Blog

Property management is a $137 billion industry in the United States. It is also, by almost any measure, broken.

If you own a rental property, you have two options. You can manage it yourself, which means fielding tenant calls at midnight, coordinating cleaners who may or may not show up, chasing down maintenance contractors, and spending hours each month on scheduling, communication, and problem-solving. Or you can hire a property manager, who will charge you 20 to 30 percent of your rental revenue to do those same things, often with inconsistent quality and limited transparency into what you are actually paying for.

Most property owners choose option one. Only about 20 percent of rental properties in the United States use professional property management. The number one reason cited by those who manage their own properties is cost. At 20 to 30 percent, the math simply does not work for most owners, especially those with one to four properties.

This is not a new problem. Property management fees have barely changed in decades. The industry has resisted disruption because the core work—coordinating the physical operations of a property—is genuinely difficult. It is local, high-variance, and dependent on real people showing up at real places at specific times.

Today we are announcing something we believe changes this equation permanently. TIDY is launching the world's first end-to-end AI property manager, delivering full property management—everything a traditional PM does—for 3.9 percent of revenue. That is 85 percent less than a traditional property manager.

This is not a chatbot. It is not a feature inside someone else's software. It is a complete property management platform—from cleaning and maintenance coordination to revenue management, guest communication, leasing, and emergency response—powered by AI and backed by human experts for edge cases.

To understand why this matters, and why we believe no one else can do this right now, you need to understand what actually goes wrong in property management and why even the biggest companies that have tried to solve it have failed.

What Happens When You Skip the Hard Part

Consider the largest publicly traded property manager in the United States. They went public via SPAC in late 2021 at approximately $10 per share. The stock peaked near $205, split-adjusted. At its height, the company was valued at roughly $4.5 billion and managed around 44,000 properties.

Then it collapsed. The stock fell 97 percent. Revenue dropped from $1.12 billion to approximately $910 million in a single year. The company accumulated more than $850 million in cumulative losses. They lost 4,000 homes—9 percent of their portfolio—in a single year as owners left over poor communication and disappointing returns. Ultimately, the company was acquired in a distressed sale for approximately $130 million and delisted from the stock exchange.

A $4.5 billion company reduced to a $130 million distressed acquisition. What happened?

The answer is deceptively simple. They solved the easy parts of property management and failed at the hard part.

The easy parts of property management are the things that can be centralized and run from anywhere. Marketing. Revenue management. Dynamic pricing. Guest messaging. Listing optimization. These are inherently digital, text-based tasks. You can run them from a single office for thousands of properties. Many companies have automated them successfully.

The hard part is the physical operations. Scheduling cleaners across thousands of properties in hundreds of markets. Handling no-shows. When a cleaner does not show up two hours before a guest arrives, someone has to solve that problem immediately. Verifying quality. Managing emergency maintenance at 2 AM. Coordinating plumbers, electricians, HVAC technicians, handymen, and dozens of other service providers, each with their own schedules, capabilities, and reliability track records.

This is what brought the company down. Their centralized operations—marketing, revenue management, guest communication—worked fine. But the local operations—the boots-on-the-ground coordination of cleaning and maintenance across hundreds of markets—could not be done cost-effectively with the tools and approaches they had. Every property is different. Every market is different. The variance is enormous, and the cost of failure is immediate and visible. A dirty property means a bad review. A bad review means fewer bookings. Fewer bookings means the owner leaves.

They tried to solve this by hiring more people. More local teams, more coordinators, more managers. The headcount grew, the margins shrank, and the quality still was not consistent enough to retain owners. The fundamental problem was that human coordination of local physical operations does not scale linearly. It scales with friction, exceptions, and compounding complexity.

The Half-Service Model: Skip the Hard Part Entirely

Others in the industry took the opposite approach. They looked at property management, identified the easy half—digital marketing, revenue management, guest communication—and built businesses around just that. They charge 10 to 15 percent, which is less than a traditional PM, and some manage tens of thousands of properties.

But if you use one of these services, you still need to find your own cleaners. You still need to coordinate your own maintenance. You still need to handle the 2 AM emergency yourself. You still need to verify that the work was done correctly. You still need a backup plan when your cleaner cancels two hours before check-in.

These models solve half the problem and leave the hardest half to the owner. That is why they can charge less than a traditional PM. They do less. And critically, they do not fundamentally change the economics for the property owner. You are still paying for coordination—just less of it—and doing the rest yourself.

This is the fundamental tension in the industry. Full-service property managers charge 20 to 30 percent because the physical coordination is genuinely expensive and difficult. Half-service companies charge 10 to 15 percent because they skip it. Neither model serves the property owner well. One is too expensive. The other is incomplete. Neither fundamentally changes the underlying cost structure.

Why We Spent 12 Years on Cleaning and Maintenance

TIDY was founded over 12 years ago. For the vast majority of that time, we focused on one thing: building the technology, AI, and operational infrastructure to coordinate cleaning and maintenance at scale.

This was not an accident. We understood from the beginning that the physical operations layer was the bottleneck for the entire property management industry. It was the part that could not be centralized. It was the part that required local knowledge, real-time decision-making, and the ability to handle thousands of edge cases. It was the part that everyone else either failed at or avoided.

Here is what 12 years of focused work on this problem produced:

• Digital twins of every property. Not a listing description. A structured digital model of the property including rooms, appliances, surfaces, access procedures, and complete maintenance history. When a tenant reports a problem, the AI already knows what is in the property, when it was last serviced, and what is likely wrong.
• AI models trained on data from over 100,000 properties. These models predict issues before they happen, optimize scheduling decisions, and make judgment calls that previously required an experienced human coordinator. When a cleaner cancels, the system does not just send an alert. It has already identified the best backup providers based on proximity, availability, quality scores, and cost, and it dispatches a replacement automatically.
• A network of vetted, performance-tracked service providers. Not a directory. A managed network where every provider has a quality score based on hundreds of data points: on-time rates, customer satisfaction, photo verification of completed work, consistency over time. The AI routes work to the best-performing providers and reduces allocation to underperformers. This is a meritocratic system that rewards quality.
• 98 percent no-show prediction accuracy. The system knows which jobs are at risk of a no-show before they happen and proactively arranges contingencies. This is the kind of capability you cannot build in a year or two. It requires years of operational data across thousands of providers and tens of thousands of jobs.
• 96 percent or higher on-time reliability. Industry average for individual contractors is approximately 70 percent. Professional firms achieve about 92 percent. We are at 96 percent and climbing, because the AI continuously optimizes provider selection and scheduling.
• Human experts for edge cases. AI handles the vast majority of routine coordination decisions. The genuinely complex situations that require judgment, empathy, or creative problem-solving are handled by dedicated human account managers. This is not a fully automated system with no human safety net. It is an AI-first system with human expertise where it matters most.

This is the foundation. The hard part. The part that brought down the largest publicly traded PM. The part that half-service companies skip. We spent 12 years building it because we knew that whoever solved local physical operations at scale would be positioned to deliver full property management at a fundamentally different cost structure.

Adding the Easy Parts on Top of the Hard Part

Once you have solved the coordination of physical operations at scale, adding the traditionally easy parts of property management is architecturally straightforward.

Revenue management and dynamic pricing are API-driven analytics tasks. Guest communication is a text-based workflow perfectly suited to AI. Leasing and tenant screening are structured decision processes. Sales calls and owner onboarding are conversational AI applications. PMS operations—calendar syncing, booking management, financial reporting—are software integration problems.

None of these are trivial. They all require careful engineering. But they are fundamentally different in nature from the local physical operations problem. They can be centralized. They can be fully automated. They do not require local knowledge or real-time physical coordination.

