Skip to content

Check Detect at a Top 10 North American Bank

valid_banner_top10_square

Industry

Financial Services

Challenge

At a Regional client bank fraud detection system was creating an unintended consequence holds placed to protect against check deposit fraud were alsolocking funds for tens of thousands of legitimate customers. The bank lacked a reliable way todistinguish high-risk deposits from low-risk ones — resulting in widespread friction for goodcustomers while only partially containing fraud.

Results

Check Detect gives banks the precision to protect their balance sheet without punishing customers. By applying item-level risk scoring at the moment of deposit, REGIONAL BANK can now make targeted hold decisions that maximize fraud protection while minimizing disruption to legitimate customers.

Key Product

Check Detect

46%
Reduction in False Positive Hold
55%
Additional Fraud Exposure Idnetified
3
Channels (ATM, Mobile, Teller)
1520096943880

The False Positive Hold Problem

Across all three deposit channels, the vast majority of holds were placed on accounts where the check was ultimately not returned as fraud.

Channel

Accounts w/ Hold False Positive Holds False Positive Rate
Teller 21,796 20,245 92.9%
ATM 5,500 3,941 70.9%
Mobile 73,800 65,160 88.3%
 
$36.8M in fraud exposure sat under hold — representing 51.1% of total fraud exposure. Yet only 7.1% of those holds caught actual fraud.

 

Root Cause

Without a reliable risk score at the item level, REGIONAL BANK's hold policy was a blunt instrument — uniform rules applied across all deposits regardless of actual risk. This led to two simultaneous failures:

• Over-holding: Low-risk accounts receiving holds, damaging customer experience
• Under-protecting: High-risk deposits sometimes cleared without intervention

The Solution

Valid Systems deployed Check Detect — an XGBoost-based real-time loss alerting model — across Regional's Teller, ATM, and Mobile channels. Each incoming check
deposit receives a real-time risk score, enabling REGIONAL BANK to apply holds precisely
where risk is highest and release holds where risk is low.

Model Performance (Test Period: Jan–Dec 2025)

Channel AUC $   KS $ Gini $
Teller 0.971 0.849 0.942
ATM 0.992 0.925 0.983

 

Fraud Dollars Captured at Top Score Thresholds

The model identifies the highest-risk deposits with exceptional precision. By reviewing only the top 1% of highest-scoring items, REGIONAL BANK captures:
• Teller: 76.5% of all fraud dollars
• ATM: 92.9% of all fraud dollars

Score Threshold (Top %) Teller — % Fraud $ Captured ATM — % Fraud $ Captured
0.5% 66.1% 73.7%
1.0% 66.1% 92.9%
1.5% 82.7% 95.8%
2.0% 87.1% 97.4%
2.5% 90.2% 98.2%

 

The Results

Applying recommended score thresholds delivers a significant reduction in false positive holds while risking only a minimal increase in fraud exposure per account — a highly favorable trade-off.

Teller & ATM Channels — False Positive Reduction Scenarios

Channel % of Accts
w/ Deposit
& Hold
Accts w/
Deposit &
Hold
% Holds w/
No Return
Accts w/
Hold & No
Return
Fraud
Exposure $
w/ Hold
Teller 1.7% 21.7k 92.9% 20,245 51.1%
ATM 2.7% 5.5k 70.9% 3,941 25.5%

 

Mobile Channel — False Positive Reduction Scenarios

Channel % of Accts w/
Deposit &
Hold
# Accts w/
Deposit &
Hold
%
Holds
w/ No
Return
Accts
w/ Hold
& No
Return
Fraud
Exposure
$ w/ Hold
Mobile 8.6% 73.8k 88.3% 65,160 55.0%

 

The Path Forward

Check Detect gives banks the precision to protect their balance sheet without punishing
customers. By applying item-level risk scoring at the moment of deposit, REGIONAL BANK
can now make targeted hold decisions that maximize fraud protection while minimizing
disruption to legitimate customers.

How Check Detect Works

  1. Real-Time Scoring — Every check deposit is scored by an XGBoost classifierthe Check
    Detect model at the moment of deposit across Teller, ATM, and Mobile channels.
  2. Item-Level Risk Assessment — The model evaluates account behavior, balance history,
    return patterns, and deposit characteristics to generate a precise fraud probability score.
  3. Targeted Hold Decisions — Holds are applied only to deposits scoring above defined
    risk thresholds, dramatically reducing false positives while maintaining fraud protection.

Key Takeaways

  • 40% reduction in false positive holds across Teller channel with score thresholds

  • $36.8M in fraud exposure under active hold monitoring — protecting 51.1% of total
    exposure

  • $7.2.8M is Charge off loss Reduction 

 

Ready to Reduce False Positive Holds at Your Bank?


Valid Systems deploys Check Detect for financial institutions of all sizes. Our check risk
models are trained on your data, validated against your fraud patterns, and deployed in
weeks not months.

validadvantage.com