Concept card

AI for Banking — AML, KYC and Credit Scoring with SAP

As of 2026-09-27

Edited by Cédric Mary, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is AI for Banking?

Model-based alert scoring flips AML triage economics — the top 10% of scored alerts capture ~80% of confirmed SAR filings, against a 95-99% rules-based false-positive baseline.

What it is

AI in SAP-anchored banking deployments splits into three regulated use cases — anti-money-laundering (AML) transaction monitoring, Know Your Customer (KYC) due-diligence augmentation and credit-scoring models — each with a different SAP product mix and a different EU AI Act compliance profile.

The problem is that financial services firms run massive transaction volumes against rules-based systems that generate enormous false-positive rates: SWIFT and Wolsberg estimates put the AML false-positive rate at 95-99 % of flagged transactions, consuming analyst time with low-yield investigations. KYC re-certification is manual, paper-intensive and mis-aligned with risk — high-risk customers trigger less frequent review than rule frequency dictates. Credit-scoring models trained on historical book data underweight forward-looking behavioural signals available in SAP's customer-interaction data.

Why it matters

Three distinct use cases (AML monitoring, KYC augmentation, credit scoring) each need a different SAP product mix (FSN/FSDM, Joule dossier synthesis, SAC Predictive feature model) — not a one-size pitch.
Credit scoring on 24 months of S/4 payment behaviour yields a 15-20% Gini improvement over bureau-only models — a concrete, quotable uplift.
AML and credit-scoring models fall under EU AI Act Art. 6 + Annex III as high-risk systems, effective 2027-12-02 (Digital Omnibus, in force since 2026-07-27) — governance must be engineered in from day one.

Key points

Three regulated AI use cases on SAP: AML transaction scoring, KYC augmentation, credit scoring.
AML false-positive baseline: 95-99 % of flagged transactions require no action (SWIFT/Wolsberg estimates).
AML alert-scoring model: top 10 % of scored alerts contain ~80 % of confirmed SAR filings.
KYC augmentation: Joule synthesises counterparty dossier from FSDM + external registries in minutes.
Credit scoring: 15-20 % Gini improvement over bureau-only by adding 24-month S/4 payment behaviour.
EU AI Act Art. 6 + Annex III: AML + credit-scoring are high-risk systems — compliance deadline 2027-12-02 (Digital Omnibus, in force since 2026-07-27).

Terms used on this page

FSDM: Financial Services Data Model — SAP's industry-standard data model for banking and insurance, providing the canonical counterparty, account, and transaction object structures for analytics.
SAR: Suspicious Activity Report — a regulatory filing required when a financial institution suspects a customer transaction constitutes money laundering or terrorist financing. Filing rates are the primary AML outcome metric.
Gini coefficient (credit): Model discrimination metric for credit scoring — the area between the Lorenz curve and the random-classifier line. Higher = better rank-ordering of default risk. Industry benchmark: bureau-only ≈ 45-55 Gini; augmented-with-SAP-behaviour ≈ 60-70 Gini.
PEP: Politically Exposed Person — a category of high-risk customer in AML regulation; transactions involving PEPs require enhanced due diligence. SAP AML models join to PEP list feeds at scoring time.
UBO: Ultimate Beneficial Owner — the natural person who ultimately owns or controls a legal entity; mandatory disclosure in EU AML 5th and 6th directives. KYC augmentation agents automate UBO chain resolution.

Sources

Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks · the facts worth quoting.

Open in the app →