AI did not fix my STR business in two weeks. It took about 12 months of building, from February 2024 through January 2025, before the system worked, and my first real proof came on June 10, 2025, when it helped me deliver 5 professional action plans in 24 hours. This is the account of that stretch, written in 2025 and updated in September 2026: what worked, what wasted my time, and the process I would follow if I started over.
In February 2024 I sat in my office and thought, "I'm J. Massey. I'll point AI at my STR business and it will organize my thoughts in two weeks." I had more than 15 years in real estate, an STR focus since 2016, and 10,000+ operators trained. I assumed the hard part was behind me. It wasn't.
What I ended up with is something I call a Cognitive Operating System: a set of prompts, data tables, and automations that holds my decision rules so I don't have to keep them in my head. It changed how I run the business and how I think about automation. It also nearly didn't happen, because I almost paid consultants to do the work instead.
How Long AI Actually Took to Work in My STR Business
The two-week plan turned into five distinct phases. Each one ended when I hit a problem I couldn't talk my way around with a better prompt.
The table below lays out those phases in order, with the length of each one counted from the dates I logged at the time.
Phase | Dates | Length | What I was doing |
|---|
Naive start | February 2024 | ~1 month | Asking AI to "organize my STR thinking" with no structure |
Integration failures | March to May 2024 | ~3 months | Typeform intake forms feeding Airtable, breaking repeatedly |
Data integrity | June to August 2024 | ~3 months | Rules for clean, validated records before anything else runs |
Workflow architecture | September to November 2024 | ~3 months | Custom workflows in Make.com connecting the pieces |
Operating system | December 2024 to January 2025 | ~2 months | Assembling v5.3 of the Cognitive Operating System |
First delivery test | June 10, 2025 | 1 day | 5 action plans delivered in 24 hours |
Add the build phases up and you get roughly 12 months against a plan of half a month. The AI part went fast. The slow parts were bad data, fragile connections between tools, and my own undocumented decision rules.
The Consultant Math That Stopped Me
Before I committed to building, I priced the alternative. I wanted a full business action plan for an STR portfolio: market analysis, positioning, revenue strategy, an operations audit, a technology plan, and an implementation roadmap. The SBA's own guide to business plans describes the same kind of structure, with market analysis near the front and financial projections behind it.
These ranges are my own estimates at the time, not a market survey. Treat them as one operator's read, not a price list.
Deliverable | My estimated cost | My estimated time |
|---|
Market analysis | $2,500 to $5,000 | 2 to 3 weeks |
Competitive positioning | $1,500 to $3,000 | 1 to 2 weeks |
Revenue optimization strategy | $3,000 to $7,500 | 3 to 4 weeks |
Operational efficiency audit | $2,000 to $4,000 | 2 weeks |
Technology integration plan | $1,500 to $3,500 | 1 to 2 weeks |
Implementation roadmap | $1,000 to $2,500 | 1 week |
Total | $11,500 to $25,500 | 10 to 14 weeks |
The low column adds to $11,500 ($2,500 + $1,500 + $3,000 + $2,000 + $1,500 + $1,000). The high column adds to $25,500. My personal estimate for the full engagement was around $50,000. That figure is my guess, not a quote I received.
I was reaching for my credit card when a different question stopped me: what if I could teach AI to think like my best property manager? That question cost me a year of evenings instead of $25,000 or more in fees. I don't regret the trade, but I want to be clear that it was a trade. Time has a price too, and a paid consultant would have delivered a plan months sooner.
If you hire outside help, remember that most consultants work as independent contractors, not employees. The IRS guidance on contractor status is worth reading before you sign anything, because the classification affects who handles taxes and what paperwork you owe.
What Didn't Work
Most of my first year was failure, and the failures were more useful than the wins. Here are the ones that cost me the most time.
Asking AI to "organize my thinking"
My first month produced long, confident, generic output. The model had no idea what my properties earned, how I screened guests, or what "good" looked like to me. Vague input produced vague output. Nothing it wrote was wrong enough to reject and nothing was specific enough to act on.
Form-to-database connections that broke quietly
From March to May 2024, my Typeform intake forms fed Airtable, and the connection failed often enough that I stopped trusting the data. Fields mapped to the wrong columns. Records arrived half-filled. Some submissions never arrived at all.
If you are wiring forms into Airtable today, read Airtable's automation documentation before you build. It states several behaviors that catch operators off guard:
Automations do not run on records that already existed when you turned the automation on.
"When a record is created" and "When a record is updated" triggers can fire as soon as someone starts typing, before a record is complete.
Failed runs count against your monthly run allowance, the same as successful ones.
Those three facts alone explain a lot of "my automation is broken" complaints I hear from hosts.
Automating before the data was clean
I kept building workflows on top of messy records. Every workflow inherited the mess. It took me three months, June to August 2024, to accept that data integrity comes first and automation comes second.
Trusting pricing and competitor output without guardrails
AI is good at producing a pricing recommendation that sounds reasoned. That is different from being right about your market on a given weekend. Any rate suggestion that goes live without a floor and a ceiling you set is a risk you are choosing to take.