We built these capabilities through a 12-agent AI architecture. Twelve specialized AI agents, each responsible for a specific domain of property management operations, working together to deliver the full scope of what a traditional property manager does. Cleaning coordination. Maintenance management. Revenue optimization. Guest communication. Owner reporting. Emergency response. Leasing. Financial management. Quality assurance. Vendor management. Sales. Account management.

Each agent is purpose-built for its domain, trained on our 12 years of operational data, and integrated with the others so that decisions in one domain inform decisions in all others. When a maintenance issue is detected, the maintenance agent coordinates the repair, the guest communication agent notifies the affected guest, the revenue agent adjusts pricing if the property will be temporarily unavailable, and the owner reporting agent updates the owner—all automatically, all in real time.

This is what end-to-end means. Not one AI feature bolted onto a traditional workflow. Not a chatbot that answers tenant questions. A complete, integrated system that handles every aspect of property management from physical operations to digital operations to financial operations.

Why 3.9 Percent Is Possible

Traditional property managers charge 20 to 30 percent because human coordination is expensive. A human property manager can effectively manage approximately 50 properties. They need an office, a phone system, support staff, and local teams in every market they serve. The coordination labor alone costs approximately $128 to $160 per property per month.

AI changes this math entirely. Our AI-driven coordination costs approximately $37 per property per month. A single human account manager, handling only the edge cases that AI escalates, can oversee approximately 150 to 200 properties. The marginal cost of adding the next property is near zero.

This is not a loss leader. This is not subsidized by venture capital. This is the actual cost structure when AI handles the vast majority of coordination decisions. The 3.9 percent fee covers our costs and generates healthy margins at scale because the underlying economics are fundamentally different from a human-labor-intensive model.

At 3.9%, on a property generating $40,000/year in rental revenue, the management fee is $1,560/year. A traditional PM would charge $8,000 to $10,000. The owner saves $6,440 to $8,440 per year. Per property. Every year.

Why This Is So Difficult to Replicate

Several companies are using the phrase "AI property manager." We researched this carefully before launching. Here is what we found.

Some startups are building AI tools for property managers. They augment human PMs. The PM company still exists, still charges 20 to 30 percent, and uses these AI tools to be more efficient. The property owner sees no price reduction.

Others are building AI-enhanced property management software. Property managers use their platform. Again, the PM still exists, still charges traditional rates.

Still others—traditional PM software platforms—are adding AI features to their existing systems. Useful features, but the property owner still needs a property manager to use the platform.

None of these companies are delivering property management. They are selling tools to property managers. The distinction matters because the property owner's experience and cost does not change. The PM might be more efficient internally, but they are still charging 20 to 30 percent.

With TIDY, there is no middleman using our tool and marking it up. The AI is the coordination layer. The property owner pays 3.9 percent and gets full-service management.

Why can we do this and others cannot? Because you cannot coordinate cleaning and maintenance at scale with just software. You need the digital twins. You need the predictive models trained on years of operational data. You need the vetted provider network. You need the quality verification systems. You need the no-show prediction and automatic contingency dispatch. These are not features you can add in a product sprint. They are the result of over a decade of focused work on the hardest problem in property management.

What This Means for Property Owners

If you currently use a property manager, you can get the same service—or better—for 85 percent less. On a property generating $40,000 per year, that is $6,440 to $8,440 back in your pocket annually.

If you currently manage your own property because professional management was too expensive, you can now afford it. At 3.9 percent, the cost is low enough that virtually every property owner can justify it. No more midnight phone calls. No more scrambling for a backup cleaner. No more coordinating maintenance yourself.

If you are a homeowner—not a landlord—and you have never considered property management because the model was never designed for you, we are building for you too. Professional coordination of your cleaning and home maintenance, powered by AI, at a price that makes sense for the first time.

The property management industry has served only about 20 percent of properties because the other 80 percent could not justify the cost. At 85 percent less, that math changes for everyone.

What This Means for the Industry

We believe AI will compress the cost of property management coordination by approximately 90 percent over the next several years. This is not a prediction about distant future technology. It is happening now. We are delivering 85 percent cost reduction today, and the AI is still improving.

When coordination costs drop 90 percent, the industry transforms. The same pattern has played out in travel agencies, stockbroking, and tax preparation. Technology collapses the cost of coordination, prices drop, adoption goes from partial to near-universal, and the market consolidates around technology-first operators.

Property management will follow this pattern. The coordination layer—what PMs charge 20 to 30 percent for—will be automated. The physical work—the actual cleaning and maintenance—will still need to be done by skilled professionals. But the coordination of that work will be handled by AI at a fraction of the current cost.

For existing property managers, this is not a threat if you choose to adapt. The local expertise, market knowledge, and relationships that good PMs have built over years are genuinely valuable, especially during this transition. We believe the best path forward is partnership, not resistance. But the standalone economics of charging 20 to 30 percent for coordination that AI can do for 3.9 percent are not sustainable long-term.

For service providers—cleaners, maintenance technicians, handymen—the work itself is not going away. AI replaces the coordinator, not the cleaner. What changes is how work is allocated. In an AI-coordinated system, performance data drives everything. The providers who do great work consistently will get more business than ever. This is a more meritocratic system, and it rewards the people who are genuinely good at their craft.

The Foundation Was the Hard Part. The Future Is Everything Else.

We spent 12 years solving cleaning and maintenance coordination at scale because we knew it was the prerequisite for everything that follows. You cannot build an AI property manager on top of a foundation that does not include physical operations. You just end up with another half-solution.

The largest PM companies proved that scaling physical operations with human labor alone does not work. Half-service models proved that skipping physical operations entirely leaves the owner with half a solution. The dozens of AI startups building tools for property managers are proving that making the existing model more efficient does not change the fundamental cost structure for property owners.

The only way to deliver full property management at 85 percent less is to have AI handle both the physical coordination and the digital coordination, end to end. And the only way to do that is to have spent years building the operational data, provider networks, predictive models, and quality systems that make AI-driven physical coordination reliable at scale.

That is what we built. That is why it took 12 years. And that is why we believe this is a genuinely new category—not an incremental improvement on what exists, but a structural change in how property management works and what it costs.

We are calling it the world's first end-to-end AI property manager because, to our knowledge, no other company is delivering a complete property management platform—including physical operations—powered by AI, at this price point. Others are building AI tools. Others are building AI software. We are delivering AI property management.

Full service. End to end. 3.9 percent.

Property management is a $50 billion per year industry in the United States alone. If you look at Carpathia Capital’s chart of industries most likely to be automated by AI, property management sits right there in orange—not yet in the deep red of "already disrupted," but clearly in the zone of "this is happening now."

TIDY has already shipped the product to automate it. We’re not waiting for AGI or some future breakthrough. The AI models that exist today are sufficient. What we wanted to figure out—and what this post is about—is: what does the compute actually cost to replace $50 billion of human labor? And how does it all fit in a single server rack by 2028?

We’re going to walk through this using short-term rentals as our worked example— it’s the segment we know best—and then extrapolate to the full industry. First, the setup cost: what it takes to get a property fully live and AI-managed. Second, the cost to serve: the ongoing weekly token spend for guest operations, pro coordination, maintenance, compliance, and everything else. Third, the industry scale: what happens when you multiply by every property in the country, and the world. And finally, the hardware: how many GPUs, how many racks, how many dollars.

Our baseline assumption: a typical short-term rental with one turnover per week. Some properties turn twice or three times a week, some every other week—but one per week gives us a clean ballpark.

