Fraud and anomaly detection for transactions, refunds, discounts and staff actions
We build fraud and anomaly detection systems. Rules and AI models check every transaction, refund, discount and voided receipt. Anything suspicious reaches the right person at once, with an explanation of what is wrong.
What hurts
Losses surface at stocktake or at quarter end. Refunds without goods, discounts for friends, receipts voided after a cash payment. By then, who did it and when is impossible to trace.
Checks are manual and selective. Security or audit reviews a sample and misses the rest. Schemes that repeat in small amounts are invisible in summary reports.
Simple rules raise too many alerts. The team stops reading them, and the real case drowns among a hundred false ones. Fraudsters quickly learn to stay under fixed thresholds.
What we build
- Analysis of transaction history and known fraud cases
- Rules for obvious schemes: refunds without a sale, discounts over limit, night operations
- Anomaly models that find unusual behaviour without predefined rules
- Behaviour profiles of customers, cashiers and locations compared with the norm
- A risk score for every operation with reasons
- Prioritised alerts to Microsoft Teams, Telegram or your case management system
- A case review workspace: confirm, dismiss, escalate
- For retail, matching POS receipts with footage from your existing cameras
What you get
- A working risk scoring system with rules and models
- Configured alerts and a case review workspace
- A dashboard of cases, losses and team workload
- Code, rule documentation and model methodology
- Training for your security or audit team
How we measure the result
- Share of confirmed cases among all alerts
- Time from suspicious operation to response
- Losses detected by the system per month
- Alerts per analyst per day
Systems and stack
Industries
Questions
How much does fraud detection cost?
From $10,000. The price depends on the number of transaction sources, the response speed you need and whether checks must run in real time. A daily review of store receipts costs less than scoring every bank transaction at the moment of payment. We fix the estimate after discovery on a sample of your operations.
What data do you need?
Six months or more of operations: transactions, receipts, refunds, discounts and who processed them. Known fraud cases help a lot, even a few. For banks we run on dedicated servers, and data never goes to third-party models. We sign a GDPR DPA before access.
How many false alarms will there be?
There will be some, because AI sometimes gets it wrong. So at launch the system runs in silent mode: it scores operations but blocks nothing. Together with your team we tune thresholds so the alert volume matches what people can actually review.
Does it work with POS and cameras?
Yes. For retail and hospitality we connect the POS and match suspicious receipts with footage from your existing cameras. We run this for a bakery chain under NDA, where every receipt is checked against video. You see the clip next to the operation.
What about regulation and explainability?
For banks and insurers we follow your regulator rules and internal security policies, and check EU AI Act obligations where they apply. We hold ISO/IEC 27001:2022, certificate No. 4160097. Every decision comes with reasons, so it can be shown to an auditor.
How do you keep up with new fraud schemes?
Fraudsters change schemes, so rules and models need updates. Under an SLA we analyse confirmed and dismissed cases, retrain models and add new rules. You keep the code and documentation.
Related services
More in this section
Describe the task. We reply within one business day, the first call is with the founder.
Describe the task. The first call is with the founder.
We reply within one business day. 30 minutes to tell you whether it is worth doing and what it costs.