Explain the risk score and evidence for a suspicious promotional order.
Promotion abuse analysis
A multi-brand retail project in Türkiye · Risk analyst
$0.026580Recorded execution cost
Client, brand and personal information has been removed. Examples include summaries and complete available conversations. Full conversations preserve every user and agent message in order, translated into English where needed. The status reflects the outcome reported at the time.
Reported a score of 73/100 and the system’s medium-risk classification for an order in a test scenario. Cited triple the usual points earning, a TRY 45,490 basket and device/IP indicators. Left the approval or rejection decision to a human moderator rather than automatically rejecting the order.
Awaiting human review
Cost scope
- Models used in this execution
- Claude Haiku 4.5
- Runs / model calls
- 1 / 2
- Input / output tokens
- 2,181 / 627
- Record date
- Jul 28, 2026, 4:53 PM
1 recorded run(s), 2 model calls. The request text, conversation and execution record were matched. No percentile filter was applied because this agent had fewer than 10 records.
This task ran as a test scenario; the amount shown is its recorded execution cost.
Agent costs vary dynamically with the scope of the request, conversation context, models and tools used, response length and usage volume. The amounts below are reference examples from verified historical execution records. External service and infrastructure fees should be considered separately.
These are selected examples, not the limits of what agents can do. Agent capabilities extend beyond the tasks, industries and regions shown here, and can be adapted to different workflows and business needs.