Why Tokens Are the Right Way to Think About This

A token is roughly ¾ of a word. Every time an AI reads context, reasons about a problem, or generates a response, it consumes tokens. Today, frontier model pricing sits around $3–15 per million input tokens and $10–75 per million output tokens, depending on the model. But these costs have been falling 10x roughly every 18 months—and that trend shows no sign of stopping.

So when we say "this task costs 50,000 tokens," what matters isn’t today’s price. What matters is the scale of the work. A 50K-token task is fundamentally small. A 5M-token task is fundamentally large. The dollar cost will keep shrinking, but the token footprint tells you what’s easy versus what’s hard.

Part 1: The Setup Cost

Getting a short-term rental property fully live and AI-managed

When a property owner comes to TIDY to get their short-term rental fully managed, there’s a one-time setup process. Let’s break down each step and estimate the token cost.

1. Analyze the Listing

The AI needs to pull in the listing from Airbnb, Vrbo, or wherever it lives, understand the property layout, amenities, house rules, and compare everything against best practices. What’s the listing missing? What could be better? Are the descriptions accurate and optimized?

The listing text itself is maybe 2,000–5,000 tokens. Best practices context and comparison frameworks add another 10,000–20,000 tokens. With a couple of reasoning loops where the AI thinks through recommendations, you’re looking at:

Listing analysis: ~50,000–100,000 tokens

That includes input tokens for reading the listing and best practices, plus output tokens for generating a detailed analysis and recommendations. At today’s prices, that’s roughly $0.15–$1.00. Trivial.

2. Photo Review & Cleanup

This is one of the more interesting setup tasks. The AI reviews listing photos to identify issues—bad lighting, clutter, poor angles—and can either suggest improvements or use generative AI to enhance them.

Photo analysis uses vision models. Each image is roughly 1,000–2,000 tokens to process (depending on resolution). A typical listing has 20–40 photos. For each photo the AI needs to assess quality, suggest improvements, and potentially generate enhanced versions.

Photo review: ~80,000–160,000 tokens (vision)
Photo enhancement: ~$0.50–$2.00 in image generation API calls

The image generation side (calling something like Google’s Imagen or similar) is priced per image rather than per token, but it’s still on the order of a few cents per image. For a full listing refresh, you might spend $0.50–$2.00 on photo enhancement total.

3. Analyze Local Regulations

Short-term rental regulations vary wildly by city, county, and state. The AI needs to research what’s required: permits, tax registrations, occupancy limits, noise ordinances, HOA rules, and more.

This involves pulling in regulatory documents (which can be lengthy—some municipal codes run thousands of pages) and extracting the relevant sections. The AI doesn’t need to read the entire code—it retrieves the relevant sections, typically 20,000–50,000 tokens of regulatory text, then reasons about applicability.

Regulatory analysis: ~100,000–200,000 tokens

This is one of the heavier setup tasks in terms of tokens because of the volume of source material. But it’s still a one-time cost, and the regulatory knowledge can be cached and reused across properties in the same jurisdiction.

4. Generate & File Regulatory Paperwork

Once regulations are understood, the AI can help generate permit applications, tax registrations, and compliance documentation. This is mostly form-filling and document generation—relatively light token work.

Regulatory filings: ~20,000–50,000 tokens

5. Set Up Dynamic Pricing

Configuring a pricing engine—connecting to market data, setting base rates, seasonal adjustments, event-based pricing, minimum stays. This is mostly software, not AI reasoning. The AI’s role is light: understanding the owner’s pricing preferences and translating them into configuration.

Pricing setup: ~10,000–20,000 tokens

6. Connect Financial Systems

Linking bank accounts, payment processors, and accounting systems. This is pure software integration—API calls, OAuth flows, webhook setup. The AI might have a brief conversation to confirm which accounts to connect, but the actual work is code, not reasoning.

Financial setup: ~5,000 tokens (essentially zero)

7. Set Up Smart Locks & Access

Connecting smart locks, generating access codes, setting up automated code rotation for guests. Again, this is primarily software—API integrations with lock providers. The AI helps configure the rules (when to send codes, how to rotate them), but the heavy lifting is traditional software.

Smart lock setup: ~5,000–10,000 tokens

8. Configure the TIDY Platform

This is setting up the scaffolding in TIDY itself—creating the property profile, cleaning checklists, maintenance schedules, pro assignments, notification preferences, and agent configurations. It’s a combination of software configuration and AI-assisted customization.

Platform configuration: ~20,000–40,000 tokens

9. Analyze Historical Messages & Extract Patterns

This is potentially the most token-heavy setup task. If the property has existing operations, the AI needs to ingest historical guest messages, pro communications, and owner correspondence to understand patterns—what questions guests typically ask, how the owner likes things handled, what issues come up repeatedly.

A property with a year of operations might have thousands of messages. Even summarizing and extracting key concepts from this history could require processing significant volumes of text. However, this can be done with summarization chains—process messages in batches, extract key themes, then synthesize.

Historical analysis: ~200,000–500,000 tokens

The range here is wide because it depends entirely on how much history exists. A brand-new listing has zero history to analyze. A property with years of guest reviews and pro communications could be at the high end. But even at 500K tokens, we’re talking about $1.50–$7.50 at today’s prices.

10. The Sales Conversation

Finally, there’s the back-and-forth conversation with the property owner during setup. Understanding their goals, preferences, deal-breakers, and specific requirements. This is a multi-turn conversation that might span several sessions.

Sales & onboarding conversation: ~50,000–150,000 tokens

Adding It All Up: Total Setup Cost

Plus roughly $0.50–$2.00 in image generation API calls for photo enhancement.

At today’s frontier model pricing (let’s use $10/M input, $30/M output as a blended average for a capable model), and assuming roughly a 3:1 input-to-output ratio:

Total one-time setup cost: roughly $5–$20 per property

Five to twenty dollars in AI compute to do what traditionally requires hours of human work—analyzing listings, researching regulations, reviewing photos, configuring systems, and having detailed onboarding conversations.

Part 2: The Ongoing Cost to Serve

What it costs in tokens to operate a property week after week

Setup is a one-time cost. The real question is: what does it cost to actually run the property, week in and week out? This is where a property manager spends their time—and where the token economics tell the most interesting story.

We’ll assume one turnover per week (~52 guests per year) and break everything into per-turn costs and amortized ongoing costs.

Guest Operations: Inquiries, Bookings & Communication

As new inquiries come in, the AI needs to evaluate them against your house rules, availability, guest history, and pricing. Much of this is handled by software—calendar lookups, rule engines, pricing algorithms—so the token budget is really just the AI applying judgment to edge cases and generating personalized responses.

Guest inquiry & booking: ~5,000–15,000 tokens per guest

During the stay itself, the token spend is modest. The average guest sends maybe 2–3 messages—check-in questions, local recommendations, a minor issue. These are mostly solved with text inputs and outputs. Each exchange runs about 3,000–7,000 tokens including context.

Guest communication during stay: ~10,000–20,000 tokens per guest

Roughly 1 in 50 guests has something that escalates—"show me the video," "I need to see what’s going on"—requiring video analysis or a more complex multi-step resolution. When that happens, it might be 100,000–200,000 tokens. But amortized across all guests:

Escalation handling (amortized): ~2,000–4,000 tokens per guest

Pro Coordination: The Biggest Ongoing Token Spend

This is honestly the most token-heavy part of ongoing operations, and it’s worth explaining why.

For every turnover, there’s messaging with your cleaning pro: scheduling confirmation, reminders, hearing about any issues they found, answering questions. Pros often ask things they could find by looking at their checklist—but they frequently don’t, and we don’t always get feedback to know what’s unclear. So there’s a steady stream of back-and-forth messaging. On average, maybe 3–5 message exchanges per turn, each consuming tokens for context and response.

