Is AirDNA (or Rabbu) Accurate?
Accurate enough for market research and for finding comps, the comparable listings you use as evidence, but not precise enough to replace underwriting a specific property yourself. Underwriting means analyzing a property's numbers in detail before you buy. Comp tools estimate earnings from public listing data, so they read a market well and can't know what your listing will do. Here's why, and how to use the number without carrying its optimism into your own analysis.
Matt NunnFounder, Builders Finance11 min readOn this page4 sections
Key takeaways
- Comp-tool numbers are modeled, not a record of what a property actually earned. Because a property's real booking records aren't public, the tools infer earnings from listing signals and other available data. That single fact explains most of where they help and where they mislead.
- They're higher-confidence for the market-level question ("is this a market worth underwriting, and what's its shape?") and for finding comparable listings. They need independent verification for the property-level question ("what will my listing earn?").
- Methodology differs by provider (AirDNA and Rabbu source and model differently), so "the estimate" isn't one thing, and their metrics don't all define revenue the same way.
- The "projected revenue" a tool shows for a specific address is a modeled estimate with a real margin of error, presented as one clean number, which is exactly how it fools people.
- The right workflow: use the tool to find your comps and read the market, then verify by inspecting real listings yourself, then haircut the estimate, discounting it for uncertainty and for the ways a market estimate can overstate your property's stabilized year. Then model the first year lower again, separately.
- A tool's number is a source to check, not an answer to accept.
What these tools actually do
Comp tools don't have your property's books. They model an estimate from public data. Actual reservation records (who booked, for how much, on which nights) are private to the platforms. What's public is the listing: its calendar (each night shows as available or unavailable), its nightly prices, its amenities, and its reviews over time. AirDNA draws on those public listing signals along with other licensed data to model what a listing earned. Rabbu works differently: it assembles comparable listings near your address (same bedroom count, widening the radius when the pool is thin), runs them through a seasonalized revenue-per-available-night calculation, and, for properties listed for sale, shows owner-reported figures alongside its estimate. Different inputs, different models: that is the first reason "is the tool accurate?" has no single answer.
AirDNA is candid about how that works. Its own accuracy page describes a blend of sources — public Airbnb and Vrbo listings, plus privately licensed data from property and channel managers, plus data contributed by individual hosts — cross-referenced to narrow the error. Some figures, like listing counts and asking prices, it can observe directly; occupancy and revenue rely on its proprietary modeling wherever direct operating data isn't available. That's a useful reminder in itself: even within a single tool, some metrics rest on stronger evidence than others. Every headline number these tools produce for a property is therefore an estimate, not a meter reading, and knowing that tells you exactly how much weight each number can bear. Two consequences follow, and they're the root of most misuse:
Occupancy is estimated, not observed, because a public calendar doesn't reveal bookings. Occupancy is the share of available nights that are booked. When a night flips to unavailable, the raw calendar can't say whether a guest booked it or the owner blocked it. The good tools don't ignore this: AirDNA, for one, runs machine-learning models over booking signals (review timing, length of stay, lead times, pricing, seasonality) specifically to tell blocked nights from booked ones. But it's still an inference, and inferences carry error. The ambiguity can distort occupancy, especially for listings that block their slow season and so look busy on the nights they chose to accept. (Even accurate market data describes properties that already have a history. The First-Year Ramp-Up (and Why Year One Isn't Your Stabilized Year) covers how to discount it for a listing that does not.)
"Revenue" doesn't mean the same thing across tools, and it isn't your take-home. Each provider defines its metric its own way, and the differences are real. AirDNA's "revenue" includes the cleaning fees guests paid (revenue that's typically offset in large part by the cleaning expense you pay out) and leaves out platform host fees. Rabbu, by contrast, builds its projection from nightly rates and excludes cleaning fees altogether. Same word, "revenue," describing two different things depending on which tool you're reading. That is exactly why you confirm what a provider's metric contains before you carry any reported ADR or revenue number into a model. (Getting a listing's revenue down to a net rate you actually keep is Step 2 of Estimating STR Revenue Without Fooling Yourself.)
As of October 2026 · Source: AirDNA Help Center (how revenue is calculated, 20 May 2026; how ADR is calculated, 22 May 2026); Rabbu FAQs (updated July 2026).
