Getting started — your first day with your AI Operator

You just hired your AI Operator. The default name is what you set during onboarding — Anni at Anni Head Spa, your Operator's name at your business. This article walks you through what to do on day one, what to expect in the first week, and where to look when something feels off.

What "Operator" means

Your Operator isn't a chatbot, an assistant, or a feature you toggle on. It's a worker with a name, a job description, and an employment contract. We use those exact words because the framing matters: workers ask before doing risky things, follow rules they were given, and learn from feedback. That's the model.

The Operator does five things on day one:

• Picks up the phone when patients call after hours.
• Drafts replies to SMS, email, and Instagram messages.
• Looks up appointments, prices, and services from your booking system.
• Flags anything outside its rules for you to handle.
• Writes a receipt for every action it takes, every reasoning step, every refusal.

What it does NOT do on day one: medical advice, medication dosing, refunds, complaints. Those are hard guardrails enforced in code, not just policy.

The first 24 hours

We've already imported your services, prices, providers, and hours from your web presence — the discovery you saw at signup. If anything looks wrong, you can edit it from Operator → Knowledge in the workspace.

For the first week the Operator runs in supervised mode. Every reply it drafts comes to your inbox for approval. You should expect 5-10 drafts per day in the first few days. Approve as-is, edit, or dismiss — every edit teaches the Operator the way you actually talk to your patients.

The voice line works differently in week 1. The Operator runs in shadow mode — it listens to calls but doesn't speak. We compare what it would have said to what you actually said, and use the diffs to tune tone and pacing. Week 2 it goes live with disclosures and real-time guardrails. You'll be notified before that switch flips.

What you'll see on Today

The Today page is your morning briefing. Three sections, top to bottom:

• Needs your attention now — drafts to approve, escalations the Operator couldn't handle, anomalies flagged overnight.
• Handled by your Operator — proof of work. Calls answered, messages replied to, appointments confirmed. Each row is a receipt you can replay.
• Pulse — calls, bookings, revenue captured. The numbers that prove the system is working.

Click any row to expand. Every entity (patient, call, appointment, receipt) has an inspector you can open without leaving the page.

The voice line

We provisioned a phone number for you. It rings the Operator first, with fallback to your existing number if anything misfires.

Disclosures play at the start of every call. The default opening is something like:

"Hi, you've reached [your business]. I'm Anni, the AI operator. This call is recorded for quality. How can I help you today?"

You can edit the disclosure in Operator → Voice → Disclosure script. State-specific compliance language (California two-party consent, Florida two-party, Illinois BIPA implications) is included by default. See the Voice call disclosure scripts article for the full breakdown.

If a caller asks "are you AI?" the Operator confirms truthfully. We never deny being AI — that's a hard guardrail. If a caller asks to opt out of recording, the Operator offers to transfer to a human.

The Trust Center

Every action the Operator takes writes a Receipt. The Trust Center, under workspace → Trust, is the ledger.

• Filters: date, skill, status, category, autonomy mode.
• Click any receipt to open the inspector. Replay the decision. See the reasoning trace.
• Pause any skill or trip the global kill switch from the same surface. The switch propagates in under a second.

The four things to do in Trust Center your first week:

1. Open one voice-call receipt. Read the transcript and the reasoning.
2. Open one reply-draft receipt. See what the Operator did with the corrections you wrote.
3. Open the Knowledge view. Check that services, prices, and providers match reality.
4. Check the kill switches. Know where they are before you need them.

When something feels off

If the Operator did something you didn't expect — wrong tone, wrong info, missed something obvious — three options:

• Report an issue in the workspace header (top right) or hit ⌘+/ from anywhere. Tell us what happened. The system auto-captures the page you were on, the last 10 receipts, and any console errors. We respond within the SLA your plan promises.
• Email support@hellodella.com.
• Founding-10 cohort only: text Nate's cell. You have the number.

Critical issues — voice line down, payment processing broken, anything tagged "urgent" — auto-page on-call 24/7.

What to do this week

Day 1: approve 5 drafts. Read one reasoning trace. Look at Operator → Knowledge to spot-check the import.

Day 2: review the voice disclosure script. Sign it.

Day 3: watch the first 3 voice shadow-mode runs. Correct anything off-tone in the inspector — your correction becomes a Correction record and feeds back into the next draft.

Day 4-7: keep approving. Look for patterns in your edits — those are policy candidates. The graduation gate to autonomous mode opens at 30 days, but you always have the final say on when to flip it.

A note on autonomy

Autonomy is per-skill, not per-Operator. You can run reactivation campaigns autonomously while keeping voice in supervised mode, or any combination. The hard guardrails (medical advice, refunds, complaints, medication dosing, PHI disclosure) are never autonomous, regardless of mode. See Setting your autonomy preferences for the matrix.

What this article doesn't cover

Things you'll want to read next:

• How to talk to the Operator — browser call + ⌘K command palette
• Setting your autonomy preferences — per-skill autonomy modes
• Reporting an issue — what the SLA looks like
• Privacy + HIPAA — how PHI is handled

If anything in this article is wrong or missing, hit Report an issue and tell us.