Cost per Unique Outcome is a practical efficiency metric for AI systems: total AI spend divided by the number of distinct useful results produced. It asks not just how much a model costs to run, but how much each new, non-duplicate answer, decision, or action costs the business. In security operations, this matters because repeated prompts or automated workflows can generate many paid model calls while adding little new value.
Cyber defenders use this metric to spot waste from redundant inference, prompt loops, and repeated retrievals in stateless services. Attackers can also exploit that weakness by flooding an AI app with near-duplicate requests, driving up cloud bills and consuming rate limits. Measuring cost per unique outcome helps teams justify caching, deduplication, semantic reuse, and tighter workflow controls. A rising value can indicate poor prompt routing, weak cache hit rates, or a system that is producing the same answer many times instead of creating genuinely new outcomes.