Testing Maestro processes

Testing approach

Testing Maestro processes rests on three principles:

1. Keep what works: Test the RPA workflows inside your Maestro process the same way you've always tested them.
2. Test each part separately: RPA, agents, and human tasks each have their own test patterns. Testing the parts individually makes it easier to find what broke when something goes wrong.
3. Test the whole process: End-to-end tests check that the steps connect correctly, that data flows the way you expect, and that the business outcome is the right one.

Maestro processes differ from RPA workflows in ways that shape this approach. RPA workflows run in seconds, and the same inputs always produce the same outputs. Maestro processes can run for minutes, hours, or days. They combine RPA, agents, and human steps in a single flow. They depend on external systems you don't control.

Maestro processes are becoming directly testable, with tooling built for them. The improvements below are how this gets there.

What you can do today

RPA workflow testing: RPA workflows used in a Maestro process can be tested with Test Cases in the RPA project, the same way as for any other RPA workflow. Test inputs verify that the output matches expectations. No Maestro-specific tooling is required.

Agent testing: Agents used in Maestro processes can be tested with eval sets in Studio Web. An eval set is a list of test inputs paired with the responses you expect from the agent. It runs the agent against each input and checks how the actual response compares to the expected one. This matters most for processes where agents make important decisions.

End-to-end process testing: A Studio Test Case can run a published Maestro process with prepared test data, then check the final output. For anyone new to testing Maestro processes, this is the most important place to start.

Debug runs: Debug run in Studio Web is the right tool for checking how a process behaves during authoring. It gives step-by-step traces, real-time variable updates, and a history of past debug runs.

How to build your Maestro testing practice

Build coverage in order:

1. Start with end-to-end tests on your most critical processes: One test per critical business outcome. These are the tests that prevent your most serious problems from reaching production.
2. Add agent tests on processes that rely on agents: Agent responses vary more than other parts of the process. Test them separately so you can catch agent-specific problems early.
3. Expand to handle failures and edge cases: Add tests for invalid inputs, failed steps, and other situations where things go wrong. Add tests for RPA workflows used across several Maestro processes.

The harder parts of testing Maestro

Four challenges are common. Plan for them upfront.

Long-running processes

Maestro processes can run for hours or days. The same process can take 30 minutes in production and 2 minutes in test, depending on the data and the responsiveness of external systems. That makes timing-based assertions unreliable. Test what the process produced and which steps it ran, not how long it took.

External system dependencies

End-to-end tests connect to real external systems. That means tests can fail for reasons that have nothing to do with your process: a connection timeout, a third-party rate limit, a system being down. Use a sandbox or test version of those systems where possible. If a connection fails randomly during development, treat it as a connection problem, not a real test failure.

Agent variability

Two runs of the same agent with the same input can produce different outputs. Equality assertions will fail intermittently. Eval sets are designed for this. Use comparison rules that allow for reasonable variation rather than requiring an exact match.

Human-in-the-Loop (HITL) steps

Tests need to run end-to-end without human intervention, but Maestro HITL steps wait for a real person to act. There is no built-in way to inject a stand-in response. The right workaround depends on how your HITL steps are set up. Your account team can help with the specific approach.