Pro messaging per turnover: ~15,000–30,000 tokens

Where the token budget really spikes is when you need a new pro. Your regular cleaner can’t make it, or you’re replacing a provider, and now the AI needs to search through potentially hundreds of candidates, evaluate profiles, check availability, negotiate rates, and onboard someone new.

A lot of the initial filtering can be done by structured database queries and hard-coded rules—location, availability, ratings, price range. But the AI needs to analyze the filtered shortlist, assess fit, and handle the outreach conversation. When this happens, it’s a 200,000–500,000 token event. But it only happens maybe 4–8 times per year for a typical property.

New pro search (amortized): ~5,000–15,000 tokens per week

Post-Clean Review: Photos & Quality Verification

After each cleaning, you’re reviewing before-and-after photos to verify quality. A typical turn generates 5–20 photos that need analyzing—checking that everything matches the checklist, spotting issues, confirming the property is guest-ready.

Occasionally there’s a 5-minute video walkthrough to review. Photo analysis runs about 1,500–2,000 tokens per image through a vision model. Video is heavier when it happens but doesn’t happen every turn.

Post-clean photo/video review: ~15,000–40,000 tokens per turn

Maintenance & Restocking

Maintenance is intermittent—a plumber here, an HVAC call there, a handyman for a broken shelf. Each maintenance job has roughly the same token profile as a cleaning coordination: dispatching, messaging the pro, reviewing results. Restocking (toiletries, linens, cleaning supplies) follows a similar pattern—detect what’s needed from checklists or photos, trigger orders or pro tasks.

For a typical short-term rental, you might have maintenance pop up every 2–4 weeks and restocking every 1–2 weeks. These are in the same order of magnitude as other job types.

Maintenance (amortized): ~5,000–15,000 tokens per week
Restocking: ~5,000–10,000 tokens per week

Compliance & Regulatory Upkeep

The initial regulatory analysis happens at setup, but there’s ongoing compliance work: periodic filings, tax payments, permit renewals, occupancy reports. These compliance systems are often old school—log into a government portal, fill out a form, send a check. If the AI is using web browsing to handle these, that consumes tokens for page content, navigation decisions, and form filling.

A typical compliance filing is roughly the equivalent of 3 minutes of web browsing—maybe 20,000–50,000 tokens per session. These happen monthly or quarterly.

Compliance upkeep (amortized): ~5,000–10,000 tokens per week

There are also noise alerts—a noise sensor triggers, and the AI needs to message the guest and handle the response. This is mostly software-driven (sensor triggers the workflow) with a small token budget for crafting the message and managing any back-and-forth.

Noise alerts & compliance triggers (amortized): ~2,000–5,000 tokens per week

Financial Monitoring & Owner Communication

Monitoring financial transactions, categorizing income and expenses, spotting anomalies, generating owner reports. Much of this is a software problem—database queries, accounting rules—with the AI handling edge cases and generating human-readable summaries.

Financial monitoring: ~3,000–8,000 tokens per week
Owner updates & communication: ~5,000–10,000 tokens per week

Marketing, Pricing & Listing Optimization

Dynamic pricing adjustments, listing syndication across platforms, upsell opportunities— this is overwhelmingly a software problem. Pricing algorithms run on structured data. Listing syndication is API calls. The AI’s role is occasional: maybe tweaking listing copy seasonally, suggesting pricing strategy changes, or identifying upsell opportunities.

Marketing & pricing: ~2,000–5,000 tokens per week

Last-Minute Changes & Exceptions

Last-minute cancellations, guest reschedule requests, shuffling pro assignments—these create incremental work but don’t happen every turnover. Maybe 1 in 4 weeks you have some kind of schedule disruption that requires the AI to recoordinate. When it happens, it’s a 30,000–60,000 token event. Amortized:

Schedule disruptions (amortized): ~3,000–8,000 tokens per week

Adding It All Up: Weekly Cost to Serve

Scaling that up:

• Per month: ~330K–780K tokens
• Per year: ~4.3M–10.1M tokens

The Annual Token Budget for One Property

Now we can put the full picture together—setup plus a year of operations:

These numbers come from production. We use blended frontier model pricing at roughly $15 per million tokens (a 3:1 input-to-output ratio at $10/M input and $30/M output), plus image processing costs.

The wide range reflects real-world variance. We’ve measured per-property token consumption as low as 5 million tokens per year for well-established properties with routine operations and optimized prompts. But we err on the side of the high end of the range—up to 10x—to increase accuracy and account for complex edge cases, heavy reasoning tasks, regulatory research, difficult pro searches, and properties with unusual requirements.

We expect these costs to decrease over time. Better models, tighter scaffolding, smarter automation, and growing confidence in lighter-weight approaches for routine tasks all push token consumption lower. The low end of our range today is where we expect the average to settle.

Compare That to a Traditional Property Manager

A short-term rental generating $50,000/year in revenue with a traditional property manager at 10–15% fees pays $5,000–$7,500 per year in management costs.

The AI token cost to do 99% of the same work? In production, we’ve measured as low as $75 in Year 1. Our conservative budget—erring high for accuracy—is up to $2,150. Even at the conservative end, the AI compute cost is less than half of traditional property management fees.

At our optimized production levels, that’s a 25–100x cost advantage in raw compute. Even at our most conservative budget, it’s a 2–3x advantage—and improving fast. At TIDY’s 3.9% fee ($1,950/year on a $50K property), there’s massive margin to deliver better service at a fraction of the industry’s traditional cost.

And remember: token costs are falling roughly 10x every 18 months. By 2028, the same operational workload might cost $7–$20 per property per year in AI compute. Multiply that across a $50 billion industry and the economics are staggering.

Where the Tokens Actually Go

Looking at the ongoing cost to serve, a clear pattern emerges. The biggest token consumers are:

1. Pro coordination (~40% of weekly tokens)
   Regular messaging, new pro search, quality verification. This is the largest single category because it involves the most human-like reasoning—understanding context, negotiating, evaluating quality, handling exceptions. Pros often don’t check their checklists and instead ask questions, which means more back-and-forth than you’d expect from a pure software standpoint.
2. Guest communication (~25% of weekly tokens)
   Inquiries, during-stay messages, and the occasional escalation. Most of this is straightforward text exchanges, with rare spikes for complex issues.
3. Everything else (~35% combined)
   Maintenance, restocking, compliance, financial monitoring, owner updates, marketing, and schedule disruptions. Individually small, collectively significant, and heavily assisted by traditional software. The AI fills in the gaps that software can’t handle alone.

What Most People Get Wrong

When people hear "AI property management," they imagine a massive compute requirement—some kind of always-on superintelligence monitoring every aspect of every property. The reality is far more prosaic.

Most of what a property manager does is coordination work—scheduling, messaging, checking, confirming—and the average turn generates maybe 100K–200K tokens of AI work. That’s roughly 75,000–150,000 words of reading and writing. For context, that’s about the length of one or two novels. Once a week. For an entire property.

A huge portion of the work isn’t even AI at all—it’s traditional software. Calendar systems, pricing algorithms, payment processing, smart lock APIs, database queries for pro matching. The AI is the reasoning layer on top, handling the judgment calls, the natural language communication, and the visual inspection. The scaffolding underneath is just code.

Part 3: The Industry-Scale Token Budget

What it costs to AI-manage every short-term rental in the US—and the world

We’ve established that a single property costs roughly 5–11 million tokens per year to manage with AI. Now let’s zoom out. What does the total token demand look like if you wanted to AI-manage every short-term rental?

