A measure of how much it costs to complete a task successfully, not merely to generate a response. It includes retries, fallback runs, safety checks, and human cleanup, which makes it a better operational metric than token cost for agentic systems.
Expanded Definition
Cost Per Resolved Request measures the true unit economics of an automated workflow by focusing on successful task completion rather than raw output volume. For agentic systems, that means counting the full operational path: retries, model reruns, escalation to a human, policy checks, tool failures, and the cleanup effort required to close the request. In practice, this makes it more useful than token cost alone, because a low-cost response that fails validation is still an expensive outcome once remediation is included.
The term is especially relevant where agents, workflows, and control checks interact. A request can appear efficient at the model layer while still being costly at the service layer if it repeatedly misses the required result, triggers guardrails, or creates manual follow-up. NIST’s Cybersecurity Framework 2.0 does not define this metric directly, but its emphasis on governance and outcome-focused risk management aligns with the need to measure operational effectiveness, not just activity.
The most common misapplication is treating token spend as a proxy for business efficiency, which occurs when teams ignore retries, exceptions, and the cost of human intervention.
Examples and Use Cases
Implementing Cost Per Resolved Request rigorously often introduces measurement overhead, requiring organisations to weigh cleaner performance insight against the effort of instrumenting every fallback, escalation, and repair step.
- A support agent handles routine password reset requests. The model’s response is cheap, but the real cost rises when failed identity verification forces repeated retries and help desk intervention.
- A security operations workflow drafts incident summaries. If analysts must repeatedly correct context, verify evidence, and rerun retrieval steps, the resolved-request cost is higher than the apparent API spend.
- An AI agent submits procurement approvals. Each policy violation, tool timeout, or manual sign-off adds cost even when the request eventually completes successfully.
- A knowledge assistant answers internal policy questions. If the answer must be regenerated after safety filtering or source validation, the effective cost per resolution includes those extra passes.
- A multi-step NHI workflow rotates secrets and updates dependent systems. The true cost reflects not only automation calls, but any rollback or cleanup when one downstream integration fails.
For identity and access-heavy environments, the metric becomes more accurate when linked to the full request lifecycle described in guidance such as NIST Cybersecurity Framework 2.0, because operational success depends on trust, control, and recovery as much as on speed.
Why It Matters for Security Teams
Security teams need Cost Per Resolved Request because agentic systems can create hidden work even when they appear to be automating successfully. A workflow that repeatedly fails validation, over-escalates to humans, or loops through guardrails can consume more analyst time than the manual process it was meant to replace. That risk is especially important in NHI operations, where requests may involve secrets rotation, service account updates, or agent permissions that must be handled precisely.
When this metric is ignored, teams may optimise for throughput while degrading reliability, increasing alert fatigue, and masking control failures behind apparently low API costs. It also helps leaders compare different models, orchestration patterns, and guardrail strategies on a realistic basis. The right question is not how much a model call costs, but how much it costs to finish the task correctly and safely.
Organisations typically encounter the real burden of this metric only after automation starts generating repeated exceptions, at which point Cost Per Resolved Request becomes operationally unavoidable to explain the surge in cleanup effort.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Agentic AI Top 10 and OWASP Non-Human Identity Top 10 address the attack and risk surface, while NIST CSF 2.0, NIST AI RMF and NIST SP 800-63 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | GV.OC-01 | Outcome-focused governance supports measuring operational effectiveness, not just activity. |
| NIST AI RMF | AIRMF emphasizes lifecycle risk management for AI systems, including monitoring and improvement. | |
| OWASP Agentic AI Top 10 | Agentic AI guidance highlights failure modes that create hidden operational cost. | |
| OWASP Non-Human Identity Top 10 | NHI governance depends on tracking the full cost of identity and secret-related workflows. | |
| NIST SP 800-63 | IAL/AAL | Identity assurance affects how often requests must be retried or manually resolved. |
Account for verification failures and step-up checks when estimating the true cost of identity-bound requests.
