Book a Demo
BlogIdentity Fraud

How a global payments platform reduced account-takeover reviews by 90%

A thin file is not a sign of fraud – but it leaves less to decide on. Adding cross-validated identity evidence changed which cases needed a human, and how quickly legitimate applicants moved through.

Joy Phua6 min read

Generative AI has made identity fraud more convincing and easier to scale. At the same time, many legitimate applicants arrive with little conventional history – new to a country, new to credit, or simply light users of traditional financial products.

For a global payments platform, those two trends met in the same place: the manual-review queue for account-takeover and thin-file cases. This article looks at what changed when the platform added broader identity evidence to its existing controls, and what fraud and risk teams can take from it.

Why thin-file applicants are hard to decide on

A thin file is not suspicious in itself. The difficulty is that it gives conventional controls little to confirm. When bureau and identity data are sparse, a fabricated identity and a genuine newcomer can look similar – and the safest available action is often to send both to a person.

That is where review queues grow. Analysts spend time on legitimate applicants who could have been approved, while the cases that genuinely need attention wait in the same line.

Evidence that fabricated identities struggle to sustain

Heka analyzed live external signals alongside the platform’s existing identity and risk controls. The value is less in any single signal than in whether they agree: a real person tends to leave a consistent footprint across independent sources, while a fabricated or manipulated identity finds that consistency difficult to maintain.

  • Digital historyHas this identity’s digital footprint developed over time?
  • Cross-platform presenceDoes the person appear consistently across independent platforms?
  • Contact intelligenceDo the email and phone belong to this person, and for how long?
  • Behavioral consistencyDoes the identity behave the way an established one would?
  • Cross-source alignmentDo independent sources agree on who this person is?
  • Breach exposureWhere relevant, has this identity’s data appeared in breached datasets?
Fig. 1 – The six signal groups and the question each helps answer.

How the added evidence changed decisions

With more to go on, fewer cases needed a person to decide them. The additional evidence helped the platform:

  • Make more confident thin-file decisions
  • Identify identity risk that legacy layers did not surface
  • Reduce the number of cases requiring manual review
  • Accelerate legitimate applicants
  • Intervene earlier in higher-risk cases

Manual-review volume, before and with Heka

Bars show relative volume. Customer evaluation result.
From thin file to decision
  1. Thin-file applicantLimited conventional data to decide on
  2. Heka identity evidence addedIndependent signals checked for consistency
  3. Faster, more confident decisionLegitimate applicants move on; higher-risk cases are prioritized
Fig. 2 – Additional evidence let legitimate applicants move on and focused review on higher-risk identities.

The results

The evaluation found a 90% reduction in manual reviews and a 99% reduction in loss exposure. Each blocked account saved $5K.

The first two figures belong together. A smaller review queue alone could mean a team is simply looking at less. Paired with lower loss exposure, it shows that the cases leaving the queue were the right ones to release – and that the ones remaining deserved the attention.

Customer evaluation result. Performance varies by portfolio and use case.

“We’ve worked with multiple vendors in the past, and Heka is the first one we encountered that actually provides the incremental value we’re looking for.”ATO Data Science Team Lead

What fraud teams can learn

  1. A thin file is a data problem, not a risk verdictTreating limited history as suspicion fills queues with legitimate applicants.
  2. Consistency is harder to fake than any single attributeFabricated identities can supply valid details; sustaining them across independent sources is harder.
  3. Measure review reduction alongside loss exposureA shorter queue only matters if losses fall with it.
  4. Use evidence to route, not just to flagThe value is in deciding which cases need a person – and which do not.
  5. Complement the existing stackThe platform’s existing controls stayed in place; Heka supplied what they did not surface.

Questions to ask in your own portfolio

  • What share of your manual reviews are thin-file applicants who are ultimately approved?
  • Which attributes can you confirm independently for applicants with limited history?
  • Do the cases your analysts escalate correlate with confirmed losses?
  • How long do legitimate thin-file applicants wait for a decision today?
  • What evidence would let you release a case without review?

Conclusion

Manual review is often treated as the cost of caution. For thin-file applicants, it is frequently the cost of missing evidence. When independent signals can confirm – or contradict – an identity, review becomes a place for the cases that genuinely need judgement, not a holding pen for everyone the data could not describe.

View the Global Fintech customer story
Book a Demo