How Many Properties Are We Talking About?

The US has approximately 1.8 million unique short-term rental properties (per AirDNA and industry data). Some sources count up to 2.5 million listings across platforms, but that includes cross-listing on Airbnb, VRBO, and Booking.com simultaneously. We’ll use 1.8 million unique properties.

Globally, Airbnb alone has about 7.7 million active listings. Add VRBO (2M+), Booking.com’s alternative accommodations, regional platforms, and direct-booking properties, and adjust for cross-listing overlap, and the global figure lands around 10 million unique short-term rental properties.

US Token Demand

Using our production range—from optimized deployments at the low end to conservative 10x budgets at the high end:

Global Token Demand

Compare That to What the Industry Spends Today

Let’s put these numbers in context. The US property management industry as a whole— short-term rentals, long-term rentals, commercial, HOA management—is roughly a $50 billion per year industry. The short-term rental slice we’ve been analyzing represents about $2–$3 billion of that (10–15% management fees on ~$20 billion in US STR revenue). Globally, STR management fees run $12–$23 billion per year (on a $125–$155 billion market).

The AI token cost to replace the STR management layer, based on our production data:

• US STR (optimized): $135M–$300M (vs. $2–$3B traditional) — a 7–22x cost advantage
• US STR (conservative 10x): $1.4B–$3B — roughly matching traditional costs today, but falling fast
• Global STR (optimized): $750M–$1.7B (vs. $12–$23B traditional) — a 7–30x cost advantage

Even at our most conservative production budget, the AI cost is at parity with traditional management—and token costs are falling roughly 10x every 18 months. By 2028, even the conservative end drops to a 7–20x advantage. At optimized levels, it’s 70–200x.

And that’s just the STR segment. The same token economics apply to the broader $50B US property management industry. The tasks are similar—coordination, communication, scheduling, compliance, quality control—they just have different nouns. The token math scales linearly.

Optimized vs. Conservative: What’s the Difference?

Our "optimized" numbers are what we’ve achieved in production on well-established properties with routine operations, tight prompts, and efficient scaffolding. Most property management tasks are straightforward—scheduling, simple messaging, photo review, financial monitoring—and run efficiently on capable models.

Our "conservative" budget is up to 10x higher. We err on this side to increase accuracy across the full range of real-world conditions: complex regulatory analysis requiring multi-step legal reasoning, difficult pro searches with nuanced candidate evaluation, guest escalations requiring judgment, properties with unusual requirements, and the long tail of edge cases that don’t show up in averages.

In practice, most properties run much closer to the optimized end. We expect the conservative budget to tighten over time as models improve, our scaffolding gets more efficient, and we gain confidence in lighter-weight approaches for tasks that currently get the heavy reasoning treatment.

Putting It in Perspective

To AI-manage every short-term rental in the United States for a full year—setup, operations, guest communication, pro coordination, maintenance, compliance, everything— requires somewhere around 9 to 200 trillion tokens, depending on where you sit in our production range.

That sounds like an enormous number. But consider: the major AI labs are already processing hundreds of trillions of tokens per month across all their customers. Even at our conservative 10x budget, the entire US short-term rental industry’s management needs represent a modest slice of current global AI capacity.

The compute exists. The models are capable. The economics work. What’s left is execution—the software scaffolding, the integrations, the refinement of every edge case. That’s what TIDY has been building for 12+ years.

Part 4: GPUs, Hardware, and Real Dollar Costs

How many GPUs does it take to manage every rental on Earth—and what does it actually cost?

We’ve established the token budget. Now let’s translate that into physical hardware. How many GPUs would you actually need? What would they cost? And what if you just bought the tokens from Anthropic instead?

First: How Fast Can Today’s GPUs Generate Tokens?

For a large, capable model (think 670B+ parameters—something in the class of DeepSeek-R1 or comparable to the frontier models used for complex reasoning), here’s what current and upcoming NVIDIA hardware can do:

• NVIDIA B200 (Blackwell): ~3,750 tokens per second per GPU on a 670B model. A DGX B200 (8 GPUs) hits ~30,000 TPS total. Cost: ~$45,000 per GPU.
• NVIDIA GB200 NVL72 (rack-scale Blackwell): With NVLink connecting 72 GPUs, effective per-GPU throughput improves to ~4,500 TPS. The full rack does ~1.5M TPS on a 120B model, less on larger models. Cost: ~$3 million per rack.
• NVIDIA GB300 NVL72 (Blackwell Ultra): 45% faster than GB200 on DeepSeek-R1—roughly ~5,800 TPS per GPU. Available now.
• NVIDIA Vera Rubin (shipping H2 2026): NVIDIA claims a 10x reduction in inference token cost and 5x raw FLOPS over Blackwell. Conservative estimate: ~15,000–25,000 TPS per GPU on a 670B-class model. 288GB HBM4 per GPU. Draws ~1,800W. Price: likely $50,000–$70,000 per GPU.

Translating Token Demand Into GPU Count

Our global STR token demand spans a wide production range: 80 trillion tokens per year at our optimized levels, up to 800 trillion at the conservative end. Let’s use the optimized midpoint for the hardware analysis—the conservative end simply scales linearly (10x the GPUs, 10x the racks):

• Average throughput (optimized): 80T tokens ÷ 31.5M seconds/year ≈ 2.5 million tokens per second
• Peak throughput (2x for daytime clustering): ~5 million TPS
• With 50% headroom over peak: ~7.5 million TPS capacity needed

Property management isn’t uniform around the clock—turnovers cluster in mornings, guest messages spike in evenings, compliance work happens during business hours. But it does span global time zones, which smooths things out. A 2x peak-to-average ratio with 50% headroom is conservative.

Option 1: Build It With Blackwell (Available Now)

Using B200 GPUs at ~3,750 TPS per GPU on a large capable model:

Twenty-eight racks at our optimized production levels. At our conservative 10x budget, that’s 280 racks—still a small footprint in a single data center. For context, a single hyperscaler data center might house thousands of racks.

Option 2: Wait for Vera Rubin (H2 2026)

Using projected Vera Rubin throughput at ~20,000 TPS per GPU:

Five racks at optimized levels, 50 at our conservative budget. Either way, you could fit the entire world’s short-term rental AI management infrastructure in a single server room. The annual cost—hardware, power, operations, everything—is about what a mid-size property management company spends on salaries.

Option 3: Just Buy the Tokens — Claude Opus 4.6 Pricing

Most companies won’t build their own GPU clusters. They’ll buy tokens from an API provider. Let’s price this out using Claude Opus 4.6—Anthropic’s most capable model—at full retail pricing. This is deliberately the most expensive option to show the ceiling.

Opus 4.6 pricing: $5 per million input tokens, $25 per million output tokens. Assuming a 3:1 input-to-output ratio (typical for property management—lots of reading context, shorter responses):

Prompt caching is particularly effective for property management because many operations share context—the same property details, checklist templates, house rules, and regulatory frameworks get loaded repeatedly. Anthropic’s caching drops input costs by 90% for cached tokens ($0.50/M instead of $5/M), and most property context qualifies.

But You Don’t Need Opus for Everything

Here’s the thing: most property management tasks don’t require the most powerful model on the planet. Scheduling confirmations, routine guest messages, photo reviews, and financial categorization work great on smaller, cheaper models. You really only need Opus-class reasoning for complex tasks: regulatory analysis, difficult guest escalations, nuanced pro evaluation, and tricky judgment calls.

