Measuring AI receptionist ROI: the four numbers that matter
Forget vanity metrics. These are the four numbers that tell you whether your AI receptionist is actually paying back — calculated from real customer data.
Most AI vendor dashboards drown you in metrics. "Conversations started", "average response time", "satisfaction score" — interesting, but none of them tell you whether the assistant pays back. Here are the four that actually do.
1. Deflection rate
The single most important number. What percentage of incoming questions does the AI handle without escalating to a human?
Janore tracks this automatically. A typical SMB sees 60–84% deflection in the first month, climbing to 80–90% by month three as the FAQ improves. Below 50% means your knowledge base is too thin — the assistant is escalating because it does not know.
To compute the cash value: multiply your deflected conversations by the average human-touch cost (€2–€8 per interaction depending on staff cost). 800 deflections × €4 = €3,200/month freed up. Janore costs €29–€199/month. The ROI is rarely subtle.
2. Booking conversion rate
For booking-driven businesses (salons, restaurants, clinics), the assistant should not just answer — it should book. Track: of conversations where booking intent was detected, what percentage ended in a confirmed slot?
A good number is 35–55%. Below 30% means your booking flow has friction (probably asking for too many details, or the next-available slot is too far out). Janore surfaces drop-off points so you can fix the right step.
3. After-hours capture
The hidden goldmine. What percentage of your bookings now come in outside office hours (19:00–08:00 + weekends)?
Most businesses see 25–40% of conversation volume after hours, and the assistant captures it instead of losing it to voicemail. If you charge €60 average per appointment and recover 20 after-hours bookings/month that would have evaporated, that is €1,200/month directly attributable.
4. Mean time to first reply (TTFR)
Not "did they reply" — "how fast". Visitors abandon a chat after 90 seconds of silence on average. Janore's TTFR is consistently under 2 seconds; a human service averages 30–120 seconds depending on staffing.
The math: every 30 seconds shaved off TTFR converts roughly 4% more visitors from "asked a question" to "completed a conversation". Over a month, that compounds.
What you do NOT need to measure
A few metrics most vendors push that are noise:
- **Satisfaction emoji clicks** — 90% of visitors click "happy" reflexively. Useless.
- **Average words per reply** — visitors do not care about word count, they care about the answer being right.
- **Total conversations** — go up over time naturally; not a quality signal.
- **AI confidence score** — fascinating internally, irrelevant to ROI.
Build the dashboard yourself
If you want a one-page ROI view, four lines:
Deflected conversations × €4 = €X saved
Bookings via assistant × €60 = €Y captured
After-hours bookings × €60 = €Z incremental
SUM minus subscription cost = ROIIf ROI is negative two months running, something is wrong — usually the FAQ is too thin or the booking flow has friction. Both are fixable in an afternoon.
What good looks like at month 6
A typical Janore SMB at six months:
- 82% deflection rate
- 47% booking conversion
- 31% of conversations after hours
- 1.4 second TTFR
- ROI 12–28× the subscription
Your numbers will differ — but if all four point in the right direction, the assistant is paying back. If one is stuck, fix the matching gap.