Lisa runs four short-term rental apartments in Vienna — three in the 4th district, one in the 7th. All four are imported via Smoobu. She's a full-time marketing consultant; the apartments are what she calls "her inheritance business" — managing them for the family, part-time, at the margins of her main career.
Before Virtual Host AI, Lisa spent roughly 12-15 hours a week on the apartments. Most of it was guest messaging — not individual messages taking long, but the constant context-switches from her main work. After six months on Virtual Host AI, she estimates 2-3 hours a week. Here's what that looks like in practice.
The numbers
Some baseline metrics from her account, averaged over the last 90 days:
- Apartments active: 4
- Guest stays per month: 28-32 across all four
- Incoming guest messages per week: 68-85
- Messages auto-handled by the AI: ~82%
- Messages escalated to Lisa: ~18% (12-15 per week)
- Time Lisa spends on guest communication: ~90 minutes per week
- Upsell accept rate (late checkouts): 27%
- Extra monthly revenue from upselling: ~€250
- Net Promoter Score change: +11 points vs pre-AI baseline
That last one was a surprise. We expected NPS to hold steady — Lisa was already a conscientious host. But response-time compression (guests now getting answers in under a minute, 24/7) moved her reviews noticeably upward.
7:42 AM — the first Telegram ping
Lisa wakes up, checks Telegram over coffee. There's one escalation from overnight: a guest at the 7th-district apartment says the oven "beeps weirdly when I try to use it." The AI escalated because "weird beeping" is outside the apartment guide's knowledge.
Lisa taps Reply on the Telegram card, swipe-types: "Hi Marco — the oven has a child-lock mode that beeps if a pot's sitting on the touch panel. Just move anything off the top surface. Let me know if that fixes it." Sent. She puts her phone down. Total time spent: 40 seconds.
Marco replies within five minutes confirming it worked. The AI notes Lisa's response and stays quiet on that thread for an hour, per our intervention shield.
9:30 AM — commute, no pings
Lisa takes the U-Bahn to her main job. Her phone vibrates twice — both calendar notifications about her own meetings. Nothing from Virtual Host AI. Four apartments, all running, and her attention is entirely on her actual job.
This is the default state. Most days, the AI handles every guest message without Lisa hearing about it.
11:45 AM — upsell approval
Telegram buzzes. Card shows: "Upsell for Sunset Studio — late checkout to 2 PM for €20 (current guest: Sarah M., checks out tomorrow). [Approve] [Skip]"
Lisa reads it during a micro-break at her desk. She knows the 4th-district apartment has no arrival until Thursday, so a late checkout is safe. She taps Approve. Two taps total.
The AI sends the offer to Sarah that evening at 7:14 PM. Sarah accepts at 8:02 PM. Lisa gets a confirmation ping at 8:02: "✓ Sarah accepted late checkout at €20. Cleaning window adjusted." Lisa doesn't even open Telegram for the confirmation — it's a glanceable notification.
3:15 PM — a complaint
This is an escalation that needs a human. Card shows: "Escalation — Park View (guest: Jens K., stay ends Friday). Message: 'The neighbors above are very loud at night, I can't sleep.'"
Sentiment-driven escalation triggered because the message mentions sleep and "can't" — distress signals. The AI deliberately didn't reply; it wanted Lisa's judgment.
Lisa steps out of a meeting, calls Jens directly. They talk for four minutes. She apologizes, offers a €50 credit against his stay, explains she'll talk to the upstairs neighbors. Jens is mollified. Lisa marks the escalation Solved in Telegram.
This interaction would have happened anyway. What the AI saved her is the 40+ routine messages that didn't interrupt her afternoon. She was available for Jens because she wasn't already drained by "what's the Wi-Fi."
Want Lisa's workflow for your apartments?
Smoobu autopilot + Telegram escalations + calendar sync. From €22.90 per apartment per month.
6:20 PM — guide update
Walking home, Lisa gets a Telegram escalation: "Park View guest asks about the rooftop terrace — is it accessible? Not in guide." The AI sent a safe response ("I'll check with the host") and pinged Lisa.
Lisa realizes this is a gap in the guide she should fix. She types /guide in Telegram, picks Park View from the inline menu, and types: "Building has a shared rooftop terrace on the 6th floor, accessible via the main stairwell. Open from 9 AM to 10 PM. No alcohol after 8 PM per building rules." She hits send.
The guide updates immediately. The next guest who asks about the rooftop gets the correct answer automatically. One-time investment of 60 seconds, permanent improvement.
Over six months, Lisa has added approximately 40 of these small guide improvements. Her guides are now better than they were before she started automating — because the friction to add information dropped from "open laptop, go to apartment settings" to "three taps in Telegram."
9:00 PM — phone down
Evening. Lisa is making dinner with her partner. Phone is in the kitchen, volume on. Three guest messages arrived in the last hour; the AI handled all of them without involving her.
Pre-Virtual Host AI, this would have been different. The old Lisa would have had her phone on the counter, checking between stirrings. The new Lisa doesn't.
Saturday — the weekly rhythm
Saturdays Lisa does a 20-minute review. She opens the web app, checks:
- The week's escalations — which ones mattered, which were false positives
- Upsell performance — how many offered, how many accepted, how much revenue
- Any unresolved guest feedback themes
- Cleaner's schedule for next week (via cleaning coordination) in case any gaps need attention
After six months, the AI has settled into a rhythm where false-positive escalations are rare. Lisa reports maybe 1-2 per week where she thinks "the AI could have handled that." Not zero — the cautious bias is intentional — but low.
What this costs her
Lisa's on the Pro plan at €26.90 per apartment per month × 4 apartments = €107.60 per month. She's captured €250/month in upsell revenue that she previously wasn't capturing at all. Net: €142/month in her pocket, excluding the value of her time. Including her time (she values her hourly rate at €60), the savings are another €500+/month in hours reclaimed.
She does not consider Virtual Host AI a "cost." She considers it revenue infrastructure.
The tradeoffs she's made
Honesty requires flagging what she's given up.
Some personal touch is gone. The "warm welcoming message from the host" that used to go out has been replaced by the AI's tonally-similar but not identical version. Lisa's NPS went up anyway, but she did miss the craft of it at first.
She checks Telegram more than she checked email. Telegram notifications are punchier than email. She estimates she glances at Telegram 10-15 times a day now, whereas previously she checked Smoobu's app 40-60 times. Net attention is much less, but it's more concentrated.
She's trusted the AI with €250/month in upsell revenue. Every offer is host-approved, but she's developed a pattern of approving most of them without much thought. This is fine — the AI's pricing and timing are good — but she consciously reviews the upsell flow once a month to make sure she still agrees with the defaults.
What she wishes worked better
One request: Lisa's old-school German-speaking neighbors occasionally need her to print and hand them a paper schedule of cleaner visits. Our cleaning coordination is fully digital. She's asked for a "print week" feature. We're considering it for Q2 2026.
Other than that, she's asked for nothing in the last 90 days. That's, by our measure, the best case study metric.
Want your week to look like Lisa's?
Four apartments, 90 minutes/week, €250 extra upsell revenue.