1. The basics, in plain English
In Bola's words.
What does a cost segregation study do, and why does it matter more for a short-term rental?
A cost segregation study breaks a property into its parts to find the ones that wear out faster than the ones that stay put. You will replace your appliances long before you replace your foundation. The tax code has a classification system built around that: 5, 7 and 15 years for the parts that wear out faster, and with bonus depreciation you can take those parts as an immediate expense.
A short-term rental is a business, and people forget that. You are running a small hospitality business. If you manage it yourself, it is a second job, but one you can do alongside your W-2. When the average stay is ≤7 days and you materially participate, the deduction from that business can offset your W-2 income.
What did the One Big Beautiful Bill Act change?
I like to say Congress was testing what bonus depreciation would do for the economy. From 2017 it was 100%, and then it phased down: 80% in 2023, 60% in 2024, 40% in 2025. The One Big Beautiful Bill Act made it permanent. If you acquired the property after January 19, 2025, you get 100%.
People get confused here. "I bought it in 2020 but didn't rent it until 2026." The law goes by when you acquired it, not when you started renting. That house stays on the old phase-down, which is 20% in 2026.
California doesn't follow federal bonus depreciation. What should a Bay Area buyer expect on the state return?
California doesn't take bonus depreciation, so the big first-year number is federal. But the study still drives value on your state return. Without a study, you take the whole house and spread it over 39 years. With one, the 5-year parts are spread over 5 years and the 15-year parts over 15. Divide by 5 instead of 39 and you are still deducting a lot more each year.
Illustrative Say a study finds $500,000 of parts: $100,000 of 5-year and $400,000 of 15-year.
| Federal | California | |
|---|---|---|
| Without a study: $500,000 ÷ 39 | $12,821 | $12,821 |
| With a study | $500,000 in year one | $46,667 a year |
California's figure is $100,000 ÷ 5 + $400,000 ÷ 15, shown on a straight line to keep it simple. The actual schedules front-load more.
What has to be true for the deduction to offset W-2 income, and what mistakes do owners make?
The average guest stay has to be ≤7 days. People tell me, "but I do mid-term rentals." You can still take weeklong or 2 week stays. The average across the year is what has to be ≤7.
Material participation usually means more than 100 hours on the rental, and at least as much as anyone else. (Over 500 hours also qualifies on its own.) At my Sea Ranch house I don't have a property manager. I have a handyman and a cleaner, and they coordinate a lot of the guest issues and day-to-day fixes. But I'm the one sending them the issues and approving the work, the whole deal. I'm restocking and messaging guests, and on our trips I'm getting the windows cleaned, the firewood restocked and the septic pumped. I log it as I go. Vacation days with the kids don't count, and a busy cleaner can out-hour you, so keep track.
What is the excess business loss cap, and when does it limit a big year-one deduction?
The deduction itself isn't capped. What's capped is how much of it can offset your wages in one year: $512,000 for married couples filing jointly in 2026, $256,000 for everyone else. Anything over that isn't lost. It carries forward to the next year. The Santa Rosa house in section 3 is a good example: $736,050 in year one, so about $224,050 carries forward.
2. The technology
The lead section.
You spent years at Airbnb building vetting and revenue systems. What did that teach you about cost segregation?
I spent over 11.5 years at Airbnb in a lot of roles, from community and business operations to go-to-market for guest services. One of them was building the vetting and revenue systems for our Experiences business. We thought about Airbnb's standards in 3 ways: safety, regulatory and quality.
Think about the breadth of it. Whether you're jumping out of a plane in Barcelona or just doing a food tour, every single city in the world, not just every state or county, has its own insurance, tax and licensing requirements. We had to solve for that at scale, with millions and millions of permutations. We started in Excel, honestly, before AI, and later used machine learning.
On the revenue side, RevOps meant managing the teams, deciding where we had a human in the loop and where we didn't, and building the systems with engineering and product. It also meant all the math: what goes to which jurisdiction, what gets paid out, what goes to insurance. On guest services, I built one of our first agentic systems. Conceptually, cost seg is the same system, verifying a house instead of verifying an experience.
