> For the complete documentation index, see [llms.txt](https://learning.contextqa.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://learning.contextqa.com/ai-agent-testing/agent-evaluation/end-to-end-outcomes.md).

# Test Post-Call and Multi-Agent Business Outcomes

Verify asynchronous post-call tasks and multi-agent handoffs with correlation identifiers, completion windows, read-only evidence, and separate assertions.

> **Documentation owner:** [Deep Barot](https://contextqa.com/academy/) · **Last updated:** November 12, 2026

## Quick answer

End-to-end agent testing verifies the full required business process, including work that happens after a conversation ends. Define each stage, its expected state, correlation identifier, and completion window. Use ContextQA conversation criteria and verifier tools where applicable, and evaluate asynchronous workers or downstream agents separately when the voice run does not capture their evidence.

## Key takeaways

* Conversation completion, transaction persistence, and post-call completion are separate assertions.
* Correlate evidence to the same interaction instead of selecting the latest record.
* Treat pending and missing downstream evidence explicitly.

## Model the workflow

For a receptionist, the process might be caller request → confirmed booking → stored appointment → post-call note → escalation or notification. A support process might be chat → specialist-agent handoff → authorized tool action → CRM update.

Create a stage table before implementing tests:

| Stage                | Expected fact                                         | Correlation                   | Evidence                  |
| -------------------- | ----------------------------------------------------- | ----------------------------- | ------------------------- |
| Conversation         | Final corrected intent and details confirmed          | Session/call ID               | Conversation and audio    |
| Primary action       | Exactly one correct authorized booking                | Booking ID and test reference | Read-only backend result  |
| Post-call processing | Required note created with the correct source context | Call ID linked to note ID     | Downstream record         |
| Escalation           | Required handoff reached its destination              | Case/handoff ID               | Destination state         |
| Notification         | Expected delivery state for the test destination      | Message correlation           | Provider or system record |

This table is a test design, not a promise of built-in connectors for every row.

## Configure checks by responsibility

Use a [Verifier](https://learning.contextqa.com/ai-agent-testing/agent-evaluation/verify-tool-actions.md) for read-only facts the judge must inspect. Use a Variable Source for values needed at run start, a Health check probe for readiness, and a Pre-Run Environment Hook only for authorized setup. Do not turn a verifier into a repair operation: creating the missing note during verification would invalidate the test.

If the voice run cannot capture the later task, create a separate case for the downstream worker with the conversation as an approved input and a specified expected outcome. Test the orchestration separately to establish that the real end-of-call event triggers that worker. Passing a worker in isolation does not prove the trigger works.

## Handle asynchronous outcomes

Record the call-end timestamp, expected trigger, task-start timestamp if available, and completion window. Poll or retrieve state through a supported integration with a bounded policy. The documentation does not prescribe an unverified polling API.

Use explicit result states:

* **Passed:** required evidence establishes completion within the policy.
* **Failed:** evidence establishes an incorrect or prohibited outcome, or a confirmed missed deadline under the policy.
* **Pending:** the allowed completion window has not elapsed.
* **Inconclusive:** evidence is insufficient to establish the assertion.

An absent record before the deadline may be pending. An absent record after the deadline is not automatically proof of a worker defect if collection itself failed. Distinguish missing business state from missing telemetry.

## Test retries and partial failure

Prepare cases where the primary action succeeds but its response is lost, the downstream task runs twice, one dependency fails, or the handoff loses a corrected value. Assert exactly-once business effects where required, even if multiple processing attempts occur. Use isolated fixtures and verify that cleanup affects only the test records.

For multi-agent systems, check the handoff’s intent, identity, authorization context, and final corrected values. Do not require or claim access to hidden chain-of-thought; observable decisions, tool events, and persisted state are the relevant evidence.

## Explain how the agent and application tests connect

Use agent evaluation to check conversation and decision behavior. Use [application tests](https://learning.contextqa.com/enterprise-application-testing/enterprise-testing.md) for relevant UI/API behavior, such as whether a created record appears in the business application. Correlate the two results deliberately. A passing browser check on an unrelated record does not validate the agent’s action.

Continue with [industry playbooks](https://learning.contextqa.com/ai-agent-testing/agent-evaluation/industry-use-cases.md) and [production analytics](https://learning.contextqa.com/ai-agent-testing/agent-evaluation/production-analytics.md) for domain scope and measurement.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://learning.contextqa.com/ai-agent-testing/agent-evaluation/end-to-end-outcomes.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
