Use test calls and real call records to check what your voice AI agent says, what information it collects, and whether its configured actions work. Review the result, provide specific feedback, correct the relevant settings, and test again before relying on a new or changed configuration.
1. Plan a few test calls
Choose scenarios that match the agent’s role and the capabilities you enabled:
- A routine question: check the answer against approved Knowledge and property information.
- An intake request: provide the details the agent should collect, then check the resulting card, board, stage, and fields.
- A question outside its knowledge or role: check how it handles the missing information or need for a person.
- A transfer, if configured: check the intended destination and the unanswered-call route.
Use your own approved test contacts and destinations. Do not place test calls to emergency services or utility emergency lines.
2. Review the call record
- Open AI Agents → Activities. You can also open an agent’s details and review Recent Activity.
- Find the call made to your agent and open the conversation.
- Review the transcript, summary, and outcome.
- Compare what happened with the expected answer or action. Check any resulting card or transfer separately rather than relying only on the summary.
3. Rate the call and explain the correction
In the call list, select the rating icon at the end of the row:
- Happy face: the agent handled the call well.
- Neutral face: the response could be improved.
- Sad face: the agent handled the call incorrectly and needs guidance.
When prompted for feedback, explain what the agent did and what it should have done instead. The sad-face rating prompts you to describe the correction.
For example: “The agent said this property was available, but it is no longer listed. Check the current published listings before offering it to a prospect.”
For a successful call, note what worked, such as: “Collected the unit and issue details accurately and created the request on the correct board.”
Call ratings automatically change the agent and also surface to the Aptly team for review. Review subsequent responses to check the result; a rating does not guarantee a particular answer on the next call.
4. Correct the right information or setting
A rating is useful feedback, but also check the source of the problem:
- Incorrect reference information: review the agent’s Knowledge and relevant property information. Add or correct the approved facts it should use. See How to Set Up AI Agent Knowledge.
- Missing or incorrect card details: review Data Collection Settings and Card - Create/Update, including the destination board, stage, and collected fields. See Create or update cards with an AI agent.
- Wrong transfer or unexpected voicemail: review the transfer rule, destination, and actual voicemail route. See How to Set Up Transfer Capability with an AI Agent.
Knowledge supplies reference information for voice conversations. It does not configure card creation, field updates, or call transfers. Change those actions in the relevant settings and capabilities.
5. Retest and keep reviewing
Repeat the scenario after a correction, then try a different wording of the same request. Review the new call record and any resulting action.
Before relying on the agent, confirm that:
- Its answers match the approved information.
- Required caller details are collected and saved where intended.
- Configured transfers and unanswered calls follow the expected route.
- Your team knows who will review requests that need human follow-up.
Continue reviewing and rating calls after launch, especially after changes to Knowledge, prompts, data collection, or capabilities.