My friends say I'm uniquely qualified to spend all day designing digital houses. I'm a host. I've had the IPO tax shock. I know systems and design, SQL is my coding language, and I'm really creative. It's that combination of experience, expertise and passion that lets me design the house and still know every micro detail behind it.
What is geospatial engineering for cost seg? Walk us from address to signed study.
Go back to ninth-grade geometry. If you know something is a rectangle and you know 2 sides, you can solve for the rest. That's the geo. The spatial is the relationships: if I know the living room, what's its relationship to the bathroom, and to the furniture in it?
First we crawl for everything we can get on the house: permits, the county assessor, the listing, photos. Then we categorize every single thing, line by line. Then we measure. Once we know what everything is and where it sits, we rebuild the house in 3D, a digital twin. What we can't see, like the wiring from the switch or the plumbing in the walls, we model, because we know how houses are built. We run it over and over, check it against the photos with people in the loop, and only then apply the tax rules.
Which data sources does the platform pull, and how do you reconcile them when they disagree?
Permit records, the county assessor, the listing and its photos, floor plans, comparable land sales, and construction cost data like RSMeans, Craftsman and Homewyse. Plus whatever the owner sends: inspection reports, appraisals, receipts.
When records disagree, we weight them, the same way we weighted signals at Airbnb. A receipt beats a published rate. A county record beats a listing site. An owner's signed answer beats our own estimate. And every pick is recorded with why it won.
What is the 3D asset map, and how do you rebuild a house without someone walking it with a clipboard?
It's like a Matterport tour of the house, except every part is tagged with what it is, what it costs and how it's classified. When you build a house, an estimator does a takeoff: every part you need to put it together. We do that in reverse, from the finished house back to its parts.
The photos confirm what's there. If I see a light switch, I know I can run wiring to it. If I don't know where your panel is, we model it, and if you give us your inspection report, we reconcile it. Every measurement has to tie back to the recorded square footage.
Where does machine vision come in? Give a concrete example.
Most traditional studies say your kitchen is 12% of the house and the appliances are mid-grade. We say: that's a KitchenAid, that specific model, and here's what it retails for. We can tell from the grain whether a floor is real hardwood, LVP or carpet.
What changes the classification is how it's installed, and that's genuinely hard. A floating floor clicks together and can come up and go with you, so it's short-life. Nailed or glued-down hardwood is part of the building. Where the photos can't settle it, the owner's answer does.
You argue AI belongs at the periphery, not in the calculations. Explain the split.
I paid my accountant for my first cost seg, and they never opened any of my files. My background is SQL and databases, so I took my spreadsheet of every dollar we spent and every detail, and built a core property table with all of the house's components. Then I set up the tax rules as a logic system that runs micro calculations on it. I could replay and rerun queries on my "house database" and get the same output every time.
AI doesn't work that way. There is no determinism and no clear logic: ask it the same question 10 times and you can get 10 answers. Pulling a study together is complex enough that we separated the two jobs. AI helps collect and verify the data. The math is deterministic, a database query over every fact we capture about the home.
If you're called into an audit, "AI said so" isn't an answer. You don't always know how AI reached a conclusion. But anyone can check whether the permit it found is real and matches the house, and then follow the logic from that fact to the number.
In an audit, we release a full export of the end-to-end calculations and their sources to your CPA and the examiner. That's essentially it.
What does "deterministic" mean for a client, and why does the same answer twice matter?
It means that if you run it again, you get the same answer, to the cent. If you get audited, you want to show over and over that this is what the output is and exactly how it got there. AI can't tell you how it made a decision. It's good at putting information together in a way that makes you feel good. That's not the same thing.
How does a CPA or IRS examiner trace a number back to its source?
Every line shows where it came from. We got this number from this source. We used the comparable-sales technique, and here are the comps. The quantity, the rate, the classification and the rule behind it are all on the line. That record is the house's own algorithm. Your CPA gets the report and the schedules, and if you're audited, we release the full calculation workbook.