What Worked
Once I stopped expecting magic, a few practices produced reliable results.
Writing down my decision rules first
The breakthrough was documenting how I make routine decisions: how I price a slow Tuesday, when I approve an early check-in, what I say to a guest who reports a broken appliance. Once those rules existed in writing, AI could apply them consistently. Before that, it was guessing.
Validating data at the point of entry
I moved validation to the front of every workflow. A record that is missing required fields never triggers anything downstream. That one change stopped incomplete records from breaking the workflows after them.
Building one workflow at a time with a test record
In Make.com, I built each workflow alone, tested it against a known record, and only then connected it to the next one. Slow, but every piece worked before it had to depend on another piece.
Keeping a human review step on anything a guest or client sees
The NIST AI Risk Management Framework is a voluntary framework, released January 26, 2023, for managing AI risk, and one of its core ideas is that people should evaluate AI output before relying on it. I apply that in a simple way: AI drafts, I approve. The 5 action plans I delivered on June 10, 2025 were drafted with the system.
The Four Areas I Built For
I focused the build on four areas. Here is what each one means in plain terms, and where I would urge caution.
1. Revenue optimization
The goal is demand-based pricing that reflects your actual market, not a generic algorithm. Platform tools already do part of this. Airbnb's Smart Pricing, for example, adjusts nightly prices based on demand within a minimum and maximum you set. My advice: set those limits yourself, review them monthly, and treat any AI pricing suggestion as a draft.
2. Guest experience automation
This covers messages and flows based on guest history and preferences. It is where many operators leave money on the table, because the same repeat guest gets the same generic welcome message three stays in a row. Start with your five most common messages and personalize those before anything else. For wording, my piece on why your STR sounds like 47 others covers brand voice.
3. Competitive monitoring
The idea is to watch comparable listings and turn what you see into recommendations. In practice, the recommendation is only as good as the comparables you choose. Pick them by hand, check them quarterly, and don't let a tool decide who your competition is.
4. Operational efficiency
This is AI that applies your decision patterns to routine choices. It is the most useful of the four and the one that depends most on the documentation work above. No written rules, no automation worth having.
Why ADHD and Autism Shaped the Systems
While building this, I noticed something about my own wiring. I have ADHD and I'm autistic, and the accommodations I built for myself turned out to be good business design.
I needed visual hierarchy so I could see what mattered first. I needed a place to offload working memory so I wasn't holding every open loop in my head. I needed decision frameworks so I didn't re-argue the same choice every week. Those three needs became the backbone of the operating system.
Any operator running multiple units has the same problems at a different volume. You forget which guest asked for a crib. You re-decide your cleaning-fee policy every time a guest complains. A system that holds those answers helps everyone, whether or not your brain works like mine.
The Process I'd Follow If I Started Over
If you want AI to carry real weight in your STR business, this is the sequence I would use today. It would have saved me most of that first year.
Write down 10 decisions you make every week. Pricing, guest approvals, maintenance calls, refunds. One paragraph each on how you decide.
Map where your data lives. List every tool: booking platform, messaging, forms, spreadsheets, accounting. Note which ones talk to each other and which ones you copy between by hand.
Clean one data source before you connect anything. Pick the table you trust least and fix required fields, duplicates, and naming.
Add validation at the point of entry. Incomplete records should stop, not flow downstream.
Build one automation, test it against a known record, and run it for two weeks. Watch the run history for failures before adding the next piece.
Give AI your written rules, not open-ended questions. Ask it to apply your policy to a specific case, then compare its answer to what you would have done.
Keep a human approval step on anything guest-facing or money-related. Review drafts, pricing changes, and refunds before they go live.
If you want a second set of eyes on where your own operation stands before you build anything, book a complimentary one-on-one diagnostic call. It is free. If you want a written plan afterward, the action plan is $997.
Frequently Asked Questions
How long does it take to set up AI for a short-term rental business?
In my case, about 12 months of building from February 2024 through January 2025, plus several more months before I trusted it for client work in June 2025. A single workflow can go live in a few weeks. A full system that applies your rules across pricing, messaging, and operations takes much longer because the data and documentation work comes first.
Can AI replace an STR consultant?
For me it replaced most of the planning work I had priced at an estimated $11,500 to $25,500. It did not replace judgment. I still review everything, and a consultant would have delivered a finished plan months faster than I built my own system.
What is the biggest mistake when automating an STR business?
Automating before my data was clean. Every workflow built on messy records inherited the mess, and I spent three months, June to August 2024, going back to fix data integrity. Clean one source first, then connect it.
Should I let AI set my nightly rates?
Only inside limits you set. Tools like Airbnb's Smart Pricing already adjust rates by demand within a minimum and maximum you choose. Set those limits yourself and review them monthly.
Do I need to know how to code?
No. I built with Typeform, Airtable, and Make.com, which are no-code or low-code tools. You do need patience for testing, and you need to read each tool's documentation on how triggers and run limits work.
What should I automate first?
Start with your most repeated guest messages or your intake form. Both are high-volume, low-risk, and easy to check. Leave pricing and refunds for later, and keep a human approval step on them when you get there.
Sources
Related Reading
