Healthcare Payors Agent Registry Cost per Case | Moring
Cost per case measures the model and tool spend behind each completed payor workflow instance, from a prior authorization review to a resolved claims exception. The registry makes the metric honest by fixing the case definition, the workflow boundary, and the owner, so the figure leadership sees reflects one governed workflow rather than a blended average across an unmanaged agent fleet.
Short answer
Cost per case measures the model and tool spend behind each completed payor workflow instance, from a prior authorization review to a resolved claims exception. The registry makes the metric honest by fixing the case definition, the workflow boundary, and the owner, so the figure leadership sees reflects one governed workflow rather than a blended average across an unmanaged agent fleet.
How it works
Cost per case is the metric that converts agent spending into operational terms a payor can act on. Aggregate spend says little; a figure expressed as cost per completed prior authorization, per resolved claims exception, or per member service interaction can be compared against the labor cost it replaces, tracked over time, and defended in a budget review. The registry is what makes the metric trustworthy, because the metric depends on definitions that the registry alone can hold stable.
The first dependency is the case definition. What counts as one case must be fixed per workflow: an authorization request processed to a determination, a claim exception taken from detection to resolution, a member interaction from first contact to outcome. Without a registry-recorded definition, teams measure differently and the numbers cannot be compared or combined. The registration entry, which already records the workflow the agent serves, is the natural home for the case definition, so every consumer of the metric reads the same one.
The second dependency is complete cost capture. Cost per case must include model inference, tool and API consumption, orchestration overhead, and the human review time that the workflow's own design requires, because an agent workflow that routes 30 percent of cases to licensed reviewers is not as cheap as its token spend suggests. Registry-linked attribution gathers the machine consumption; the human component comes from the workflow's design parameters recorded at registration. Together they produce a figure that survives scrutiny, whereas token-only figures systematically understate what a governed workflow actually costs.
The third dependency is variance discipline. A cost-per-case figure is useful when it moves and explainable when it does. Registry-recorded version changes give the explanation: when a prompt revision, a new tool, or a policy change alters the workflow, the metric's movement should coincide with a recorded change, and unexplained drift should trigger investigation. Common causes are looping on a tool call, processing cases outside the registered purpose, or a retrieval change that inflates context size. Each of these is also an operational or compliance signal, which is why cost-per-case belongs on the same dashboard as quality and scope indicators rather than in a finance-only view.
The metric also disciplines vendor and model choices. When cost is visible per case, the comparison between model options, or between an agent workflow and the manual process it augments, becomes concrete. Healthcare agent economics are dominated by the governance and integration layer rather than the model, and cost per case is the figure that exposes that structure: a workflow with modest inference spend but heavy human checkpoint time has a different optimization path than one burning tokens on redundant retrieval.
Benchmarking completes the picture. Per-case figures across comparable workflows let a payor see which agents earn their keep, and registry attribution makes the comparison fair by ensuring each workflow's figure covers the same cost elements. Leadership decisions about expansion and retirement then rest on consistent evidence, and the agents that survive those reviews are the ones whose owners can explain their numbers.
The practical starting point: pick the two or three highest-volume workflows, register their case definitions and cost components, publish cost per case alongside quality metrics, and require that every material change to a workflow be accompanied by its expected and observed effect on the figure. That habit, more than any tooling, is what makes cost per case a governance instrument rather than a dashboard decoration.
Frequently asked questions
Why must the case definition live in the registry?
Cost per case is only comparable when one case means the same thing everywhere: an authorization processed to a determination, a claims exception taken to resolution, or a member interaction carried to outcome. Recording the definition in the registration entry fixes it per workflow, so teams cannot measure differently and figures can be combined across the fleet. Without it, the metric degenerates into blended averages nobody can defend. A payor that standardizes definitions early avoids relitigating them in every budget conversation afterward.
What belongs in the cost numerator for a payor workflow?
Model inference, tool and API consumption, orchestration overhead, and the human review time the workflow design requires. A workflow routing a large share of cases to licensed reviewers is not as cheap as its token spend implies, so registry-linked machine attribution plus recorded human checkpoints produce the figure that survives scrutiny. Token-only numerators systematically understate governed workflows.
How should unexplained cost-per-case drift be handled?
Investigate it against registry-recorded version changes first, since drift that coincides with a prompt revision, new tool, or policy change is explained, and drift that does not is a signal. Typical causes are looping tool calls, processing outside the registered purpose, or retrieval changes inflating context. Keep the metric beside quality and scope indicators, because the same anomaly often carries operational and compliance meaning.
How does cost per case discipline vendor and model decisions?
Visible per-case cost makes comparisons concrete between model options and between the agent workflow and the manual process it augments. Healthcare agent economics are dominated by the governance and integration layer rather than the model, and the figure exposes that structure, showing whether optimization should target inference spend, redundant retrieval, or human checkpoint time. Fair benchmarking across workflows depends on registry-attributed, consistent cost elements.