Fraud & Risk Pattern

Detect Fraud Rings in Real Time

A payment can look legitimate on its own. Memgraph keeps accounts, transactions, and more linked in a live graph as events arrive, so your fraud stack can query multi-hop patterns and add graph-derived signals to existing rules and models.

Detecting fraud ring patterns

Three accounts share one identity and device, then transact with the same merchant within ten minutes.

Shared entitiesLinked accountsRecent transactionsCommon merchantRisk signal

Graph-enhanced fraud detection pipeline

Kafka / Pulsar events

Ingest events + update graph

Trigger runs the Cypher pattern match

Apply fraud rules & ML scoring

Alert or investigation

Each incoming event updates the graph. A Memgraph trigger fires when that commits and runs the fraud-ring match against the latest relationships, with no polling and no separate service. Matches are flagged in the graph and feed your existing fraud rules or ML scoring.

Fraud-ring match criteria

Inside the Cypher matching step

Time window

The other transactions in the fraud ring fall within 10 minutes before the new transaction.

Graph pattern

(a:Account)-[:REGISTERED_TO]->(i:Identity)
(a:Account)-[:USES]->(d:Device)
(a:Account)-[:MADE]->(t:Transaction)-[:AT]->(m:Merchant)

Match condition

At least three distinct accounts a share the same i, d and m.

Query output

Mark the matched Identity, Device, Merchant and Account nodes, or create an :Alert node, for downstream scoring or investigation.

Results in production

Capitec Bank

Banking · Payment fraud

3.5M+

records scored daily

2.1

false positives per confirmed fraud case, monthly average

Read the full success story

Leading US Insurer

Insurance · Claims fraud

135%

increase in fraud detection efficiency

$1M+

in previously missed fraud on the same datasets

Read the full success story

See how teams add graph signals to their fraud stack.

Explore Fraud & Risk Detection