Desert Sol: 6 Months with AI Operations

Desert Sol: 6 Months with AI Operations

130+ vacation rental properties in Palm Desert, California. A team managing calls, messages, checkout requests, and payment follow-ups across Airbnb, VRBO, and direct bookings — all while trying to grow the portfolio.

Six months ago, that team deployed Dimora AI. Here is what happened.

This is not a projection. Not a benchmark. Not a case study written from survey responses. These are the actual numbers from a live production deployment. For the complete technical breakdown, see the Desert Sol case study.

The Starting Point

Desert Sol Real Estate manages 130+ vacation rental properties across the Coachella Valley — Palm Desert, La Quinta, Indian Wells, and surrounding communities. Their properties run on Guesty. Guests book through Airbnb, VRBO, and directly. Properties range from compact condos to large homes in gated communities with HOA rules that vary by neighborhood.

Before Dimora AI, the operations picture looked like this:

Phone coverage: Staff available during business hours, voicemail after 8 PM. The 2 AM lockout call meant someone getting woken up. Coachella weekend meant all-hands coverage. There was no version of "every call answered" that did not require staffing it.

Inbox volume: Airbnb messages, VRBO messages, email inquiries. Each response required pulling the right property information — not from memory, but from Guesty, which meant context-switching dozens of times per day. A 4-person team across 130+ properties during peak season. Messages waited hours.

Revenue operations: Late checkout and early check-in requests were handled when guests asked. Gap nights — those one- and two-night openings between reservations — required someone to spot them manually and reach out. Nobody had time to do it systematically.

Payment tracking: Balance reviews happened on an ad-hoc basis. No daily scan. No automated follow-up sequence.

The ceiling was structural. Growing the portfolio required proportionally growing the team.

The Deployment Timeline

Week 1: Integration and Go-Live

The Guesty API integration pulled all 130+ property records into the Dimora knowledge base. Property descriptions, house rules, lockbox codes, WiFi credentials, parking instructions, saved replies — all of it indexed and available.

Voice AI activated on the Desert Sol business number. From day one: first-ring pickup, 24 hours a day, seven days a week.

No warm-up period. No "training calls" that went to voicemail. The first call after go-live was answered by the AI, and it resolved the guest's check-in question in under 90 seconds.

Week 2: Inbox AI Training Phase

The Inbox AI entered draft-and-review mode. The system generated message drafts — full response drafts, not templates — and the Desert Sol team reviewed, edited, and sent.

Every edit the team made was captured. The AI Learning module analyzed each one: what the AI drafted, what the PM changed, why the change improved the message. These comparisons built the golden examples dataset — the foundation of the self-improvement loop.

By end of week two, 30+ golden examples existed. The Revenue Engine activated simultaneously, scanning the calendar for gap nights and upsell windows the moment it had reservation data to work with.

Coachella Weekend: The Real Test

February brought Coachella booking season. Late checkout requests, lockout calls, HOA noise rule questions, parking inquiries, gap night availability requests. The kind of volume that, in prior years, meant overtime and a team stretched thin.

The staff did not work overtime for phone coverage. That had never been true before.

Voice AI answered every call. First ring, every property, simultaneously. It knew which property each guest booked, had the access code for that specific unit, and provided parking instructions from the actual listing record — not a generic answer.

The inbox team reviewed AI drafts. Nothing waited more than 30 minutes.

6 Months of Numbers

Voice AI: 1,800+ Calls Handled

1,800+ calls answered across the 130+ property portfolio since activation. Every one answered on the first ring.

The call breakdown by type reflects exactly what you would expect from a vacation rental portfolio at this scale: lockout and access calls, check-in instruction requests, late checkout and early check-in inquiries, amenity questions, WiFi and technology issues, booking availability requests.

All resolved directly. No voicemail accumulation. No staff hours consumed on routine inquiries. The small percentage of calls that genuinely required a human — maintenance emergencies, owner calls, complaints — were escalated with full context and a call transcript.

One number to hold onto: the previous voicemail backlog was a constant operational drag. After activation, it reached zero and stayed there.

Inbox AI: 6,300+ Drafts Generated

6,300+ message drafts generated since the Inbox AI went active. Each one in under 10 seconds.

The 7-agent architecture routes each incoming message to the appropriate specialist before drafting a response:

• Property Info agent handles questions about amenities, rules, and property features
• Door Code agent handles access and check-in instructions
• Availability agent handles booking inquiries and calendar questions
• Early/Late Checkout agent handles timing requests and sends offers
• Offer Acceptance agent processes accepted upsells
• Escalation agent identifies situations requiring human judgment
• General QA agent handles everything else

The draft quality improvement over six months is measurable. In week two, the team edited roughly half of all drafts before sending. By month six, most drafts go out with minor refinements or none at all — a direct result of the AI Learning loop processing those early corrections into the golden examples dataset.

Revenue Engine: 470+ Upsell Offers Sent

470+ upsell offers sent automatically since activation. Zero manual identification. Zero manual outreach.

