Token ROI is now a governance problem, not just a cost issue

At a glance

What this is: Tetrate argues that token spend is now large enough to require governance, but the real problem is proving whether that spend produced business value.

Why it matters: IAM, NHI, and AI governance teams should treat token attribution and runtime identity as part of cost control, because spend without workload-level accountability cannot be governed reliably.

Context

Tokens are the metered unit of modern AI spending, but they do not behave like cloud instances or software licenses. They are consumed in different forms, priced differently, and often attributed too late for meaningful governance. That makes token economics a measurement problem as much as a cost problem.

The article also has a clear identity angle: agent credentials, workload labels, and runtime ownership determine whether AI spend can be traced to a responsible system or team. Without that linkage, token governance becomes an exercise in reconstruction after the fact rather than control at the point of use.

Key questions

Q: How should security teams govern AI agent token spend without losing accountability?

A: Treat token spend as an identity control problem. Give each agent its own credential, scope its tool access tightly, and use session boundaries so every model call can be attributed to a specific executor and task. That makes cost governance, access review, and revocation part of the same control plane.

Q: Why does token ROI become difficult when attribution happens after the bill arrives?

A: Because the bill is too coarse to explain who used the tokens, for which workload, and under what operating context. By the time the invoice arrives, agents, keys, and team ownership may already have changed. Governance has to move to request time so the spend record and the identity record are created together.

Q: What breaks when AI cost is tracked only at the invoice level?

A: Invoice-level tracking hides the request, retry loop, and agent step that created the overspend. It also collapses different workflows into a single total, which makes ownership unclear and slows remediation. Teams end up explaining the bill after the fact instead of controlling the behaviour that generated it.

Q: Should organisations prioritise token governance or model cost reduction first?

A: Token governance first. Cost reduction is useful only when you can attribute spend to a workload and prove whether the output justified the cost. Without runtime attribution, cheaper models may simply hide waste instead of fixing it, and finance teams will still lack a defensible answer at budget review time.

Technical breakdown

Why token spend is harder to govern than cloud spend

Token economics breaks familiar FinOps assumptions because the unit of consumption is abstract, variable, and context dependent. Input, output, cached, and reasoning tokens can carry very different costs, so a single invoice total hides the real drivers of spend. Unlike cloud resources, tokens do not have stable infrastructure identifiers that automatically map back to a team, workload, or objective. That means cost governance has to move closer to the request path, where model selection, workload context, and consumption type are still visible. See the Identity and NHI Security Business Case Guide.

Practical implication: Treat token spend as a runtime governance problem, not a month-end accounting exercise.

Why attribution fails when teams wait for the invoice

Attribution decays quickly because the key, the agent, the workload, and the human owner can all change before the bill arrives. By then, a token line item no longer tells you who caused the spend, what the request was for, or whether the usage was productive. This is the same control problem that appears in [[L1|secret sprawl]] when credentials outlive the context that created them: once identity and usage drift apart, governance becomes forensic. Token ROI therefore depends on capturing ownership at issuance time, not reconstructing it later. Read Guide to the Secret Sprawl Challenge.

Practical implication: Bind every AI credential to an owner, team, agent, and workload at creation time.

Why routing is the control point for token governance

A routing layer can see the request before it is priced, attributed, or aggregated into a coarse monthly number. That makes it the only practical place to record which agent used which model, how many tokens were consumed, and whether the request belongs to a workload that has a meaningful outcome metric. This matters for both financial control and identity governance because the router becomes the policy boundary for who can spend, on what, and under which operating context. In effect, token governance is an access-control problem expressed through metered usage. Read the Ultimate Guide to NHIs: What are Non-Human Identities.

Practical implication: Move token metering, policy enforcement, and ownership capture into the runtime layer.

Threat narrative

Attacker objective: The objective is not theft in the classic sense but uncontrolled consumption, where AI usage becomes expensive, unauditable, and hard to justify.

1. Entry occurs when agents or workloads obtain API keys or other credentials that let them consume model capacity.
2. Escalation happens when those credentials are reused across models, workloads, or teams without strong attribution or policy boundaries.
3. Impact is budget opacity and governance failure, because teams cannot reliably prove which AI activity produced value or waste.

