Your finance team can already tell you what the organization spent on AI last month by tool, team, and model. What they usually can't tell you is whether any of it worked. That's not a flaw in your cost model. It was never designed to measure whether AI spend produced value.
Key Takeaways
- TBM and FinOps can show what AI costs and how it’s consumed. They don’t measure workforce proficiency or independently prove that AI spend produced a business result.
- TBM can allocate AI costs across known applications and business units, but it can’t account for AI usage that never enters the application inventory, including shadow AI and personal accounts.
- FinOps was built for variable, consumption-based spend. The harder questions are workforce proficiency, human-versus-agent attribution, and connecting spend to business outcomes.
What TBM and FinOps Actually Do Well
Technology Business Management and FinOps are related, but they serve different purposes. TBM, maintained by the TBM Council, is a classification standard. The TBM Taxonomy organizes technology cost, consumption, and resources into cost pools, towers, applications, and services so IT, finance, and business leaders share a common language and can trace spend to business value. The Council describes the current Taxonomy as extensible to cloud, AI, Agile, and FinOps, not something built only for static, pre-cloud technology.
FinOps, maintained by the FinOps Foundation, was built for exactly the kind of spend AI produces. The Foundation's 2024 definition update describes FinOps as bringing financial accountability to cloud's variable, usage-based cost model and explicitly extends that approach to other variable-cost and consumption-based services. Both disciplines build toward business value: the FinOps Framework includes unit economics and KPI-driven benchmarking aimed at connecting spend to value, and the TBM Taxonomy traces technology spend to business value.
What Cost Data Still Can’t Tell You
TBM and FinOps can show where AI spend goes. They don’t tell you whether the investment is paying off.
Attribution Gets Harder Before It Gets Easier
The FinOps Foundation's AI working group notes that a billing dashboard will show API costs, but without added tracing instrumentation, it can be difficult to connect those costs back to specific user interactions or individual agent runs. Practitioner accounts published by the Foundation describe teams building manual workarounds today, such as partnering with HR to tie license data to a person, while attribution tooling for agentic workloads continues to develop.
Consumption is also far less predictable than seat-based software. A 2026 study of eight frontier models on SWE-bench Verified found that identical coding tasks produced token usage that varied by as much as 30 times between runs of the same model. That kind of run-to-run variability in agent consumption is exactly the pattern a flat, seat-based license count was never designed to capture.
The Application Inventory Is Never Complete
A TBM model allocates cost against a known set of applications. Shadow AI, including personal account use, and AI features embedded inside already-approved SaaS tools generate real cost and risk outside a procurement-driven catalog. Larridin's State of Enterprise AI 2026 Report found that 45% of AI adoption happens outside IT's visibility, which limits how complete any cost-allocation inventory can be regardless of the framework applied to it.
Cost Visibility Isn't Value Visibility
This is the most consequential gap, and it's one the market is only starting to address. IBM says its Apptio AI TCO & Usage product tracks total cost of ownership, token consumption trends, and adoption across business units. It doesn’t claim to measure whether usage translated into a business result. IBM addresses that separately with Apptio AI Value & ROI, built specifically to connect AI spend to business outcomes.
On August 6, 2026, IBM announced the public preview of Apptio AI Value & ROI, with general availability planned for the third quarter of 2026. IBM cited Gartner research finding that 84% of finance leaders have not been able to measure the ROI of their AI initiatives and, among those who tried, roughly two in five succeeded.
That product split makes the distinction clear. AI Value & ROI connects spend to outcomes by having customers select proof metrics, such as cycle time or cost avoided, for each initiative and track them against a baseline. It doesn’t independently measure how deeply or proficiently people use AI day to day.
AI ROI Requires More Than Cost Visibility
None of this means that TBM or FinOps should be set aside for AI spend. The allocation rigor those practices bring still matters, especially for the parts of AI spend that behave like conventional software: seats, platform fees, and committed infrastructure.
AI ROI adds another layer: whether the people and agents using AI are doing so productively and whether that usage is moving a business metric that matters.
Question | TBM / FinOps Answers This | Needs a Different Layer |
|---|---|---|
What did we spend, by application and business unit? | Yes | |
Is spend growing faster than budget? | Yes | |
How is AI being consumed across models and business units? | Yes | |
Which teams are using the AI tool, and how much? | Yes | |
How deeply or proficiently are they engaging with it? | Yes | |
Is usage split between humans and autonomous agents, and by how much? | Partially (emerging) | Yes |
Are the people using AI getting proficient at it? | Yes | |
Did that spend produce a measured business outcome? | Yes (or a connected tool) |
A mature TBM or FinOps practice can answer the cost and consumption questions above. AI ROI gets harder because it requires utilization, proficiency, attribution, and outcome data, not just a more detailed cost report.
Frequently Asked Questions
Can our existing TBM or FinOps platform measure AI ROI?
It can measure AI cost and consumption well, which are necessary inputs to ROI, not the ROI answer itself. IBM's AI TCO & Usage product's scope is cost, usage, and adoption tracking. The company built a separate product, AI Value & ROI, specifically to connect that data to business outcomes.
Doesn't FinOps already handle variable, usage-based costs?
Yes, and that's exactly why it isn't the obstacle here. FinOps was built for cloud's variable spend model from the start. The open problem is that billing data alone often can't connect costs back to specific user interactions or individual agent runs without added tracing instrumentation, not a limitation of the variable-cost model itself.
Should we replace our TBM or FinOps platform with an AI-specific tool?
Not necessarily. The allocation discipline a TBM or FinOps practice brings is still useful for the licensed and infrastructure portions of AI spend. The practical approach is pairing that cost rigor with a measurement layer that covers utilization, proficiency, attribution, and outcomes, the parts of the AI ROI question a cost-allocation platform isn't built to answer on its own.
What's the best enterprise tool to measure AI ROI if we already have an IT financial management platform?
Look for one that builds on the cost data your TBM or FinOps platform already provides, adding utilization depth, human-versus-agent attribution, proficiency, and business outcomes. That gives you the additional measurement needed for AI ROI without replacing the financial-management system you already use.
How Larridin Complements TBM and FinOps
Larridin measures AI usage and impact at that layer. Token Spend & Insights consolidates and attributes AI spend, AI Adoption and AI Fluency measure utilization and proficiency, and AI Impact connects usage to engineering and sales outcomes such as deployment velocity and deal cycles. None of this replaces a TBM or FinOps practice; it answers the question those practices were never built to answer alone.
See how Larridin connects AI spend, usage, proficiency, and business outcomes across the enterprise.