By Sagar Shankaran, Founder of CallSphere
What counts as routine and what counts as hard in a vacation rental inbox, who owns the escalation list, and the cost math on 41,000 guest messages a year.
Key takeaways
Which of these two messages, sitting eleven seconds apart in your Guesty inbox on the afternoon of 3 July, deserves the same amount of your company’s money and attention?
Hi! What is the wifi password for Dune Ridge 12?
The a/c has been blowing warm since noon, it is 96 out, my wife has MS and cannot be in this heat, and if this is not fixed tonight I am disputing the whole charge with my credit card company.
Today they cost the same, because the same guest services agent reads both, in the order they arrived, and the second one waits behind eleven wifi questions.
In a 200-door company, roughly four out of five inbound guest messages are the same twenty questions: check-in time, gate code, parking pass, does it have a coffee maker, which beach access, is the pool heated, can we bring the dog, where do we put the trash, what time does the shuttle run. They are answerable from the house manual and the reservation, and they are answerable in four seconds.
The remaining fifth is where your season is decided. A failed HVAC compressor on a holiday weekend. A guest claiming bed bugs. A double-booking created when the channel connection dropped and Vrbo and your direct site both sold the same week. An HOA president threatening to report a party to the city, which in a permit market means your owner’s short-term rental registration. A refund demand that exceeds what your property management agreement lets you approve without the owner’s consent.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for hospitality in your browser — 60 seconds, no signup.
Model routing is the practice of sending the cheap, repetitive requests to a fast inexpensive model and reserving the expensive, careful one — and the human behind it — for the cases where being wrong costs you a stay, a permit or an owner. Cisco built exactly this into the personal agent it is rolling out to roughly 90,000 employees, for the same reason you would: capability where it matters, cost control everywhere else.
Every industry that tries routing gets stuck on the same question, and the answer is never generic. Here it is for vacation rentals, and note that none of these lines are about how hard the sentence is to read — they are about who bears the loss if the answer is wrong.
flowchart TD
A["Guest message lands in the channel inbox"] --> B{"Answerable from house manual or reservation?"}
B -->|Yes| C["Fast model drafts and sends, note on reservation"]
B -->|No| D{"Money, habitability, permit or owner involved?"}
D -->|No| E["Strong model drafts, guest services agent approves"]
D -->|Yes| F["Straight to on-duty guest services manager"]
C --> G{"Guest re-asks or pushes back?"}
G -->|Yes| E
G -->|No| H["Closed, counted in the weekly routing review"]
E --> H
F --> H
Not the vendor. The escalation list belongs to your guest services manager, it lives in a document your staff can open, and it gets reviewed the first Monday of every month. The reason it cannot be the vendor’s default is that the line moves by market and by calendar. A power flicker in a Smoky Mountains cabin in January is a hard case, because the guest is on a mountain in the dark. The same flicker in a Scottsdale condo in March is routine. Outside temperature over 90 makes an air conditioning complaint hard in Destin from June through September and routine in Big Sky.
The line also moves with occupancy. At 96% occupancy on the Fourth of July, you have nowhere to move anyone, so a habitability complaint escalates to the general manager instead of the guest services manager, because the decision is going to be about money rather than a relocation. Write that into the list, seasonally, and stop treating it as a judgment call your newest seasonal hire has to make at 7 p.m.
Assumptions: 200 doors, 41,000 inbound guest messages a year across Airbnb, Vrbo, Booking.com, SMS and your direct inbox. A guest services agent handles about 22 messages an hour, fully loaded cost $27 an hour, so roughly $1.23 per message handled by a person. The routing split below is typical for a company with a decent house manual; yours will differ.
| Tier | Share | Messages | Cost each | Annual |
|---|---|---|---|---|
| Routine, fast model, sent automatically | 76% | 31,160 | $0.004 | $125 |
| Middle, strong model drafts, agent approves | 18% | 7,380 | $0.41 | $3,026 |
| Hard, straight to a person | 6% | 2,460 | $1.23 | $3,026 |
| Routed total | 100% | 41,000 | — | $6,177 |
| All-human comparison | 100% | 41,000 | $1.23 | $50,430 |
The saving is real but it is the less interesting number. The interesting one is response time on the 6%. When 76% of the queue stops sitting in front of the air conditioning complaint, the median time to a human on a habitability case in one of these companies typically falls from over an hour on a holiday weekend to under ten minutes — and the difference between a 90-minute wait and a 10-minute wait on 3 July is frequently the difference between a partial refund and a chargeback plus a two-star review that sits on the listing all season.
Prove it with two counts, not with a vendor dashboard: median minutes to first human touch on escalated cases, and the number of guest-initiated refund requests per hundred stays, measured for the same weeks a year apart.
Still reading? Stop comparing — try CallSphere live.
See the hospitality AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
The dangerous failure is not a wrong wifi password. It is a hard case dressed as a routine one. Is the pool heated? is routine unless it comes from a guest who booked specifically for a heated pool, wrote it in the special requests field, and is about to discover the heater failed — and that context lives in the reservation notes, not in the sentence. Make sure the routing reads the whole reservation and the whole message thread, not the newest line, and make anything with a special request attached escalate automatically.
The second one is emotion. A polite message can be a furious guest, and this trade’s worst outcomes come from guests who were polite three times and then went to their credit card company. Route on the third message in a thread to a person regardless of what it says. It is a cheap rule and it catches most of them.
Finally, keep a person on anything involving a minor, an accessibility need, an allergy, or a police report. Not because the model would answer badly, but because if it goes wrong you want a name on it, and you want that name to have been awake and paying attention.
The guest services manager, with a monthly review that includes the property care manager and your revenue manager. Owners are usually tempted to write it themselves and then never touch it again. It goes stale within one season, because your markets and your occupancy change.
Occupancy-based rules. Above roughly 90% occupancy, widen what counts as hard and staff for it — a holiday weekend at full occupancy has no relocation options, so more decisions become money decisions. The routing should read your occupancy, not just the message.
They should. A line saying a member of the team is taking this over personally, with a name, buys more patience than anything else you can write. Guests forgive slow far more readily than they forgive being managed.
Read a sample every week — 25 messages, ten minutes, done by your guest services manager, logged. Also read 100% of anything that escalated after an automatic reply, because those are the cases where the line was drawn in the wrong place.
CallSphere builds AI voice and chat agents that answer business phone lines and web chat, book appointments and capture leads around the clock. The same split applies on the phone as in the inbox: the gate-code call and the no-air-conditioning call should not wait in the same queue, and the second one should be in a manager’s ear with the unit number and the stay dates already attached.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
Sixty-one pages of addendum land three days before a DOT letting. How model routing gets an estimator the quantity changes that actually move the bid.
The pre-season exterior survey eats four weeks of May and never gets finished. What autonomous inspection routes change for coastal vacation rental managers.
Model routing sends routine product requests to a cheap model and hard ones to a strong one. Here is who draws the line in a B2B software company, and how.
Cheap model for potholes, strong model for water quality, a pager for sewage in a basement. How a public works superintendent writes the triage table.
How water systems use a cheap-model first pass and a strong-model escalation so nobody drives out at 2 a.m. for a SCADA alarm that already cleared itself.
Sort courier calls before answering: cheap model for status and reschedules, strong model for damage claims and missed medical runs. Costs, rules and limits.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI