Skip to main content

Evidence Framework

TL;DR
  • Tier 1 = Direct evidence to verify: Confirmed unauthorized use or identity misuse tied to the case, not a device or IP match alone
  • Tier 2 = Supporting signals: New contact details, device anomalies and unusual behavior need context
  • Decision: Use signals to route investigation. Classify fraud only when the evidence supports it; unresolved stays unknown
  • Write it down: What you found, where it came from and why you decided, the day you decide

Deciding whether a suspicious order is fraud? Sort what you've got into two tiers. Neither a signal count nor a shared identifier proves that this customer committed fraud. Rows on SSNs, bureau data and never-pay apply if you lend or extend credit. Skip them if you don't.

Overview​

Not all fraud signals are equal. Sort them by confidence and you'll call the same case the same way twice.

Core Principle

Tier 1 = Direct evidence, verified against the case
Tier 2 = Supporting signals, checked for legitimate explanations

Tier 1 Indicators​

Check what each source actually shows about this case. A data error or a stolen identity record doesn't by itself make the applicant or this order fraudulent:

IndicatorHow to Confirm
Identity details conflict with official recordsVerify the record, correct errors and offer a secure escalation path
Identity confirmed as fraud victimPolice report, affidavit, bureau alert
SSN belongs to deceased individualBureau data, death records
Submitted SSN, name and date of birth don't matchAn authorized SSA route, such as eCBSV, with written consent and eligibility. It can return a death indicator. Resolve the specific mismatch, because a no-match alone doesn't prove identity fraud
Document forensically invalidDocument verification technology
IndicatorHow to Confirm
Device/IP linked to confirmed fraudVerify the link to this case; shared devices and networks can include legitimate users
Address used by known fraud ringConsortium data, internal analysis
Exact application data matches confirmed fraudCompare the application's name, address, phone, email and device with the confirmed case; a specialist or your fraud tool does this at scale
Account explicitly claimed by identity theft victimAffidavit, manual review

When direct evidence establishes unauthorized use in this case, follow your incident procedure and document the reason. If it only points to an anomaly, keep investigating. Don't turn a match into a fraud label.

Tier 2 Indicators​

Supporting signals can justify investigation. Several correlated signals may describe one legitimate circumstance, so adding them up doesn't prove fraud. The weights and counts below are illustrative settings for routing work. Test them on cases you've resolved as fraud or legitimate, and on the unknowns, all old enough for the outcome to have come in.

Application Signals​

IndicatorWeightNotes
Email first seen recently, on your provider's observed-age field (AtData is one)⚠️ MediumA first-seen date isn't a creation date, and new customers have new addresses too
Phone recently ported⚠️ MediumSIM swap indicator
Address mismatch (stated vs. bureau)⚠️ LowMay be recent move
Employment unverifiable⚠️ MediumCheck method matters
Income stated >> bureau income indicators⚠️ MediumCould be recent change

Behavior Signals​

IndicatorWeightNotes
Never-pay (0 payments from origination)⚠️ HighNot conclusive (see first-party fraud)
Bust-out pattern (utilization spike)⚠️ HighMay be financial hardship
Device seen on prior fraud (1-2 cases)⚠️ MediumCould be shared device
Velocity anomaly⚠️ MediumContext dependent
IndicatorWeightNotes
Same phone on multiple identities⚠️ MediumCould be family
Same device on multiple identities⚠️ MediumCould be shared device
Address velocity (for example, 3+ apps, same address, 30 days)⚠️ HighCan point to a ring. Could also be a shared building or household

Decision Matrix​

Classification Rules​

EvidenceClassification
Verified evidence of unauthorized use tied to this caseConfirmed fraud, with rationale documented
Identity or record anomaly not yet resolvedInvestigation; keep the outcome unknown
Multiple supporting signalsPrioritize review using case value, risk and capacity
One supporting signalCheck its reliability and legitimate explanations
No listed indicatorsUnknown until you find the cause. No signals doesn't prove credit loss

Example Scenarios​

Scenario A: Clear Fraud

  • The verified account holder confirms unauthorized use of this specific transaction through a trusted channel
  • Classification: FRAUD, with the confirmation and case link documented

Scenario B: Credit Loss

  • Verified employment, verified income
  • No velocity anomalies
  • No device/address flags
  • Customer payment pattern consistent with financial hardship
  • Classification: CREDIT LOSS, on the hardship pattern you verified. The missing flags alone wouldn't prove it

Scenario C: Needs Investigation

  • Email first seen recently (Tier 2)
  • Never-pay pattern (Tier 2)
  • Employment unverifiable (Tier 2)
  • Action: Gather more evidence, time-box decision

Scenario C is the one that eats your team. Time-box it before you open it, because an investigation with no deadline becomes a case nobody closes.

Documentation Requirements​

Every fraud classification needs a paper trail. Here's what goes in it:

  1. Indicators present: List all relevant signals
  2. Evidence sources: Where each signal came from
  3. Decision rationale: Why classification was made
  4. Reviewer: Who made the decision
  5. Date: When decision was made

Write it the day you decide. A rationale reconstructed six months later is a guess with a date on it.

Next Steps​

Setting up evidence framework?

  1. Define Tier 1 indicators: High-confidence signals
  2. Define Tier 2 indicators: Supporting signals
  3. Create decision matrix: Classification rules

Investigating a case?

  1. Check direct evidence: Verify what it proves about this case
  2. Assess supporting signals: Check context, not a fixed count
  3. Manual review: Human investigation for complex cases

Documenting decisions?

  1. Review documentation requirements: What to record
  2. Use example scenarios: Apply to your case
  3. Classify per decision matrix: Make the call

Something wrong on this page? Email matt at paymentsandrisk.com and say what needs fixing.