A realistic model mix might be:

• ~70% of tokens on a fast, efficient model (like Haiku-class at $0.25/$1.25 per MTok) — routine scheduling, simple messages, standard reviews
• ~25% on a mid-tier model (like Sonnet-class at $3/$15 per MTok) — pro coordination, guest communication, photo analysis
• ~5% on Opus-class ($5/$25 per MTok) — regulatory analysis, complex escalations, new pro search evaluation

The Full Comparison

Let That Sink In

The US property management industry spends $50 billion per year on human labor. Just the short-term rental slice—$2–$3 billion—could be replaced with AI. At our optimized production levels with smart model routing and batch processing, the US cost is $18–$23 million. Even at our conservative 10x budget, it’s $180–$230 million—still a 10x cost advantage.

At optimized levels, a Vera Rubin cluster could AI-manage every short-term rental in America for $2–$3 million per year with five server racks. At the conservative end, that’s 50 racks and $20–$30 million—still a fraction of the industry’s cost.

Globally? $11–$140 million per year depending on where you are in the production range. Even the high end is a rounding error compared to $12–$23 billion in traditional management fees. And that’s just the STR segment—the same economics apply across the full $50 billion US property management industry.

Important Caveats

These GPU calculations assume you’re running open-source or self-hosted models. You can’t run Claude Opus on your own hardware—it’s a proprietary model available only through Anthropic’s API. The GPU cluster numbers represent what you’d need if running an equivalently capable open model (like DeepSeek-R1 or future open alternatives).

The API pricing path is the realistic one for most companies. The GPU path illustrates the underlying physics: this workload is not computationally large. The gap between the ~$12M self-hosted cost and the ~$95–330M API cost represents the model provider’s margin, their R&D costs, and the premium for the best models.

Also: none of this includes the software platform itself—the scheduling engine, the integration layer, the pro network, the compliance tracking, the digital twin. The tokens are just the AI reasoning layer. The scaffolding underneath is what makes those tokens useful.

A Rounding Error on Global GPU Production

Here’s perhaps the most striking way to frame all of this. NVIDIA’s Vera Rubin platform entered full production in early 2026—ahead of schedule—with volume shipments starting in H2 2026. Analyst estimates project roughly 30,000 Vera Rubin NVL72 racks shipping in 2026 (that’s ~2.16 million GPUs), ramping to ~100,000 racks (~7.2 million GPUs) in 2027.

At our optimized production levels, we need 5 racks to AI-manage every short-term rental on Earth. At our conservative 10x budget, 50 racks.

5–50 racks out of 30,000 produced = 0.017–0.17% of Year 1 Vera Rubin production

The entire global short-term rental property management workload is a rounding error on NVIDIA’s output. Even if you expand to all property types—long-term rentals, commercial properties, HOAs, the full scope of property management worldwide—and use our conservative budget, you might need 500 racks. That’s still 1.7% of Year 1 production.

The hardware constraint doesn’t exist. The compute to automate a $50 billion industry is being manufactured at a scale that makes the workload invisible. The only question is who builds the software scaffolding to put those tokens to work—and that’s been TIDY’s focus for over a decade.

Why We Expect These Costs to Decrease

Our production range is wide—up to 10x between optimized and conservative—because we deliberately err on the side of higher token budgets to increase accuracy. But we’re actively working to narrow that range downward, and multiple forces are converging to make it happen:

• Better models: Each generation of frontier models delivers better reasoning with fewer tokens. Tasks that required long chain-of-thought reasoning a year ago now complete in a single pass.
• Tighter scaffolding: Every month, we optimize our prompts, reduce unnecessary context loading, and improve the software layer that handles what doesn’t need AI at all. The scaffolding gets more efficient with every deployment.
• Smarter automation: As we accumulate production data, we identify which tasks can be handled by smaller, cheaper models—or by pure software with no AI reasoning at all. The model mix shifts toward efficiency.
• Growing confidence: Early in a deployment, we over-invest in reasoning to ensure quality. As we validate performance on specific task types, we can safely reduce token budgets without sacrificing accuracy.

The direction is clear: our conservative budget tightens toward our optimized levels, and our optimized levels continue to improve. Meanwhile, the underlying cost per token keeps falling roughly 10x every 18 months. Both curves bend the same way.

Why 99%, Not 100%—and the Human Cost of the Last 1%

We’ve been careful to say 99% automation, not 100%. This is deliberate. The last 1% isn’t a technical limitation—it’s a design choice.

If the house catches fire, do you want the AI to handle it without notifying the owner? If there’s a major insurance claim, a lawsuit from a guest, or a decision to terminate a long-standing pro relationship—these are moments where owners may genuinely want to be involved. Going much beyond 99% automation in this industry isn’t just unnecessary, it’s arguably not even desired.

So what does the remaining 1% cost in human labor?

The US property management industry employs roughly 250,000 people in the short-term rental space. At 99% automation, that’s 1%—about 2,500 peoplehandling the edge cases that need a human touch.

The Full Organization: Not Just Automation

Automation doesn’t mean there are zero humans. Even a fully AI-managed property management platform needs people—just far fewer, and doing different work.

You still need account managers and sales representatives to help property owners get set up, ease the transition from traditional management, and provide the human relationship that some owners want. You need people building and improving the software platform itself. The whole organization might run on roughly $20M in G&P labor costs.

There’s also a significant channel market: an estimated 10,000+ property managers and partners who, for various business reasons, need to be the ones using the software rather than the property owner directly. They may manage portfolios for absentee owners, handle commercial relationships, or serve markets where direct-to-owner AI management isn’t the right fit today. This partner channel needs its own sales force—maybe another $20M in labor.

At optimized production levels, call it $200 million per year, all-in, to AI-manage every short-term rental on Earth—compute, humans, sales, engineering, everything. At our conservative budget, $1–$1.3 billion. The global STR management industry currently spends $12–$23 billion. The broader US property management industry? $50 billion.

That’s a 10–115x cost reduction on STR alone, depending on where you sit in our production range. And the range is tightening downward as models improve and our scaffolding gets more efficient.

The Product Already Exists

Everything we’ve described isn’t theoretical. TIDY has shipped this product. It’s in market today. We charge 3.9% per completed job—compared to the industry average of 10–15%. The cost savings of a software-first approach make this price point possible, and we expect it to go down further as token costs continue to fall.

It took us 12 years to build the scaffolding. We don’t think it will take competitors 12 years to match—the AI tools are better now, the playbook is more visible. But the product exists today. It hasn’t saturated the market yet. There will be competitors. And within five years, any property management business that hasn’t adopted AI automation will be dead.

This requires zero innovation from AI model providers. No breakthroughs needed. No AGI required. The models we have today are sufficient. The scaffolding just needs to exist—and it does.

That’s just property management. Just one segment on Carpathia’s chart. One fraction of one day of GPU production. How many other $50 billion industries will fall to the same math?

A Personal Note

Building this has been simultaneously the most fun of my career and one of the most terrifying things I’ve witnessed. People debate what happens if we get AGI, what happens if AI can do everything. Those conversations are important.

But they’re missing the point. Sufficient intelligence is here today. A $50 billion industry can be automated with a single rack of GPUs and the right software. Not theoretically—we’ve done it. Most industries are not safe. Not from some hypothetical future superintelligence, but from the AI that exists right now, combined with years of domain-specific scaffolding. The disruption isn’t coming. It’s here.

Jevons’ Paradox: Why Cheaper Means Bigger

Here’s what makes this story even more interesting. Today, roughly 80% of the addressable market is not using a property manager at all. They manage their own rentals because traditional property management is too expensive. When you’re paying 10–15% of revenue, plenty of owners decide it’s not worth it and do everything themselves.