How does your approach map to the IRS Cost Segregation Audit Techniques Guide?
The guide is interesting because it allows so many different methodologies. We outlined every one of them, and our algorithm explores each supportable method for every house and keeps the one with the highest deduction the record can prove. The guide asks for detailed engineering over estimates, cited costs and a reconciliation back to what you paid. That's how we're built.
What is wrong with the $395 to $500 AI-generated studies, and with the traditional percentage method?
If an AI study is only $500, ask what exactly it's doing and what you're paying for. If the IRS asks how it made its calculations or what links it has, it has none of that. Honestly, I'd go with a traditional firm over an AI-driven one, because of the audit.
Not every provider does the engineering. My first study cost $6,600 and was an estimate; no one opened my files. The study on my second property was slightly better, but it still worked in high-level categories: kitchen 12%, appliances 5%. It was a more robust estimate, but still an estimate. Good engineering firms and Unlevered use the same kinds of evidence: inspections, plans, photographs, permits and invoices. Where actual costs aren't available, we estimate from quantities and sourced prices, including retail pricing and construction-cost references like RSMeans.
What's different is how those facts work together. We build a spatial graph that connects each component to its location, the components around it and the evidence behind it. We reconstruct the property as a 3D twin, so you can see that model. Then our property algorithm calculates costs and allocations, evaluates the supported tax treatments, and records the alternatives and the result it selected.
The study comes out of that connected model. You can follow any number back to its component, quantity, cost source and classification, and when new evidence arrives or the property changes, we update the model and rerun the calculations.
When does itemizing beat a top-down percentage allocation?
One of my investors has a home in Healdsburg. Part of how we raised money was saying: let's run our own cost seg and compare it with the one they got from a reputable, traditional provider.
They had invested in a barrel sauna, a plug-in hot tub and a gym in an ADU that had been moved more than once on the property. The traditional firm treated all 3 as if they were built into the house and put them in the 39-year building. Because we separate and itemize every individual item, we moved the sauna and the hot tub, rightfully, to personal property.
How do you protect client data?
We keep client facts minimal. We have a property model, the study and client facts, kept separate, and the client facts are your contact details plus the documents and photos you send us. Stripe handles payment, so we never hold card numbers. Data is encrypted and hosted in the United States, only you and the people you invite can open your study, a CPA gets in only when you grant it, and every action on a study is logged.
Where is the platform going next?
We want to help you maximize your return on the property for as long as you own it. The model of your house stays with it. When you remodel, sell or exchange, you and your CPA start from a house that's already measured and proven.
3. The local example: a Santa Rosa house
Shared with the owners' consent. Bought June 2026 for $1,724,600 and ready for guests October 1, 2026. Figures are from the current accepted study, October 19, 2026.
What did the platform produce for this address?
From the record alone we pulled 99 documents, including 83 listing photos, the floor plan, 14 county permits, the deed and an aerial. We rebuilt 31 rooms across a 3,363 square foot main house and a 468 square foot pool house, and itemized 1,162 lines.
It flagged what didn't fit. The assessor said land was 72.9% of the value, far above any nearby lot sale. One source had the house as 1 story; it's 2. The pool depth wasn't on record.
What is the year-one deduction?
| Purchase price | $1,724,600 |
| Land, from comparable lot sales (23.9%) | $411,778 |
| Depreciable basis, with $20,731 of purchase costs | $1,333,553 |
| Year-one deduction, 100% bonus | $736,050 |
| Usable against wages in 2026, joint filers | $512,000 |
| Carried forward | about $224,050 |
Approximate. Leaves out the rental's own income and expenses.
Which components drive the number, and how is each classified?
| Component | Class | This house |
|---|---|---|
| Pool, pool deck and equipment | 15-year | $245,723 |
| Lawns, planted beds, trees and irrigation | 15-year | $138,500 |
| Fences, deck, paths, courtyard, driveway | 15-year | $63,225 |
| The well | 15-year | $9,831 |
| Appliances and removable finishes | 5-year | $121,551 |
| The house, pool house and septic system | 39-year | $600,711 |
The sleeper hit is landscaping and site work. Here the outdoor work is $457,279 before the premium, nearly 4 times every appliance and removable finish in the house. I always say buy the house with the pool, buy the house with the garden. It's all accelerated right away, and updating it is what guests love and what books.
