFO.
Nearer to house, 54% of Asia-Pacific respondents in an IDC examine for Expereo reported value overruns or missed ROI, 80% in Malaysia.
In a Cognizant survey of 250 FinOps leaders, 72% had exceeded their AI budgets previously yr.
The problem isn’t the worth of a token. It’s the working system round AI: who consumes it, for which final result, underneath what coverage, on which mannequin, and with what proof of high quality and return.
Visibility is the primary management
AI spend enters a financial institution via many routes, from AI mannequin APIs to embedded software program options and self-hosted infrastructure. The mannequin bill is just a part of the invoice: vector databases, embeddings, orchestration, observability and knowledge motion can account for 40% to 60% of an AI function’s value.
That makes a month-to-month token complete a weak administration metric.
Value per onboarding file processed, servicing request resolved or alert reviewed tells a financial institution way over value per million tokens. Failed and discarded runs should keep within the denominator.
Patrons already know this: 23% of State of Tokenomics respondents requested suppliers for extra granular utilization knowledge. Solely 4% requested for decrease costs.
Management should comply with each request
As soon as consumption is seen, it may be steered. Not each process wants a frontier mannequin—and one can value 50 to 100 instances extra per token than the smallest mannequin from the identical supplier.
Routing, the follow of dynamically directing an incoming immediate or process to essentially the most appropriate AI mannequin, goes mainstream: 86% of respondents use or are evaluating a mannequin router, and customers are 4 instances extra more likely to present the CFO measurable worth.
However routing isn’t only a value lever. It’s a resilience and governance management. Delicate workloads keep contained in the enterprise. Visitors strikes when a supplier slows or fails. Coverage travels with each request, somewhat than counting on customers to decide on accurately.
Governance should produce proof
For banks, these disciplines are converging with regulation. The Financial Authority of Singapore’s AI danger administration tips, printed on 7 October 2026, name for clear accountability, an correct stock of AI use, risk-materiality assessments, lifecycle controls and oversight of third-party AI.
Governance expectations apply from October 2027, and lifecycle controls from October 2028.
A financial institution can’t maintain a dependable stock when AI exercise is fragmented throughout suppliers. It can’t assess materiality with out understanding the mannequin, knowledge, proprietor and goal behind every use case. And it can’t proof oversight with no document of approvals, modifications, efficiency and exceptions.
Agentic AI raises the stakes: 97% of organisations in Cognizant’s FinOps survey had an AI agent session escalate past anticipated value or exercise. MAS plans to seek the advice of on agentic AI in 2027.
Governance can’t be a gate on the finish of improvement. It has to run constantly, within the stream of labor.
The identical proof that satisfies auditors and regulators ought to assist groups elevate high quality, reuse what works and cease funding what doesn’t.
Three strikes to start out
- See the economics. Construct one cross-provider view of AI accounts, API keys, instruments and infrastructure, and attribute each value to an proprietor, workload and measurable final result.
- Steer the workload. Route every request by high quality, value, latency, resilience and data-control wants, with budgets and guardrails set per workload, not simply per supplier account.
- Show and enhance the consequence. Maintain an audit path of fabric exercise and human approvals, measure high quality beside unit value, and switch the best-performing patterns into enterprise requirements.
The place Cognizant AI Nexus suits
Cognizant AI Nexus was designed for this problem: a modular, multi-model working layer round whichever fashions a financial institution already makes use of. Banks can begin with their most urgent constraint and add modules as they mature.
- Reveal maps tokens, instruments and {dollars} to folks, tasks and outcomes.
- Route directs every request to the mannequin with the correct stability of high quality, value and management.
- Coach evaluates AI work in context and turns higher practices into shared requirements.
- Interact provides engineers and enterprise customers one ruled workspace.
- Orchestrate converts enterprise processes into audited agentic workflows, with human approval retained for crucial selections.
The purpose isn’t one other AI device. It’s a ruled surroundings the place profitable work is measured, reused and improved throughout groups.
Mannequin functionality will nonetheless matter, however fashions will maintain altering. Sturdy benefit will come from how briskly a financial institution can flip AI spend into outcomes it will probably clarify, govern and repeat.
For CIOs, CFOs, CROs and COOs throughout ASEAN banking: if the board requested tomorrow what your AI funding returned, who owned the consequence and which controls utilized, how lengthy wouldn’t it take you to reply?
Featured picture credit score: Edited by Fintech Information Singapore, primarily based on picture by byzain by way of Magnific