This is a classic case of Jevons’ Paradox: when you make something dramatically cheaper, demand doesn’t stay flat—it explodes. A 90% cost reduction doesn’t just save money for existing customers. It pulls in the vast majority of property owners who were previously priced out.

We expect to see two massive waves of new demand:

1. Rental property owners coming off the sidelines. The 80% who manage their own properties today will start adopting AI-powered management when the cost drops from thousands of dollars per year to hundreds or less. Why handle scheduling, guest messages, and pro coordination yourself when AI can do it for 3.9%?
2. Consumers entering the market for the first time. Homeowners who have never used a property manager—because property management was historically only for landlords and investors—will start using AI-powered management tools for their own homes. Cleaning scheduling, maintenance coordination, vendor management. These are problems every homeowner has. They’ve just never had an affordable solution.

The addressable market isn’t 1.8 million US properties or 10 million globally. When the cost of management drops 90%, the market could be 5–10x larger. TIDY is already positioned for this with our consumer offering—being one of the first platforms to bring the benefits of property management to regular homeowners at a price point that makes sense.

2028: One Rack to Automate an Entire Industry

The numbers above already feel absurd—five Vera Rubin racks to manage every short-term rental on Earth. But there’s a real possibility it gets even more extreme.

NVIDIA announced Groq LPU integration into the Vera Rubin platform, which is designed to deliver up to 35x more inference throughput for certain workloads compared to standard GPU inference. Groq’s Language Processing Units are purpose-built for sequential token generation—the exact workload that dominates property management AI (reading context, generating responses, reasoning through decisions).

Combine that 35x inference improvement with the ongoing algorithmic and architecture improvements in model efficiency—distillation, mixture-of-experts, speculative decoding, quantization advances—and the math starts to look like this:

• Today (Blackwell): 28–280 racks for global STR management
• H2 2026 (Vera Rubin): 5–50 racks
• 2028 (Vera Rubin + Groq + algorithmic gains): Potentially a single Feynman rack

At optimized levels, a single rack. At our conservative budget, maybe ten. Costing $10–$100 million. To AI-manage the entire global property management industry.

A $50 billion industry of human labor, replaced by $10–$100 million in hardware. A 500–5,000x cost reduction for the US alone—not in theory, not in a decade, but within 2–3 years at current trajectories.

There are reasons this might not play out exactly this way. Model quality could plateau. Groq integration might not deliver the full 35x for real-world mixed workloads. Regulatory or market dynamics could slow adoption. But from a pure token-cost and hardware perspective, the math says it’s very possible. The trend lines are not subtle—they’re exponential, and they’re converging.

The Bottom Line

Property management is a $50 billion industry. Here’s how the math works out:

• One property, one year: 5–113M tokens in production, $75–$2,150 at today’s blended pricing. We’ve run as low as the bottom of that range and budget at the top for accuracy.
• Every US short-term rental: 28–280 Blackwell racks today, 5–50 Vera Rubin racks in H2 2026
• Every STR on Earth: $11–140M/year self-hosted, $95M–$1.2B/year via API — compared to $12–$23B in traditional management fees
• By 2028: Vera Rubin + Groq + algorithmic gains could compress the entire global STR workload into 1–10 racks costing $10–$100M
• Share of global GPU production: 0.017–0.17% of Year 1 Vera Rubin output — the hardware to automate a $50B industry already exists in absurd surplus

The product already exists—TIDY shipped it. The market hasn’t saturated yet. Competitors will come. But the math is the math: a $50 billion industry of human labor can be replaced by a handful of server racks and 12 years of software scaffolding—and the costs are heading down, not up.

Take care of your properties like it’s 2026, not 1996.

Sources & References

• AirDNA US Short-Term Rental Outlook Report — 1.8M unique US properties, market performance data
• Anthropic Claude API Pricing — Opus 4.6, Sonnet, and Haiku token pricing
• NVIDIA Vera Rubin Platform Announcement — 10x inference cost reduction, HBM4, H2 2026 availability
• Inside the NVIDIA Vera Rubin Platform — Technical blog with detailed specs and performance claims
• NVIDIA GB300 NVL72 (Blackwell Ultra) — Specs and AI reasoning performance
• CoreWeave GB300 NVL72 DeepSeek-R1 Benchmarks — 6x+ performance gain on DeepSeek-R1 workloads
• NVIDIA Groq 3 LPX Technical Blog — Groq LPU integration, up to 35x inference throughput per megawatt
• NVIDIA’s $20B Groq Acquisition Analysis — Next Platform deep dive on the deal
• NVIDIA Feynman Architecture Roadmap — 3D-stacked GPUs, Rosa CPU, 2028 timeline
• Jevons’ Paradox — Economic principle: efficiency gains increase total resource consumption
• a16z: LLMflation — LLM Inference Cost Trends — ~10x cost decline per year for equivalent AI performance
• Epoch AI: LLM Inference Price Trends — Rigorous analysis showing median 50x/year decline in token costs
• Grand View Research: Short-Term Vacation Rental Market — Global market sizing ($134.5B in 2024)

TIDY 7.0 delivers meaningful improvements to the chat experience and job management views, along with a range of bug fixes that address issues customers have been running into across scheduling, messaging, and mobile navigation.

New Features & Improvements

• Expandable Chat Improvements: The in-app chat now surfaces suggestions in a clearer, more trustworthy format, making it easier to understand what actions can be taken on your properties. Response handling has also been improved for faster, more reliable replies.
• Loading Indicators on Action Buttons: Action buttons now display a loading state while your request is being processed, so you always know something is happening and can avoid accidentally tapping twice.
• Improved Job Page & Notification History: The job detail page has been redesigned for better clarity around quotes, job scenarios, and status updates. Notification history now shows a more complete timeline of events — including pro cancellations and schedule changes — directly on the job view. The Pros page is also now available on mobile.

Bug Fixes

• Fixed an issue where the Schedule List view showed "No scheduled items found" even when jobs were scheduled.
• Fixed future jobs incorrectly displaying a "No Show" status.
• Fixed the Sign In button overlapping the Add a Property form on native apps.
• Fixed an error when opening a conversation in the Messages view ("Error sending message. Please try again later").
• Fixed an issue where cancelling a job caused other job cards on the page to stop loading until a manual refresh.
• Fixed the Subscription section not loading on certain accounts.
• Fixed the feedback rating card pushing content off-screen when a user tried to rate a message.
• Fixed pro names intermittently disappearing from the Messages page conversation header.
• Fixed a case where pro cancellations were not appearing in the job notification history.
• Fixed a token conflict that could cause login issues when a user was signed into both the homekeeper and customer apps on the same device.

TIDY 6.9 adds a Hostfully integration and fixes several front-end issues across booking, authentication, workflows, and device compatibility.

New Features & Improvements

• Hostfully integration: You can now connect your Hostfully account to TIDY, syncing property management data the same way Guesty and other PMS integrations work.

Bug Fixes

• Booking & auth in guest mode: Fixed several issues where users in guest mode could reach the booking view without completing sign-up. Also fixed the auth modal being unresponsive for up to 20 seconds when booking from the schedule list page.
• Job frequency display (Pro app):Fixed an issue where all plan-type job opportunities showed “Recurring” instead of the actual job frequency.
• Workflow editor on mobile: Fixed multiple issues in the Workflows section on mobile and native apps — editing workflows, loading trigger outputs, and adding template workflows all work correctly again.
• Chat loading in private browsing:Fixed an issue where the chat panel would get stuck on “Loading conversations...” when opened in an incognito or private browsing window.
• Google & Apple sign-in reliability: Fixed an issue where signing in via Google or Apple would fail on the first attempt. Sign-in now works on the first try.
• Samsung Galaxy Z Fold layout: Fixed navigation and scroll issues on Z Fold devices, including the properties list not scrolling and the schedule page not rendering correctly.