The 15-year rows are itemized parts; the 15-year share of the Purchase Premium, what the buyers paid above the itemized cost, adds $154,012. The 5- and 39-year rows already include their share. Because it's rented mostly to transient guests, we treat this short-term rental as 39-year nonresidential property, not 27.5-year residential. Speakers, TVs and furniture are furnishings and sit in a separate inventory.
How does the platform handle land on a large lot with mature oaks and vineyard views?
We separate everything that was done on top of the land, the site work, from what's native and part of the land. I love Jeff Lewis and remodeling shows, and if you want a mature olive tree on your property, that can run into the tens of thousands of dollars. That's what cost segregation asks: what went into this property, how much of your investment belongs to each part, and how quickly can each part be depreciated?
The land itself doesn't move. Your soil is land, and so are the oaks that grew there on their own. But the rhododendron you paid $300 to plant by the front door is not land. It's site work: an investment you made to improve the land, and if you ripped it out, it would cost money to put back. People have long counted that kind of investment as land. It isn't. We value the land itself from sales of comparable vacant lots, which carry none of the planting, paving or building. Here that put land at 23.9%, against the assessor's 72.9%.
How do you treat earlier remodels in the price versus renovations the new owner does now? Can you just expense a remodel?
When you buy a house that's already renovated, you bought the renovation. It's in your price and in your study. Anything you add after that is separate, and it's placed in service when the work is finished.
People say, "I did a huge remodel, can I just expense it?" Repairs, yes, and small items under the IRS de minimis rule. A real remodel gets capitalized and classified part by part. You don't need a whole new study; we add the invoices to the one you have. When the house is 39-year nonresidential property, most interior improvements qualify as 15-year property (enlargements, elevators and internal structural framework don't), so with bonus they're deducted the year the work is done. And because the first study already priced what you tore out, you can write off what's left of its basis, plus what it cost to remove it.
People ask me, should I redo my kitchen or my yard? When you're buying, I'd pick the house with the great yard, because landscaping and site work are accelerated right away while built-in cabinets sit in the 39-year building. After you buy, it changes. On a short-term rental most interior remodel work is 15-year property, so a kitchen redo can be written off the year it's done, and the old cabinets you tear out get written off too. So be strategic: know what you're removing and keep every invoice.
What would the owner need to add to sharpen the study?
Anything you can send improves both the deductions and how defensible they are: recent photos, receipts for anything you've done (even furniture, because people forget things), the inspection report, where we see things we normally don't, the closing statement and the appraisal.
Owners who live an hour or more away: what do they need to document, and how does personal use affect the deduction?
Keep a log as you go: date, hours, what you did. Messaging, pricing, restocking, lining up the cleaner and the handyman, maintenance trips.
And watch personal use. If your personal days are more than 14, or more than 10% of the days it's rented, whichever is greater, the house is treated as a residence. Days you spend working substantially full time on repairs and maintenance don't count as personal.
For me, pure personal use is rare. We go for the holidays, and we make maybe 8 or 9 trips to The Sea Ranch a year and 2 or 3 to our Cape Cod cottage, and I rarely go without doing work. Even ordering cleaning supplies counts toward my hours, though a holiday with the family is still a personal day.
How does this compare with your Sea Ranch example?
Our own Sea Ranch house is the best example of what the algorithm does. Same house, same evidence, 3 studies:
| An industry-average split | $361,305 |
| Our first version, remodel left in the 39-year building | $492,818 |
| With the algorithm: interior remodel as 15-year (delivered) | $560,545 |
Before the algorithm, we weren't trying every technique to find the most defensible maximum. Once we turned it on, it found the remodel qualified as 15-year property, and we amended our 2024 return. You can amend when it's your first year; after 2 years of returns it's a change in accounting method instead. Furnishings add another $73,629, and Sea Ranch was on 60% bonus. A house bought today gets 100%.
