Tacnode™

Fraud holds that reach authorization too late

Card authorization approves six purchases on an account already on hold, because the hold is still moving through a pipeline.

Photo: Philipp Katzenberger / Unsplash
→→At4:10
The story

An identity check flags a new application from account R1’s device.

  • Application R2 comes in from the same device as R1, an open card account at the lender.
  • The lender’s identity check links the two and flags R2 for the fraud team.
  • Card authorization checks account status before every purchase. It reads that status from a cache, which a pipeline keeps in step with the case system.
The authorization monitor
What’s happening underneath
readsposts approvalsCase systemCase managerR2 linked to R1’s devicePipelineDebezium CDC, Kafka, Flink jobnormal loadAuth cacheRedisR1: activeAI shopping agentFraud ringCard authorizationAuthorization serviceapprovingCard ledgerIssuer processorposting
1 / 9

Illustrative scenario. Timings show the shape of the problem, not a measurement of any one company.

Watch the walkthrough

Video
4:14
Card Fraud: Six Purchases Approved on an Account Already on Hold
Alex Kimball walks through this replay: a fraud hold placed in time reaches card authorization eight seconds late, and the same hold with current context.
Watch on YouTube

Your stack today, and with the Tacnode Context Lake™

Today

Case system
Case manager
Pipeline
Debezium CDC, Kafka, Flink job
Hold still in transit
Auth cache
Redis
Card authorization
Authorization service

Most lenders keep fraud decisions in a case system and card authorization in another, joined by a pipeline: change capture, a stream, a stream job, a cache. That was fine when fraud moved at human speed. Rings that open accounts with stolen identities now run AI agents that spend the moment an account is live, in bursts that last a few seconds. Authorization reads account status from the far end of the pipeline, and the agent spends during the delay.

With the Tacnode Context Lake™

Case system
Case manager
Tacnode Context Lake™
R1 on hold
Card authorization
Authorization service
  • The hold was readable once committedThe case system wrote it where authorization reads, with no pipeline in between.
  • Agents spend in secondsA few seconds between a decision and the system that acts on it is enough to empty a card.
  • Every decline was savedThe next check on that device or that ring reads all six.

Questions to ask your team

  1. 1When your fraud team places a hold, how long until card authorization declines on it?
  2. 2Have you seen bursts of card-not-present purchases on an account in the seconds after it was flagged?
  3. 3Do your fraud decisions and your authorization path read account status from different systems?