TIDY 6.1 is a focused follow-up to our 6.0 launch. This release addresses dozens of bugs and adds quality-of-life improvements across the app — particularly around scheduling, checklists, mobile, and the native app experience.

New & Improved

• Chatbox overhaul: Links in chat are now clickable and navigate within TIDY. Photos display inline with an expanded view. Chat switching is more reliable, and the interface has been cleaned up on both desktop and mobile.
• Global search: Search has moved to the top nav bar and now works across the entire app. TIDY logo added to navigation for better branding.
• Referral codes: Partner referral links now show a confirmation banner and auto-apply the discount at checkout.
• Schedule filter improvements: The filter bar on the Schedule page now handles multiple active filters without overlapping content.
• Past job checklists: Checklist text on past jobs is now selectable. Removed unnecessary edit button and cleaned up the UI.
• Accessibility: Improved support for screen readers and assistive technology across the app.

Bug Fixes

Checklists & To-Dos

• Checklist name and description changes now reflect immediately without a page refresh.
• Custom fields on checklist items can now be edited reliably, including select-type fields.
• Marking items as important via the star icon now persists correctly.
• All lists now appear on the Default Checklist page — previously some were missing.
• Fixed a flash of the "Add Property" button when opening a checklist.

Scheduling & Jobs

• Cancelling or rescheduling a job with filters active no longer requires a page refresh to see updates.
• Schedule filters now persist correctly when navigating away and returning.
• Quotes now display on the Job page even when "Best Available Pro" is the placeholder.
• Share Job links now work correctly (previously used a broken path).
• Fixed visual glitch with grey area on the book job info page (mobile).
• Custom field entries on properties now save correctly.

Account & Authentication

• Password reset page no longer blocked by the account selection modal.
• Sign-in options are now visible after dismissing the account selection on mobile.
• Fixed "For Service Pros" link on account selection — now points to the correct URL.
• Privacy link in the auth modal now works.
• Onboarding messages no longer reappear after signup if previously dismissed.
• Fixed an issue where Pros couldn't check in due to an incorrect identity verification redirect.

Mobile & Native App

• Back button from Automations now returns to the More menu instead of Integrations.
• Long property names no longer overlap header icons on Overview.
• Deleted addresses in Guest mode no longer persist on the Overview page.
• "Request Pro" button on native no longer becomes unresponsive after selecting time/frequency.
• "All Properties" dropdown on Overview no longer causes cards to overlap with the property list.
• Delete address confirmation dialog is no longer cut off on native.
• Audit Logs now have a back button on native.
• Quick suggestion text in chat now displays in full before sending.
• Fixed layout issues in agent onboarding (message alignment, button overlap).
• Pros link in search now navigates to the correct page.

First, a little history. With the launch of TIDY 5.0, we achieved best-in-industry results at being able to customize and operate your cleaning and maintenance at any scale—from one property to thousands. We're grateful to the great customers who've been working with us along the way.

However, at TIDY we've been working since our founding to fully automate cleaning and maintenance, with the idea that one day things will be done with robotics. Along this journey we've taken several major milestones, and the latest is a big one.

Agents Operating Throughout Your Properties

With TIDY 6.0, you can now have agents operating throughout your properties to help you deliver amazing cleaning and maintenance results. We have 10 agents today that span the full gamut of everything you need to manage your cleaning and maintenance operations. This includes scheduling agents for helping schedule, maintenance agents (something we previewed in October) to manage messaging with your pros, restocking agents, and more.

This allows property owners and managers, for the first time, to take a fully hands-off approach to cleaning and maintenance orchestration.

Customize Absolutely Everything

As always, we allow you to customize absolutely everything. We know every property manager has their own unique workflow requirements, checklist requirements, vendor requirements, and more. TIDY can accommodate any of those—completely autonomously.

Feature Highlights

That's the high-level vision. Let's talk about the features. They're too many to list individually, but here are some highlights:

• Checklists & Maintenance (Renamed)
  We've changed the name of To-Dos and Tasks to Checklists and Maintenance to clarify the differences between these things. Checklists are things you want people to get done. Maintenance is any type of item that comes up—either reported by a pro, reported by a guest or tenant, or added manually by you.
• Smarter Checklists
  Checklists are now dramatically easier to work with. You can more easily use the same checklist across all your properties with smart matching for restocking items.
• Agents Everywhere
  As we mentioned, on almost every page there's an agent now. You can specify what triggers the agent and what actions it needs to take—in plain English. Our team will review these, make adjustments, and let you know as needed. We know it can be a little tricky to figure out exactly what you should say in any circumstance.
• Pro Message Automation (Now in Public Beta)
  This is now leaving our private beta and going into public beta. Anyone can now start using message automation with their pros. We think this is the biggest time-saver we've ever built.
• Easy Feedback, Everywhere
  We've made it really easy to chat with us, with a message box on many pages. We believe strongly that the primary thing you should be doing in most cases is telling us what you want—what's going wrong, what's not happening the way you want. We can most easily make changes that way.

How Are We Able to Do This?

The advancement in agentic LLMs is a huge part of it. We couldn't do what we're doing today without them. However, we also have our outstanding concierge team, which acts like a team of cloud humans to resolve issues. This has been an important way that we've developed the product, and we continue to be able to provide excellence even in those cases where LLMs aren't quite good enough yet.

TIDY's core infrastructure is the backbone.Without TIDY's core infrastructure, you would not be able to do this. Nobody can just have a black box of ChatGPT managing your properties and hope to have it do anything—it doesn't work like that. You need strong scaffolding with many, many features that we've introduced over many years to allow people to customize things. We're strong believers that every home is a unique snowflake with unique requirements. There's no such thing as an easy or simple job—they just don't exist.

So the combination of software scaffolding, combined with LLM reasoning, and a fallback to humans means that we can achieve superhuman results for a fraction of the cost.

Simpler Plans, Dramatically Lower Cost

With these developments, we've been able to improve the quality of our Autopilot service as well as reduce the cost. As part of this change, we simplified our plans and made them even easier. Before, you had to pay for automation with flat fees or other fees per message and such. Now it's much simpler:

We charge just 3.9% per completed cleaning and maintenance job.

Compare that to the industry average of 10-15%. This can be a whopping 60-90% cost reduction over a traditional property manager—and all of this is with higher quality standards: advanced AI to predict issues, advanced AI to respond more quickly, humans available 24/7, and things that are just not possible with traditional property managers.

What's Next

With TIDY 6.0, we're certainly not done yet with our journey to fully automating cleaning and maintenance. We're happy to be pretty transparent with our roadmap: throughout 2026, you can expect a lot of improvements to the agentic experience. Both the UI to help you understand what your agents are doing, making sure we can actually accommodate every case possible, and making it easier than ever to set up and get going.

We've made it easy to give us feedback in the app, so send us a message about anything you'd like changed. We look forward to bringing the future of cleaning and maintenance into today.

The Industry Is Shifting

Property management is undergoing a dramatic shift. As things automate, we expect costs to come down, which is going to dramatically change the industry. TIDY is a partner you can count on to stay ahead of the curve and make sure you're competitive as the world around you moves to AI. Using TIDY gives you a competitive advantage.

Lower costs also allow us to serve consumers—being the first, really, to ever allow consumers to access the benefits of property management. A single place to call and ask about any cleaning and maintenance needs, and have it getting done for you, at one low cost.