---
title: "Fraud Detection: SUMWARE spots POS fraud in the conversation"
description: "SUMWARE Fraud Detection listens at the POS microphone and detects whether anyone spoke, with nothing recorded. GDPR-compliant: only a 0/1 pattern of the last 180 seconds. Silent returns and four-eyes gaps become visible."
canonical: "https://sumware.at/en/solutions/fraud-detection/"
language: "en"
alternates:
  - { hreflang: "de", href: "https://sumware.at/loesungen/fraud-detection" }
  - { hreflang: "en", href: "https://sumware.at/en/solutions/fraud-detection" }
  - { hreflang: "x-default", href: "https://sumware.at/loesungen/fraud-detection" }
source: "Generated from the published HTML page; the HTML at the canonical URL is authoritative."
---

Solution: Fraud Detection

# You can hear fraud. *Without* listening in.

SUMWARE Fraud Detection listens at the POS microphone and detects whether anyone spoke at the point of sale: employee, customer, both or nobody. Nothing is recorded; all that is stored is a pattern of ones and zeros covering the last 180 seconds. When a receipt is finalized, the POS pulls the conversation pattern. A cash refund without a single spoken word becomes visible the moment it happens.

[Book a Fraud Detection demo](https://sumware.at/en/#contact) [See the full feature set →](https://sumware.at/en/solutions/fraud-detection/#catcher)

## From the microphone to the red flag.

A dedicated service per POS station captures the sales conversation as a pattern and returns it to the POS on demand. GDPR compliance at the core, self-learning from day one. The features below show how silence becomes a signal.

- Listening without recording

  ### Listens at the POS microphone. Stores not a single word.

  The service detects human voices through the POS microphone, even over music and background noise. Nothing is recorded: no audio file, no transcript, no content. All that is stored is a pattern of ones and zeros covering the last 180 seconds. GDPR compliance is designed into the core, not retrofitted with a deletion policy.
- Voices, not identities

  ### Detects that someone is speaking. Not who.

  The system tells voices apart without identifying people. Each voice it hears is classified purely by how often it occurs. Frequent voices count as staff, rare ones as visitors. Who is behind a voice remains unknown and is never stored.
- Visitors are forgotten daily

  ### Staff voices take root. Customer voices evaporate.

  Employee voices consolidate into confirmed profiles over several days. Visitor voices are discarded again every single day. No customer profile is ever built and no recognition carries beyond the day, just the here and now of the sales conversation.
- A two-track pattern of the last two minutes

  ### A vector of zeros and ones. That is all there is.

  The service keeps a rolling window of the last 180 seconds, encoded second by second: employee, customer, both or silence. That is fully sufficient for fraud detection, and nothing more ever leaves the device.
- The POS asks, the service answers

  ### Every finalized receipt carries the conversation pattern.

  When the POS stores a receipt or triggers a sensitive operation, it queries the service and receives the conversation pattern of the last two minutes. One HTTP call, an answer in milliseconds, no extra hardware at the counter.
- Silent returns turn red

  ### Cash going out with nobody talking: that stands out.

  A normal sale has a pattern: the cashier speaks, and usually the customer does too. A return with a cash payout where nobody said a word is unusual. Exactly these transactions are flagged as suspicious the moment they happen, instead of surfacing only at day-end close.
- Four-eyes principle, verified acoustically

  ### Two people approving? Then two voices should be heard.

  Wherever two employees jointly approve a transaction, the system checks whether two known voices from the staff pool were actually heard. If it hears only one, or none, the four-eyes principle was probably satisfied on paper only.
- Learns on its own, from day one

  ### No voice samples, no per-person setup.

  The system learns from live operation. After the first few receipts it knows the regular crew at the POS and is fully operational. Staff changes are picked up automatically, with no profiles for anyone to maintain.
- Analysis stays at the POS station

  ### No audio stream. No path to the cloud.

  Detection runs locally on the device. No audio signal leaves the POS and no stream goes to the cloud. What leaves the service is the finished assessment as a numeric pattern, nothing else.
- Receipt-level context for loss prevention

  ### Zero in on suspicious receipts.

  The conversation pattern travels with the transaction and is available to loss prevention and audit later: voids, price changes, cash payouts, manager approvals. The reviewer sees whether anyone spoke, without ever hearing what was said.
- ### Where does your shrinkage happen?

  We demo SUMWARE Fraud Detection live on your specific risk: silent returns, void patterns, cash payouts or four-eyes approvals. GDPR-compliant, no recording, in 60 minutes.

  [Book a demo →](https://sumware.at/en/#contact)

---

*HTML version: https://sumware.at/en/solutions/fraud-detection/ · Site index: https://sumware.at/llms.txt · Sitemap: https://sumware.at/sitemap-index.xml · API: https://sumware.at/openapi.json · Developer docs: https://sumware.at/docs/*