AI Copilot

The Copilot is an AI chat that knows your account: ask it anything in plain language and it answers with real data from your calls, campaigns, and agents. It can also propose changes (create an agent, adjust a campaign, add a user) — but nothing is applied without your confirmation.

Starter blocks and interactive cards

When you open the Copilot it greets you with tappable action blocks (list agents, voice test, call a phone, recent calls, campaigns, plan & usage) — you only see the ones your role allows. And data answers arrive as interactive cards, not plain text: the agent list has buttons to view details, test or edit; a call detail includes the recording audio and the transcript; campaign analysis brings the funnel; and your plan shows a minutes meter.

Reports with charts that actually read well

Ask for "give me the results report" (you can narrow it to one agent or a number of days) and you get a card with the key numbers, a sentiment donut and goal met vs not met bars in green and red. Below them, the card always spells out what each number means, so nobody misreads the data.

The card comes in two layers you should not mix. First Telephony · who answered: answered by a person, no answer, voicemail, busy, rejected and failed, with the contact rate. Then the agent's goal: out of the calls where there WAS a conversation, how many achieved what you asked for.

Two useful details about that layer: "no answer" includes calls the carrier reported as answered but where nobody spoke (the agent talked to silence) — those used to count as contact and they were not. And voicemail is only detected on agents that have voicemail detection enabled: if you see zero, it may be that no agent has it on, not that there were no voicemails.

To see it over time, ask for "a chart per day of answered, no answer and voicemail": it draws one line per metric in the same chart, with its legend.

Why do your calls end?

Ask exactly that and the card shows you two lists. Agent disposition: the reasons in YOUR script's own words ("Reminder successful", "Transferred-Financial hardship"), which is what your team recognizes. And How the call ended: the technical cause the platform records (customer hung up, silence, transferred, voicemail, duration limit).

You can also ask for "show me the contacts of campaign X": the card brings the table with phone, status, attempts and sentiment, plus the columns from the file you uploaded (ID number, amount, due date… whatever your list has). It pages itself, without spending extra questions.

Voice-test and call, without leaving the chat

• Voice test: say "I want to voice-test agent X" and a card with a microphone appears so you can talk to the agent right there, in the chat.
• Real phone call: say "call +57…" and it prepares the card with the number and the call data. The call only goes out when YOU tap "Call" — the Copilot never dials on its own.
• Card buttons chain the conversation: tap "View details" on a call and the Copilot brings it up instantly.

What you can ask

• "How are my collections campaigns doing this week?" — it builds a summary with numbers and even a chart.
• "How is call sentiment this month?" — it breaks down positive/neutral/negative and success vs failure, globally or per agent, with a chart.
• "Audit my sales agent" — it checks the coherence between the script and the real configuration and proposes the fixes.
• "Create a receptionist agent for my clinic" — it proposes one ready to go; you review and confirm.
• "Why did this call end badly?" — it analyzes the transcript and gives you its read.
• "Save this call as a test" — turns a call that went wrong into a reusable test (see Agent tests).
• "Add a webhook for when a campaign finishes" — it prepares the action and waits for your OK.

Build and improve agents by chatting

From an agent's editor, the "Improve with AI" button opens a dedicated chat: describe what you want and the Copilot writes the script, picks the voice, configures tools, variables and analytics, and shows each change as a draft with its differences. You apply what convinces you — whole or piece by piece — and save.

When a problem shows up in several points of the script (which is the norm: a non-existent tool is usually mentioned across many steps), the Copilot tells you how many times it appears and in which sections, and fixes each point separately, keeping the step, its message and its disposition.

• Ask it to "Audit my agent": it checks the coherence between the script and the real configuration and proposes a fix for every finding.
• When it needs more information, it asks with tappable options — you answer in one tap.
• After a test call, ask it to analyze it: it tells you how it went and what to adjust.
• Every proposal takes 👍/👎 — it learns from what you approve and avoids what you reject. The Versions tab shows which script version converts best.