Use cases

Real decisions, with real delays

Each one is a question an app asks a decision model, the outcome that comes back later, and the target its policy has to meet. Between them they cover all three question types, an escalation band, labels that take months, and a cost model.

noulDemo available Refund fraud An online shop screens refund requests before paying out. “Is this refund request fraudulent?” Outcome · a chargeback, 30 to 90 days later Target · false passes ≤ 2% with 95% confidence; the uncertain middle goes to the risk team choiceDemo available Support ticket triage An online shop routes tickets to billing, shipping, returns or tech. “Which team should handle this ticket?” Outcome · the team that finally handled it, hours later Target · misroutes ≤ 25% among auto-routed tickets in the demo scoreComing soon Model routing A chat product sends easy requests to a small model and hard ones to a large one. “How hard is this request?” Outcome · a thumbs-down, a retry, or an LLM judge on a sample Target · the cost of a large-model call against the cost of a bad answer noulComing soon PII guardrail An internal chat assistant checks every prompt before it reaches an LLM. “Does this text contain personal data?” Outcome · the privacy team reviews a sample, days later Target · false passes ≤ 1% with 95% confidence noulComing soon Agent tool approval An ops agent wants to run a database write or send an email. “Is this tool call safe to run without a human?” Outcome · a human approves or denies; incident reports Target · unsafe passes ≤ 0.5% with 95% confidence scoreComing soon Content moderation A marketplace checks new listings before they go live. “How serious is the problem with this listing?” Outcome · a moderator decision or a user report, days later Target · a review queue under [N] items a day