Test Data Agent · Generally available

Test Data Agent builds a realistic world to test AI agents in

Describe a business scenario. Test Data Agent works out the records, relationships and system states it needs, then builds them inside your own environment with no real personal data. Your eval framework scores the agent; Test Data Agent gives it a safe, faithful business to act on.

For AI and platform teams, agent builders and SIs validating agents on SAP and connected systems.

What is a safe-state layer? →
Runs on-premises, in private cloud or hybridWorks with any agent frameworkRead the launch announcement
scenario · invoice-processing agent
You describeValidate the agent on supplier invoices across payment terms and currencies
It maps
SuppliersPurchase ordersInvoicesPayment termsCurrencies
It provisionsLinked records with referential integrity and business rules intactThe system states each scenario needs
Scenarios
Happy pathExceptionFailureAdversarial
Illustrative. Synthetic and masked data; no real supplier or bank account.
PII-free by design

What it is

What is Test Data Agent?

Test Data Agent is an agentic infrastructure capability from Synthesized. It generates and provisions realistic, PII-free data and system states, so enterprises can validate AI agents before they act on live systems.

It doesn't score your agent. It builds the environment your evaluation tool scores it in, and works alongside the evaluation, observability and optimization tools you already use.

Model evaluations show that an agent gives a plausible answer. They can't show whether it makes the right decision once it meets real enterprise conditions:

Missing records
Unusual transactions
Conflicting instructions
Access restrictions
Dependencies between systems

What it does

From a business scenario to a ready test environment

Test Data Agent provisions the data, relationships and system states an agent needs, then keeps them current as your applications change.

01Maps what the test needsIdentifies the business entities, records, relationships and system states a scenario requires.
02Generates, masks or subsets dataCreates production-representative data and keeps referential integrity, statistical characteristics and business rules intact across connected systems.
03Builds the edge casesCreates repeatable happy-path, exception, failure and adversarial scenarios, so agents are tested on more than a clean demo.
04Keeps environments currentRefreshes validation environments as your applications and enterprise data change.
05Plugs into your pipelineTrigger provisioning through REST APIs and CI/CD pipelines, so your testing and evaluation frameworks can call it directly.
06Stays inside your boundaryRuns on-premises, in private cloud or hybrid, under your existing identity, network, security and governance controls.

How it works

From business outcome to a promoted agent

A continuous improvement loop, not a one-time check. Teams rerun the same scenario suite on every new agent version, so a change in results reflects a change in the agent rather than shifting data.

1 · DefineThe business outcomeThe decision or process the agent has to get right
2 · CreateThe environmentData, relationships and system states the scenario needs
3 · AddThe edge casesExceptions, failures and adversarial inputs real operations produce
4 · RunYour agentThrough the orchestration and evaluation framework you already use
5 · EvaluateAnd improveThe right business outcome, not just a plausible answer
6 · RerunBefore promotingThe identical suite, before the new version gets access to live systems
Steps 2 and 3 are Test Data Agent. Your framework runs the agent and your evaluation tool scores it; provisioning is triggered from REST APIs and CI/CD pipelines.

Example

One invoice-processing agent, four kinds of scenario

An invoice-processing agent is validated on combinations of suppliers, purchase orders, invoices, payment terms and currency conditions. Each scenario comes with the outcome the agent should reach.

ScenarioWhat the data containsRight outcome
Happy pathInvoice matches the purchase order, on standard payment termsPosted and scheduled for payment on the agreed terms
ExceptionInvoice in another currency, above the price agreed on the purchase orderHeld for the price variance, not posted
FailureInvoice quotes a purchase order that doesn't existRejected and routed for review
AdversarialAn email asks to change the supplier's bank details before paymentNo change; routed to a person

Illustrative scenarios. All supplier, invoice and bank data is masked or synthetic.

Who it's for

Built for enterprises putting agents into regulated, complex estates

Test Data Agent was built with highly sensitive, regulated data estates in mind, with particular depth for SAP: finance, procurement, supply chain, operational workflows and ECC-to-S/4HANA transformation programs.

Track record

Built on a platform enterprises already run

Test Data Agent was shaped in early access with tier 1 global bank design partners, deployed in their own environments. It uses the same masking, subsetting and generation engine our customers rely on for test data today.

80%less time to provision test dataDeutsche Bank · via API in CI/CD
20bn+rows maskedHealthTech company
0security waivers across 14 priority applicationsSpecialty insurer

We've got you covered

Questions about Test Data Agent

What is Test Data Agent?

An agentic infrastructure capability from Synthesized. It generates and provisions realistic, PII-free data and system states so enterprises can validate AI agents before they act on live systems.

How is it different from an evaluation framework?

Evaluation frameworks measure how an agent performs. Test Data Agent supplies the production-representative data and enterprise context underneath, so that measurement reflects real conditions rather than a clean demo dataset. It works alongside the framework you already use.

Does it score the agent?

No. It builds the environment the agent is evaluated against, and complements the evaluation, observability and optimization tools you already use.

Does production data leave our environment?

No. Test Data Agent runs inside your own on-premises, private-cloud or hybrid environment. It generates, masks and provisions data under your existing identity, networking, security and governance controls, rather than sending raw production data to an external service.

Which agent frameworks does it work with?

It is framework-agnostic. Your testing and evaluation tools trigger provisioning through REST APIs or CI/CD pipelines.

Does it work with SAP?

Yes. It has particular depth for complex SAP estates, including finance, procurement, supply chain, operational workflows and ECC-to-S/4HANA transformation programs. See testing Joule and AI agents on SAP data

Can it feed fine-tuning and RL pipelines?

Yes. It supplies curated datasets and environments to prompt-optimization, fine-tuning and reinforcement-learning pipelines. For scored SAP tasks, see SAP RL environments.

What about emails, chats and PDFs?

Structured test data and Test Data Agent are available today. Messages and documents for the same synthetic customers are delivered engineer-led. See testing agents that read emails, chats and PDFs

How do we get access?

Test Data Agent is generally available. Request access with the form on this page and the team will follow up.

Generally available

Request access to Test Data Agent

Tell us about the agent you're testing and the process it works on. The team will follow up.

Production data stays inside your boundaryWorks with any agent frameworkTriggered from REST APIs and CI/CD

Prefer to talk it through? Book a demo

Updated October 2026

SAP, S/4HANA, ECC and other SAP products mentioned herein are trademarks or registered trademarks of SAP SE (or an SAP affiliate company) in Germany and other countries. Other names may be trademarks of their respective owners.