Release Notes: v3.4
Why it matters for your business
- Save money: automated grading replaces manual call reviews and catches missed revenue before it compounds.
- Work faster: pass rates and recovery statuses surface issues without digging through every transcript.
- Standardize service: define criteria once and apply them consistently across calls, WhatsApp, and SMS.
- Delight every Guest: recovery follow-ups help Guests finish bookings and requests they started but did not complete.
Summary
- Set your own criteria for successful conversations.
- Grade new calls and WhatsApp/SMS threads automatically.
- Enable recovery messages when a Guest leaves a process unfinished.
- Review pass rates and individual results in AI Evaluations, Overview, History, and Guest timelines.
- Choose which evaluations run for each agent.
AI Evaluations
Version v3.4 adds checks for conversation quality. Each evaluation has a title and a description of the behavior you expect.
Examples:
- Stay booked: each time the caller wanted to book a stay, the agent booked it.
- Callback offered: when the agent could not help, it offered a callback.
Voicetta checks the transcript and returns:
- Pass: the expected behavior was shown.
- Fail: the situation occurred, but the agent did not meet the criterion.
- Not applicable: the situation did not occur, or there was no usable transcript.
Each result includes a short explanation.
Configuration and defaults
Evaluations belong to the workspace. You choose which agents use each one. Agent badges show where an evaluation is active.
Workspace settings control text grading and recovery (open AI Evaluations → gear):
- Text thread grading delay: 0–60 minutes after the latest message.
- Max per text thread: 1–5 recovery messages. Default: 1.
- Max per voice call: 1–5 recovery messages. Default: 1.
- Max per Guest per day: 1–10 recovery messages across all channels in 24 hours. Default: 1.
These limits are shared by all evaluations with recovery enabled. Disconnect follow-up (added later) uses the same Evaluation settings page — see AI Evaluations.
Calls and text threads
Calls are graded when they end.
WhatsApp and SMS messages are grouped into threads and graded together after the workspace's configured delay. Each graded thread includes a short summary. If the Guest replies later, the thread opens again and is re-graded, so a previous Fail can become Pass.
Recovery follow-up
Recovery can be enabled separately for each evaluation. When a Guest starts but does not finish a process, such as a booking, Voicetta can send a short message in the Guest's language.
The follow-up channel depends on the conversation:
- Phone call: SMS first. WhatsApp is used only when its 24-hour messaging window is open.
- WhatsApp voice call: WhatsApp first when the window is open; otherwise SMS.
- WhatsApp/SMS thread: follow-up continues in the available text channel.
Guests who opted out do not receive recovery messages. Test calls and browser widget calls do not trigger them.
Managers can distinguish the first result from the final outcome:
- First-contact pass: the criterion was met without a follow-up.
- Recovered: the first result was Fail, but the Guest completed the process after a follow-up.
- Awaiting reply: a follow-up was sent and Voicetta is waiting for the Guest.
- Needs attention: the issue is still unresolved.
- Final success: the criterion was met either immediately or after recovery.
The initial result stays visible after a conversation is recovered.
Results and reporting
The main menu now includes AI Evaluations. From this page, you can:
- create, edit, and delete evaluations,
- enable or disable an evaluation for each agent,
- review pass rates and result counts,
- open individual results,
- export results as CSV.
Evaluation results also appear in:
- Overview: workspace results for the selected period.
- History: results and explanations in conversation details.
- Guest timelines: results shown on conversation cards.
Other changes
- The app version badge now shows
v3.4. - Models used for summaries, intent, sentiment, and evaluations are now managed centrally by Voicetta. Model updates do not require an app update or action from workspace administrators.
Important limitations
- Evaluations apply only to conversations that end after the evaluation is created and enabled.
- Existing conversations are not graded retroactively.
- Grading depends on the available transcript. Calls without a usable transcript are marked Not applicable.
- WhatsApp recovery messages must stay within the 24-hour messaging window.