Where they're higher-confidence — and where they aren't
Match the tool to the question. It's strong at market-level and discovery work, and needs verification before it decides a single property.
| Good uses for comp tools | Verify independently before underwriting |
|---|---|
| Discovering comparable listings to inspect yourself | A specific address's headline "projected revenue" |
| Directional market read — strong vs. weak market, seasonality shape, supply trend | Precise occupancy — it's modeled from public signals, not observed |
| Relative rate benchmarking inside a dense, well-matched comp set | Thin or loosely-matched markets — an average built on few or dissimilar listings is a guess |
| A sanity check against your own bottom-up estimate | Your property's real quality, view, and layout premium the model can't see |
The pattern: the tool is at its best where a modeled average over many comparable listings is genuinely informative (is there demand here, when does it peak, roughly what do comparable places charge), and it needs a human check exactly where you need specificity about one property in a possibly-thin sample. A tool that's very good at "should I be looking in this market?" is being asked to do something else entirely when it's used to answer "what will this house make?" The pattern holds for a reason: two identical properties next door to each other can perform very differently, because the operator and the execution, which no model can see in advance, do a lot of the work.
Read the confidence, not just the number
Every estimate rests on a sample, so read both how big it is and how good it is. Most tools surface some signal of reliability: the number of comparable listings behind an estimate, a market grade or score, a data-quality note. Two things degrade an estimate, and you should check for both. Size: a projection extrapolated from eight listings in a small mountain town deserves a fraction of the trust you'd give one drawn from four hundred in a mature beach market. Quality: four hundred loosely-comparable listings aren't necessarily better than twenty-five genuinely similar ones — more data isn't better data if it's the wrong data. When the sample is thin or badly matched, the tool isn't wrong so much as guessing precisely, and a precise guess is more dangerous than an obvious one, because it doesn't look like a guess.
And because different tools use different inference models, they disagree, which is useful. Pull the same market or property through two tools. When they roughly agree, that's an additional point of corroboration, not proof. Two products can share similar public source data, similar sets of comparable listings or similar modeling biases, and both read high together. When they diverge widely, treat the spread itself as evidence that the accurate answer is "uncertain," and trust your own verified comps over either projection.
The right way to use them
Use the tool for discovery and direction, then do the verification it can't. The single highest-value thing a comp tool does is surface the comparable active listings you can then open, read, and judge for yourself, rather than hand you a revenue projection. In practice: let the tool assemble the candidate comps, then inspect them yourself (their real photos, their review depth, their actual calendars, how their quality stacks up against your property) and build your own comp set from what survives that look. Now the number you carry forward is one you sourced, saw, and can defend, with the tool as the starting point rather than the conclusion.
Verifying the tool's figure this way is the first move of the Builders Finance Underwriting Method, which The STR Deal Underwriting Manual sets out in full.
A tool's estimate is a source to verify, not an answer to accept.
The value of a comp tool is that it finds the comparables and reads the market, work you'd struggle to do by hand. The mistake is letting it also make the decision. Source the number from it, then go verify the listings it's built on with your own eyes; the tool earns your trust on the market, never on the verdict.
The mistake is pasting a tool's "projected revenue" for your address straight into the pro-forma, the projected income statement for the property. It's the fastest way to bring in every source of overstatement at once: modeled occupancy, a revenue metric that may bundle cleaning fees, an average pulled across professional operators and part-timers, and a possibly-thin or loosely-matched sample, all hidden inside a single confident number. The tool's projection is where your revenue work starts, not where it ends.
Your action plan
- Use it to find comps, not to skip them — let the tool surface comparable active listings, then open and inspect each one yourself.
- Read the confidence — check the sample size and the sample quality behind any estimate; distrust precise numbers built on thin or ill-matched data.
- Cross-check a second tool — different models, different biases; treat rough agreement as corroboration and wide disagreement as a warning.
- Confirm what the number includes — get each provider's "revenue" down to a net rate you'd actually keep before it means anything (see Estimating STR Revenue Without Fooling Yourself).
- Discount the occupancy — remember it's modeled; take it down for blocked-calendar effects, and model the first-year ramp as its own, lower year.
- Sanity-check, don't substitute — use the tool's figure to test your own bottom-up estimate, not to replace it.
- Haircut what's left — treat the verified number as a source and apply the Underwriting Method's haircut before it enters the model.
The bottom line
Is AirDNA or Rabbu accurate? For the job they're built to do, reading a market and surfacing comparables, they're genuinely useful. For the job people wish they did, telling you what one specific property will earn, they're modeled estimates, not measurements, because they infer earnings from public data and average across operators you'll never be. So use them the way a professional uses any data source: to find the evidence, not to skip it. Let the tool point you at the comps, verify the comps yourself, haircut the result, and the tools become an asset instead of a trap.
Educational information only — not individualized tax, legal, or investment advice. The worked example is an illustrative model, not a projection or a recommendation.
