Get the whitepaper!
The Test Data Imperative: Towards SAP Autonomous Enterprise

AI agents · Joule

Test Joule and AI agents on safe, realistic SAP data

AI agents read SAP data and act on it: they create orders, post documents and answer questions about customers and employees. Testing them needs realistic business data, many repeat runs and a clean state each time, without real personal data in prompts, logs or test systems.

Book a health checkSee the platform
Available on SAP Store
Synthetic data generation configured in the Synthesized platform
Same state every run
No real people in prompts
Built for agents
Data the agent can act on, then reset
  • Whole business processes, masked
  • Edge cases on request
Built for the SAP & Enterprise QA toolchain
SAP S/4HANASAP AribaSAP HANASalesforceUiPathTricentis

Why agents are different

Four ways AI agents change test data needs

01
They change dataAn agent that posts a goods issue or approves an invoice leaves the system in a new state. The next run needs the old one back.
02
They vary run to runThe same prompt can take different paths, so each test runs many times and the results are compared.
03
Data ends up in prompts and logsWhatever the agent reads can surface in prompts, traces and evaluation logs. Real customers shouldn't be there.
04
They cross systemsAgents call SAP and the systems around it, so test data has to match across all of them.

How it works

A masked baseline, reset before every run

Agent test suitePrompts and expected outcomes
Joule or your agentReads and posts documents
SAP test systemMasked, realistic data
SynthesizedResets the data state
Results and logsNo real personal data inside
fresh state before each run
next run
Agents change the data they act on. Every run starts from the same masked state, so results can be compared.
  1. 01

    Choose the scenarios

    Order-to-cash, procure-to-pay or the HR processes your agents work in, with the edge cases they must handle.

  2. 02

    Prepare a masked baseline

    Subset and mask production into the agent test system, or generate the records production doesn't have.

  3. 03

    Reset before each run

    Call the Synthesized API from your test harness, so every run starts from the same state.

  4. 04

    Compare runs

    Because the data is the same each time, differences come from the agent, not from the data.

Platform

Mask, subset and generate the data agents act on

MaskSubsetGenerate

Find and mask personal data

A scan flags personal data in SAP tables and the systems around them, then masking replaces it with realistic values, the same way everywhere.

  • Sensitive columns found by the scan, including Z-tables
  • Same replacement values in SAP and the CRM
  • A record of every run
Automated masking with a sensitive data scan report

Copy only what your tests need

Choose company codes, date ranges or business objects, and the subset keeps every related record so tests still run end to end.

  • Business keys and date ranges in scope
  • Document chains kept whole
  • Smaller, faster test systems
Automated subsetting with system statistics

Create the data production doesn't have

Generate customers, orders and edge cases for scenarios production doesn't contain yet, configured as code or in the UI.

  • Edge cases on request
  • Configuration as YAML or in the UI
  • Runs from the API, CLI or a pipeline
Automated generation configured as YAML

EU AI Act

Where the AI Act touches agent test data

For high-risk AI systems, the AI Act sets rules for the data used to train, validate and test them. Most finance and logistics agents won't be high-risk; AI used in recruitment or to manage employees can be.

Regulation (EU) 2024/1689 · Article 10

Data and data governance for high-risk AI

Training, validation and testing data sets meet quality criteria for their purpose.

They are relevant and sufficiently representative and, as far as possible, free of errors and complete.

Special categories of personal data are used only exceptionally, with safeguards, for detecting bias.

Paraphrased. The dates for high-risk obligations are under review; check the current timeline.

Read the source: Regulation (EU) 2024/1689 (AI Act) on EUR-Lex

What it means in SAP

HR agents in SuccessFactors
Recruitment and employee-management uses are listed as high-risk. Test them on anonymized employee data.
GDPR still applies
Testing agents on real personal data is processing under GDPR, whatever the AI Act says.
Keep the record
Each refresh records the rules applied and who ran it, next to your test results.

We've got you covered

Questions about testing AI agents on SAP

How do you test SAP Joule?

Joule works on the data in the SAP system it's connected to, so test it in a non-production system that holds masked, realistic data. Reset that data before each run and compare the results across runs.

Can we use production data to test AI agents?

Copying production into an agent test system puts real personal data into prompts, traces and logs. Masked data that keeps the business process intact gives the same coverage without that exposure.

Why reset the data between runs?

Agents change what they act on. Without a reset, the second run starts from a different state, and you can't tell whether a difference came from the agent or from the data.

Does the EU AI Act apply to our SAP agents?

Only if an agent counts as high-risk, for example in recruitment or employee management. Then it sets rules for training, validation and testing data. Your legal team decides, and the dates are under review.

Does Synthesized send our data to an AI model?

No. The masking and generation engine runs in your environment and gives the same output for the same input. The optional AI assistant can be switched off, so nothing calls out.

Next step

Give your agents data they can't leak

We prepare one masked agent test system for a process you choose, and show a reset between two runs.

Runs in your environmentNothing installed in SAPRead-only access to SAPSecurity and deployment
Book a health checkSecurity and deployment

Related SAP pages

AutomationTest data for SAP test automationTosca and UiPathProcessSAP order-to-cash test dataOrder-to-cashSecuritySecurity and deploymentSecurity
Updated October 2026

SAP, S/4HANA, SAP HANA, SuccessFactors, Ariba, Concur and other SAP products and services mentioned herein, as well as their respective logos, are trademarks or registered trademarks of SAP SE (or an SAP affiliate company) in Germany and other countries. All other product and service names mentioned are the trademarks of their respective companies.