Testing

How to test agents using the built-in testing interface and automated Simulations.

Every Spara agent has a built-in testing interface that lets you interact with a draft before it goes live. Testing always runs against your current draft — not the published version — so you can iterate freely without affecting real leads.

The testing interface is accessible in two places:

  • In the Agent Editor — the right-hand panel, toggled alongside Ask Spara AI

  • On the Testing page — a standalone view for side-by-side comparison of multiple agents

The Testing panel in the Agent Editor showing a simulated chat conversation.

Test interface by agent type

Agent type
Test interface

Chat

Simulated chat conversation in Navigator or Smartbar

Voice

Simulated phone call (browser-based)

Email

Send a simulated incoming email, review the draft reply

Text

Simulated SMS exchange


Using the test interface

Running a test

Open the right-hand panel in any Agent Editor and start interacting. For chat agents, type messages as if you're a lead. For voice agents, click to start a test call. The agent responds using your current draft instructions.

Test conversations are isolated. Messages sent during testing do not appear in lead history, trigger workflows, or generate analytics events.

Reviewing a response

Each agent response can be inspected inline:

  • Lightbulb icon — Explains why the agent responded the way it did, citing specific instructions it followed.

  • Thumbs down icon — Flags the response as incorrect or unhelpful, and offers line-by-line suggestions for improving the prompt to prevent the issue.

Use these tools to understand model reasoning, especially when a response doesn't match your expectations.

Resetting the test

Click the refresh or clear button to start a new test conversation. Each test session begins with a fresh lead context — no prior messages, no previously captured fields.


Simulations

Simulations let you define automated test scenarios and run them in batch — a faster way to validate your agent against a set of known interactions without manually testing each one.

The Simulations tab showing three passing simulations and a 100% pass rate.

Creating a simulation

Each simulation has two parts:

  1. Simulated user behavior — A plain-language description of how the simulated lead should act. For example: "The lead is interested in enterprise pricing and wants to know if there's a volume discount."

  2. Success criteria — A list of outcomes the agent must achieve for the simulation to pass. For example: "The agent asks for the lead's team size before providing pricing details."

Simulations can be as narrow (testing a single objection) or as broad (testing a full qualification flow) as you need.

Running simulations

Simulations can be run individually from their detail view, or in batch from the Simulations list. Batch runs execute all simulations concurrently.

Each simulation run generates a pass or fail result. Failed simulations include an explanation of what went wrong and, where relevant, point to the specific instruction the agent didn't follow — or flag when the simulation's expected behavior conflicts with the agent's instructions.

When to use simulations

Simulations are especially useful when:

  • You've made a significant change to your instructions and want to verify it didn't break existing behavior

  • You want to test edge cases (unusual questions, hostile leads, out-of-scope requests) without relying on live traffic

  • You're onboarding and want confidence before going live


FAQ

Does testing affect my analytics or workflow triggers?

No. Test conversations are isolated from production. They don't create lead records, trigger workflows, or appear in analytics.

Can I test with a specific lead's context?

Not directly from the editor. You can manually set fields at the top of the test panel (e.g., first_name, company_name) to simulate a personalized experience. For more advanced testing with realistic lead data, use Simulations.

Can I run simulations against a published version?

Simulations always run against the current draft. If there's no draft, a copy of the published version is used as the starting point.

How do I use Ask Spara AI during testing?

Toggle to Ask Spara AI in the right-hand panel. It has access to the test conversation and can answer questions like "why did the agent respond this way?" or suggest prompt edits to fix a behavior you saw in testing.

Last updated