Breaches seen in the wild

• Salesloft OAuth token breach: hackers stole OAuth tokens to access Salesforce data via Salesloft.
• Internet Archive breach 2024: An exposed GitLab token opened Internet Archive code and 31 million user records; unrotated Zendesk tokens let the attacker back in weeks later.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

Token economics is now an identity governance problem as much as a finance problem. Once a machine credential can initiate model spend, the question is no longer only what was consumed but who or what was authorised to consume it. That creates a governance bridge between NHI management and FinOps that most organisations have not formalised yet. Practitioners should treat AI spending identities as governed assets, not anonymous billing events.

Attribution must move from the invoice to the issuance point. The article is right to say the invoice is too coarse to explain token ROI, but the deeper lesson is that post-hoc accounting always loses context. When ownership, workload, and agent identity are attached after the fact, the control has already failed. The spend is still real, but the governance signal has already degraded.

Token governance is becoming a policy boundary for agentic systems. As agents consume more model capacity, the ability to constrain and explain token use will matter alongside the ability to authorise tool access. That is why runtime enforcement matters: it turns abstract cost into a controlled operating condition rather than a surprise on the ledger. Practitioners should align token control with identity control from the start.

Secret sprawl and token sprawl are converging risks. The same operational habits that leave API keys unowned also leave token spend unbounded and untraceable. If a team cannot say which workload created the cost, it also cannot say who should be accountable for the access that created it. The right response is to govern both credential lifecycle and consumption telemetry together.

FinOps for AI will not work as a standalone discipline if identity is missing. The standards conversation matters, but standard benchmarks only help when organisations can map spend back to an accountable workload. That means AI cost governance must inherit the same discipline used in NHI governance, where identity, lifecycle, and authorization are never treated as optional metadata. Practitioners should build for accountable consumption, not just comparable pricing.

What this signals

Token cost governance will increasingly sit beside workload identity governance. Organisations that already manage non-human identities can extend that discipline to AI spend by treating model requests as accountable system actions, not anonymous transactions. The organisations that do this well will be able to compare cost, usage, and outcome on the same control surface.

Runtime attribution will become the decisive control for AI economics. The practical shift is away from reconstructing spend later and toward capturing it at the moment of use. That is the same structural lesson identity teams learned with service accounts and API keys: if the identity is not captured where the action happens, governance arrives too late.

For practitioners

• Instrument token spend at the routing layer Capture model, workload, agent, owner, and token type at request time so the cost record is usable for both finance and governance.
• Bind AI credentials to accountable ownership Require every API key or agent credential to inherit a named owner and workload on issuance, and revoke keys that cannot be attributed cleanly.
• Separate pricing analysis from outcome analysis Measure fair price with tokens per dollar, then evaluate effectiveness with workload-specific output metrics such as tickets closed, PRs merged, or decisions completed.
• Review model selection against real workload quality bars Test expensive default models against a defined acceptance threshold and downgrade workloads that meet the bar with lower-cost models.

Key takeaways

• Token spending has crossed the threshold where it needs governance, but the harder problem is proving whether the usage produced value.
• The article shows that invoices alone cannot explain AI consumption because attribution, model choice, and workload context decay before finance sees them.
• Practitioners should move token control to the routing layer and bind every AI credential to an accountable workload and owner.

Related NHI Mgmt Group resources: Guide to the Secret Sprawl Challenge · Ultimate Guide to NHIs: What are Non-Human Identities · Identity and NHI Security Business Case Guide · Guide to NHI Rotation Challenges

Key terms

• Token Economy: A token economy is a usage-based pricing and control model where AI service consumption is measured by the amount of model processing, not by a fixed seat licence. For identity teams, it creates a direct link between who or what is using the system and how fast cost and risk can grow.
• Runtime Attribution: The ability to explain which identity initiated an action, what it was allowed to do, and how that action propagated across systems. For autonomous or semi-autonomous AI, runtime attribution is essential because after-the-fact logs may show activity without revealing the decision path.
• Workload Identity: The identity assigned to a software workload, such as a containerised application, serverless function, or microservice, enabling it to authenticate to other services without storing static credentials.
• Token Sprawl: Token sprawl is the accumulation of too many active, forgotten, or overlapping tokens across SaaS and automation workflows. It creates visibility gaps, increases the chance of over-privilege, and makes revocation slow when an incident forces a response.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, workload identity, and secrets management. It helps practitioners connect identity controls to the broader security programmes they run across cloud and AI.