The end-of-day email already told you what happened. Now it tells you what's getting better — and what to fix. One glance, overall score, weakest dimension, the calls that fell short, and the top pending improvements waiting for your approval.
The end-of-day recap email has been going out since the day we launched. Call count, how many booked, voicemails at the top, avg duration. Operational stuff. Useful, but it only told you what happened.
Starting today, it also tells you how the AI is doing — and what's worth fixing.
Open tonight's 5 PM recap and you'll see a new block: Agent performance — yesterday.
Overall score: 4.2 / 5 · 12 calls scored
Weakest dimension: Intent resolution (3.4)
Calls that fell short
Booking · 2.0/5 · Maria — Agent never understood the caller wanted
to cancel
Pricing · 3.0/5 · Kate — Booking didn't land
Suggested improvements
Flow · 14 calls
Acknowledge multi-service bookings in first turn
Knowledge · 6 calls
Add parking guidance to knowledge layer
Review & approve in the dashboard →
That's the whole block. Takes five seconds to scan. Tells you three things:
If your AI had a clean day, the section hides itself. No empty-state box. No noise on a day where everything worked.
We'd already built all the pieces. Every call gets scored overnight on seven dimensions. The system proposes specific fixes. The dashboard shows everything.
The problem: nobody was checking the dashboard. Salon owners are on the floor during business hours. Evenings they scroll through email. The feedback loop was open, but the feedback wasn't reaching them.
Now it does. The 5 PM email is the channel owners already read. Moving the signal there means you see yesterday's agent report card without changing anything about your routine.
We draw the line at an overall score below 4.0 (on a 1-5 scale). Above that, the rubric says the call was solid — no need to surface it. Below that, something went wrong enough to look at.
Every call that makes the list gets the first issue from the QA scorer surfaced as a one-line reason. If the scorer found multiple issues, the email shows the one flagged as most severe. You can click through to the dashboard for the full breakdown.
Typical reasons you'll see:
Sometimes the thing that "fell short" is a graceful failure. We still surface it so you see the pattern.
Up to three pending improvements show up, sorted by how many calls they affect. These are the patterns the nightly QA system spotted across multiple calls — not one-off issues, but things that keep tripping the AI.
Each one tells you:
You don't approve from the email — click through to the dashboard, read the full proposed text, and accept with one click. The change applies to your live agent within seconds.
Today's email is report-only. The AI doesn't apply fixes on its own overnight — you still review each suggestion.
That's deliberate for now. We want you to see what the model is proposing before we let it self-apply. Once you've accepted a few cycles' worth of suggestions and the quality is obviously right, we'll open the door to auto-applying the safest ones (the ones affecting 5+ calls with high confidence). For today: the human stays in the loop.
We're also not running a regression check yet. If you accept a suggestion on Monday and it makes things worse by Thursday, today you'd have to notice that yourself. That's the next piece we'll ship — 24 hours after an accepted suggestion goes live, we'll compare scores on the affected call type and either confirm the improvement or flag a rollback for your review.
The Agent Performance section appears automatically in your next 5 PM email — assuming yesterday had scored calls. If your 5 PM email was quiet today, it'll be quiet again; this doesn't flood your inbox on days when there's nothing to say.
If you've opted out of the end-of-day recap and want it back, Settings → Notifications.
Your receptionist's nightly review just landed in your inbox. Take 30 seconds tomorrow evening and see how yesterday went.
Learn more: /docs/owner-notifications.