4. When it makes sense, and when it doesn't
Below what basis does a study not pay off?
My rule of thumb is $350,000 and up, and that's the building, not the price. You'd be shocked. A $2.7 million home in LA can have less basis than a house in Fort Bragg at half the price, just because of the land.
Explain your pricing. Which should a Sonoma County STR buyer choose?
The Basic Study at $2,500 is your basic cost seg. It's defensible and it meets the IRS guidelines, engineered from your photos, delivered as a PDF. At $5,000, 3D GeoEng + Audit, we substantiate everything, run every supportable method, give you the 3D model and stand behind you in an audit. Multifamily & Hospitality starts at $10,000.
I always say go for $5,000. It's a weird thing to give up audit support and the model to save money when the study is saving you hundreds of thousands.
What is a lookback study, and who should consider one?
If you've had a rental and been taking standard yearly depreciation, you can accelerate what's left. A lookback study catches up everything you missed in one year, with a change in accounting method filed with that year's return. No amended returns, just a form your CPA submits. An owner who's run a Russian River rental since 2019 can take it this year.
What are the risks: recapture at sale, audit exposure, losing STR status? How does a 1031 exchange fit in?
Recapture is real, but it's smaller than people think. When you sell, the 5-year parts are taxed at ordinary rates only on what they're worth by then, and they've worn out. Depreciation on the building comes back at up to 25%, and site work can have its own rules, so plan the sale with your CPA. Skipping the study doesn't avoid recapture either: depreciation you could have taken on the building counts whether you took it or not.
Illustrative A study finds $300,000 of 5-year parts, all deducted in year one. You sell 8 years later and those parts are worth $40,000, so $40,000 comes back as income. You deducted $300,000 and gave back $40,000.
In tax terms at 37%: about $111,000 back in year one, about $15,000 paid at sale.
A 1031 exchange can defer the tax, recapture included, by rolling the gain into your next property. Personal property generally doesn't qualify, so plan the exchange with your CPA before you list. The audit risk is a study you can't defend, which is why every number in ours traces back. And if your average stay goes over 7 days (the exception for stays up to 30 days needs significant personal services, which few hosts provide), or you stop participating, that year's loss stops offsetting your wages.
How long does a study take, and what is the latest point in the year a buyer can close and still benefit?
About 30 days. We have people in the loop because reviewing and redoing it matters so much. I always say start as soon as possible. The deadline isn't closing. The house has to be ready for guests by December 31 to count for that year, and the study can finish before you file.
My rule of thumb: close at least 30 days before the end of the year, and be live and hosting with at least 2 weeks left.
5. Working with Unlevered
What does the client need to provide, and how much of their time does it take?
Honestly, just your name, your address, what you paid, when you placed it in service and your listing: Airbnb or VRBO for a short-term rental, Zillow or other rental evidence for a long-term one. We love inspection reports, appraisals, newer photos and receipts, especially if you've remodeled.
How do you work with a buyer's existing CPA? Who signs what?
We sign off on the study as Unlevered Inc. You can share it with your CPA directly in the portal. A lot of times people forget they've already taken depreciation, so their CPA can give us those inputs and we rerun. Your CPA files your return; we're not a CPA firm.
What is the one question every buyer should ask a cost segregation provider?
Ask them to show you the calculation.
Then ask why they do it. I know it sounds weird. This industry is rooted in traditional commercial businesses, and hospitality is so different. I was at Airbnb for a long time, and I saw how much it matters to care about design, experience and the customer first.
Bio
Bola Akinsanya is the founder of Unlevered, a geospatial cost segregation platform. In over 11.5 years at Airbnb, she held executive roles across the company: she ran community and business operations for Homes, one of its first digital marketing campaigns and a taxonomy team, opened its office in India, led global teams, helped launch Experiences and ran go-to-market for guest services. She hosts at The Sea Ranch and on Cape Cod.
Geospatial Engineering for Cost Seg™