---
title: 'Danske Bank: Machine-learning and deep-learning transaction fraud scoring'
slug: danske-bank-ai-fraud-monitoring
stable_id: a3327c8f8ced4e18
company: Danske Bank
function_code: financial_crime
pattern_codes:
  - continuous_decisioning
evidence_strength: reported
publication_tier: showcase
freshness: watch
reviewed_at: '2026-08-23'
updated_at: '2026-08-22'
source_quality_summary: 1 independent, 1 vendor; publication outcomes are reported.
caveat_summary: >-
  Outcome figures originate from Danske executives and Teradata materials; independent
  Forbes reporting quotes the same executive, not a separate audit.
collections: []
bundle_version: 1.0.0
bundle_fingerprint: sha256:c23c6cc2b88153d008ea8fda928f632ce0011fc2d4c5036672a16e5d895bab93
canonical_url: https://brianletort.ai/transformations/danske-bank-ai-fraud-monitoring
---

# Danske Bank: Machine-learning and deep-learning transaction fraud scoring

That handcrafted rules are the primary way to identify digital payment fraud.

Function: Financial crime. Patterns: Continuous decisioning. Evidence: reported.

Freshness: watch. Reviewed: 2026-08-23. Updated: 2026-08-22.
Watch status: verify the cited source and deployment condition before reusing this case.

Source quality: 1 independent, 1 vendor; publication outcomes are reported.

## Before

1. **Rules engine** — Apply human-authored fraud rules to transactions. (control: Static rule thresholds)
2. **Fraud investigator** — Review up to 1,200 alerts per day, most of which are false positives. (control: Human investigation)

## After

1. **ML and deep-learning engine** — Score millions of transactions in real time using learned patterns and latent features. (control: Sub-300ms scoring target)
2. **Fraud investigator** — Investigate the smaller, higher-yield alert queue and decide customer or enforcement action. (control: Human adverse-action authority)

## Decision rights

The model prioritizes and scores; investigators determine whether activity is fraudulent and what action to take.

## Exception path

Investigators review alerts, use model explanations, and escalate or clear cases; the public sources do not disclose automated blocking rights.

## Outcomes

- **False-positive alerts and true-positive detection** (reported): Up to 1,200 false positives per day; 99.5% of investigated cases not fraud; about 40% fraud detection → Vendor case study reports 60% fewer false positives and 50% more true positives. Outcome figures originate from Danske executives and Teradata materials; independent Forbes reporting quotes the same executive, not a separate audit.

## Executive lesson

The transformation metric is investigator yield, not model accuracy in isolation; a lower false-positive queue changes how scarce human judgment is allocated.

## Anti-pattern

Automating account action from an opaque score or presenting vendor-reported uplift as independently audited.

## Questions for leaders

- What is the residual false-negative risk?
- Which actions require human confirmation?

## Sources

- [Danske Bank Uses Tech To Prevent Digital Fraud](https://www.forbes.com/sites/tomgroenfeldt/2017/10/30/danske-bank-uses-tech-to-prevent-digital-fraud/) — Forbes
- [Danske Bank Fights Fraud with Deep Learning and AI](https://assets.teradata.com/resourceCenter/downloads/CaseStudies/CaseStudy_EB9821_Danske_Bank_Saves_Millions_Fighting_Fraud_With_Deep_Learning_and_AI.pdf) — Teradata