Skip to main content
You can use the Test tab to validate an agent’s responses before (and after) going live. Here, you create short, targeted test cases that simulate a snippet of conversation and check whether the agent’s reply matches an Expected Output using a Similarity evaluator. You can run a single case or your whole suite, tune thresholds, and quickly debug failures. Testing Interface

​
Inside a Test Case (Editor)

  • Name the case (top-left). Keep it action-oriented: Identify caller, Respect DNC, etc.
  • You can either setup the conversation yourself or optionally, select a previous call to seed messages from a real transcript. This transcript can also fully be edited for the test.
  • Build the Conversation with short turns:
    • user → what the human said
    • bot → what your agent already said (context)
    • user → the prompt that triggers the behavior
  • Add one final bot turn and toggle Expected — this is the target reply used for scoring.
  • Use Add Message to insert turns; keep to 2–4 lines total.
  • Click Get AI Response to preview what your current agent would say (helps tune the Expected Output).
  • Save, and then your test case is ready to run.
Example: 
user: Strada Insurance. Amy speaking.
bot: Hi Amy! Is Stephanie Voltz available to chat?
user: Can I ask who's calling?
bot (Expected): Of course! This is <agent>, a virtual assistant with <brand>. I'm following up on an email we sent recently.
Tips:
  • Assert one behavior per case.
  • Avoid volatile details (dates, IDs).
⌘I