SAP S/4HANA procurement fraud
SAP S/4HANA proof of concept that sends anonymised purchase-order risk indicators to Jev and classifies orders to release, approve or block.
Overview
This proof of concept, documented in Portuguese, assesses fraud risk in SAP S/4HANA purchase orders with Jev via OpenRouter. ABAP computes risk indicators from order, supplier and buyer data; a Fiori Elements app sends the anonymised indicators to Jev with seven typed questions; and ABAP receives the probabilities, applies thresholds and classifies the order as release, approval or block. Everything runs in shadow mode: evaluations are logged but no order is blocked.
The five Noul questions cover a fictitious supplier, payment diversion, overpricing, order splitting and internal fraud, plus a 1-5 atypicality score and a Choice suggesting an action for comparison only. A blocked supplier is a hard rule in ABAP and never goes to Jev. The final classification is always ABAP's: maximum risk below 0.20 releases, up to 0.60 goes to approval, above blocks, and the README calls these thresholds a hypothesis to calibrate. No names, supplier codes, users or order numbers are sent, and the README estimates about US$0.09 per thousand orders. Without a key, a local simulator is used.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | fornecedor_ficticio: Is this a fictitious supplier? | — |
| Yes / no | desvio_pagamento: Is there a payment diversion? | — |
| Score | atipicidade: How atypical is this order versus history (1-5)? | 1 · 2 · 3 · 4 · 5 |
| Choice | acao: Suggested action (for comparison only) | release · approval · block |
How it works
- ABAP computes anonymised risk indicators
- The Fiori app sends them to Jev with seven typed questions
- ABAP applies thresholds (0.20 / 0.60) to the maximum risk
- Classify as release, approval or block in shadow mode and log
Strengths
- No identifying data sent to the model
- Code owns the final classification
- About US$0.09 per thousand orders per the README
Caveats
- Proof of concept in shadow mode
- Thresholds uncalibrated
- Requires on-premise S/4HANA with RAP and OData V4
Getting started
Import the ABAP objects in layer order, run ZCL_POC_JEV_SETUP, then npm install in app/ and add an OpenRouter key (or use the local simulator).At a glance
- Type
- Project
- Status
- prototype
- Language
- ABAP
- Runtime
- SAP S/4HANA + Fiori Elements app (Node)
- Has code
- yes
Use cases and integrations
How curated lists describe it
ERP fraud: a proof of concept that flags suspicious purchase orders in SAP S/4HANA with a Jev classification.
yibie/awesome-jev
Frequently asked questions
What does Jev decide?
Probabilities for five procurement fraud patterns, an atypicality score and a suggested action.
Does it block orders?
No; it runs in shadow mode and only logs evaluations.
What data is sent?
Only anonymised indicators such as days and counts, not names or document numbers.
Related projects
Crypto exchange risk gateway where Jev triages deposits and withdrawals while hard rules and Java code make the final call.
ServiceprototypeJavaUiPath coded agent for AML alert triage where an LLM reads and writes but Jev makes every red-flag, risk and disposition decision.
AppprototypePythonWorkflow design article: one Jev Noul asks whether a domain name refers to a protected brand, with the TLD kept out of the input.
ArticleproductionFrench library ranking ADEME emission factors for an activity label, splitting code rules from Jev's semantic match.
ProjectprototypeJavaScriptEvent-stream layer labelling signups, logins and checkouts as human, AI agent, crawler or bot, with a bad-intent probability.
AppprototypeJavaScriptSpanish oncology decision-support prototype that walks ESMO guideline trees and uses Jev only for judgment calls.
ProjectresearchPython