The system runs two workflows continuously:

Early/Late Checkout: When a reservation is eligible — sufficient turnaround time, no adjacent arrival blocking it — the system sends a personalized offer. $35 for Legacy Villas properties. $50 for other properties. 11 AM checkout is always included free. The timing calculation uses real turnaround data: the latest possible late checkout equals the next guest's arrival time minus 3 hours.

Gap Night Extensions: When a one- or two-night gap exists between reservations, the system identifies both adjacent guests and sends discounted extension offers. Fill the gap at a reduced rate or leave it empty — the math is clear, and the offers go out automatically.

470+ offers. All generated from calendar data. All personalized with actual property names, specific time windows, and accurate pricing. Not one of them required a PM to open a spreadsheet, identify the window, draft the message, and send it.

AI Learning: 52 Golden Examples

52 verified golden example pairs built over six months. Each one represents a real interaction where the PM's edit improved the AI's draft — and that improvement is now applied to every future message in a similar context.

The self-improvement loop works in one direction only: forward. The AI does not get worse over time. Each golden example raises the floor for draft quality in that category.

By month six, the knowledge base reflects Desert Sol's actual communication style — their tone, their policies, their way of explaining HOA rules to guests who have never stayed in a gated community before. You cannot build that from a template library. You build it by capturing the judgment calls a good communicator makes hundreds of times.

Payment Audit: Daily Scans Active

Daily 9 AM scans across all active reservations. Outstanding balances flagged, automated follow-up sequences initiated, escalation triggers set for unresolved cases at check-in.

Before the Payment Audit module, balance reviews happened whenever someone had time to run them. That meant some balances slipped through until the guest arrived. Now the scan runs every morning without exception, and flagged cases receive consistent follow-up — five-message escalation sequence before a human review.

Before and After: What Actually Changed

The team still reviews AI drafts before they send. The system operates in draft-and-review mode by design — not because the AI cannot be trusted, but because PM oversight during the review phase is what makes the AI measurably better over time. For guests, the experience is a response drafted by an AI that knows their reservation, reviewed by a human who manages their property. That combination is faster than human-only drafting and more accurate than AI-only sending.

What the Modules Look Like in Practice

A Typical Lockout Call

11:47 PM. Guest calls. They cannot get into the property.

Before AI: staff member's phone rings, they pull up Guesty, find the property record, find the access instructions, call the guest back. Four minutes minimum. Often longer.

After AI: the call is answered on the first ring. Before the guest finishes saying "I can't get in," the AI has already pulled the reservation by phone number, identified the property, and loaded the access instructions. The AI walks the guest through the entry process conversationally. The call is resolved in 90 seconds. No one wakes up.

A Gap Night Offer

A reservation ends Tuesday. The next reservation starts Thursday. One-night gap.

Before AI: someone would have to notice this gap in Guesty, identify both adjacent guests, draft personalized offers, and send them. In a 130-property portfolio, this happened inconsistently.

After AI: the Revenue Engine identifies the gap automatically. An offer goes to the departing Tuesday guest asking if they want to extend one night. If they decline or do not respond, an offer goes to the Thursday guest asking if they want early arrival. Both offers include the property name, the specific dates, and accurate pricing. If either accepts, the offer acceptance workflow fires and the reservation is flagged for review.

The Inbox on Coachella Weekend

180+ messages in 48 hours across Airbnb and VRBO. Every one gets a draft in under 10 seconds. The team reviews and sends.

The drafts reference specific property details — the parking arrangement at that particular unit, the pool heat instructions for that specific listing, the HOA quiet hours for that neighborhood. Not generic hospitality copy. Actual answers pulled from the knowledge base.

The Pricing Context

Desert Sol runs on the Pro plan at $9 per property per month. At 130+ properties, that is roughly $1,170/month for Voice AI, Inbox AI, Revenue Engine, AI Learning, Payment Audit, and the Dashboard — all six modules, no feature tiers.

The Revenue Engine alone generated 470+ upsell offers. At an average acceptance rate and average offer value, the math on ROI from the Revenue Engine alone covers the platform cost in the first month of operation. The other five modules are the remainder.

What Comes Next

The operations ceiling is gone. Desert Sol can onboard new properties onto Dimora without changing their operations team headcount. Voice AI handles calls for 150 properties the same way it handles 130. The Inbox AI drafts messages for 200 properties with the same response time. The Revenue Engine scans 200 calendars with the same automation.

This is what "AI operations layer" means in the context of a real portfolio. Not a tool that makes the existing workflow faster. A system that handles the operational load, scales without linear cost growth, and improves over time as it accumulates more data.

Six months in, the team is managing more properties than they were in Q4 2025 with the same headcount. The math has changed.

---

Read the full case study with detailed module-by-module breakdowns: Desert Sol Case Study →

Explore the modules behind these results: Voice AI | Inbox AI | Revenue Engine | View Pricing