Skip to documentation content
Agentplace
Browse documentation

Building your agent

Test and refine your agent

Use the preview to test complete customer tasks, including changes and failures. The Builder can test and improve what it creates, and you can ask for revisions in plain language. Your review should confirm the business result and connected actions, not only the appearance of the page.

Agentplace product documentation Updated

Start with a small set of realistic sessions

  • A typical customer who provides the expected information.
  • A vague request that needs clarifying questions.
  • A customer who changes a requirement after seeing a result.
  • Missing, conflicting, or unsupported information.
  • A disconnected app or an action that cannot complete.
  • Two customers whose private information must remain separate.

Check both the page and the action

Check the numbers, choices, explanations, and next steps against your rules. If the agent says it sent an email or booked a call, verify the corresponding result in the connected system.

Use suitable test data and test modes where the provider supports them. Do not treat a successful-looking preview as evidence that a real payment, booking, or handoff was completed.

A useful revision request

When a client lowers the budget, keep the original estimate visible next to the revised option and explain which items changed. Ask for confirmation before sending either version to my team.

Review before publishing

Confirm the intended audience, access rules, current knowledge, permitted actions, and handling of unresolved cases. Check the mobile page and voice flow if those are part of the experience.

After publication, test the actual shared link with the access level a customer will use. Revisit these checks when you change a connected workflow or an important business rule.

How can I prove the value before a larger rollout?

Start with one customer task and one main success measure. Prepare the relevant business materials, rules, and examples of real requests. Define what a useful result must contain before spending time on a wider rollout.

  1. Choose one outcome, such as a complete service inquiry, a useful preliminary estimate, or a qualified appointment request.
  2. Build the first version and test typical requests, exceptions, and incomplete information.
  3. Run a small pilot with the intended audience. Record completed outcomes, corrections needed, and actual usage for the selected model.
  4. Compare the result with your current process. Review completion rate, result quality, time saved, and cost per useful outcome before expanding.

A showroom pilot

Start with helping visitors prepare an appointment request. Check whether the agent captures the customer’s requirements and whether the showroom team can use the summary. Measure completed, useful requests and their actual cost. An increase in visits or sales is something to measure, not assume.