- Investors want to set up a short- or mid-term rental. Finding the right property — one where the numbers actually work — is the hard part, and nothing on the market evaluates a listing the way an STR investor does. Perch is the AI workflow that does that evaluation.
- The goal was never an app. The workflow screens every listing near a park entrance — price, financing and DSCR, condition from photos, setup cost, nearby Airbnb ADR — and the output is a weekly newsletter: a ranked, underwritten shortlist sent to people who want to invest in STR.
- Perch is the clearest example of the studio's lifecycle — and of the stop criteria that decide whether a product keeps its engineering time.
Context & problem
Plenty of people want to set up an Airbnb. Very few can find the right property to do it with. A listing that looks fine as a home is a bad short-term rental if the nightly rate doesn’t cover the mortgage, the occupancy in that market is seasonal, or the city has quietly restricted STR permits. Evaluating all of that per listing is a half-hour of work with four browser tabs and a spreadsheet, and the investor has to do it for dozens of listings to find one worth touring.
Generic listing search (Zillow, Redfin) doesn’t understand yield. Professional analytics platforms do, but are priced for funds, not for someone buying their second property. The individual investor falls in the gap: they spend evenings evaluating, and still hear about the right listing after it’s pending.
The markets that matter most are the ones hardest to evaluate from a distance: cottage country. Lake Tahoe, Joshua Tree, the Smoky Mountains — places where demand is real but seasonal, where a cabin two miles from the one you’re looking at rents for half as much, where the county and the town have different STR rules, and where the investor is almost never local. Those are exactly the markets where a spreadsheet and a Zillow tab fail, and where a workflow that already knows the comps, the occupancy curve and the permit map earns its keep.
So I built the evaluation as an AI workflow: the thing the investor does by hand per listing, run automatically across every listing in a target market, every week.
Insight
The user doesn’t want a better search; they want to stop searching. Every interview converged on the same thing — “just tell me what’s worth looking at this week.” That’s a push product, not a search product, and it doesn’t need to be an app. The right interface is a weekly newsletter, and the right engine is a workflow that runs the investor’s underwriting on their behalf.
It was also obvious that the shortlist’s value could be tested before an agent existed to produce it.
Mini PRD
| Problem | Small STR/MTR investors spend hours a week manually scanning listings and still miss high-yield opportunities. |
| Users | Individual investors — often remote — targeting short- and mid-term rentals in vacation markets such as Lake Tahoe, Joshua Tree and the Smoky Mountains. |
| Goals | A weekly newsletter with a ranked, underwritten shortlist · Replace the investor’s spreadsheet · Build an audience of STR investors before building any software they have to install. |
| Non-goals | An app as the product · Offer generation · Property management · Agent/brokerage features (that’s Studio 3Sixty5). |
| Core requirements | Investor profile: target markets, budget, financing, minimum yield · Listing search anchored on park proximity · Financing model with DSCR · Photo-based condition evaluation · Setup and renovation cost formulas · Nearby Airbnb ADR and occupancy comps · Per-listing P&L with visible assumptions · Weekly newsletter delivery (with a companion mobile surface). |
| Success metrics | Activation: profile completed and first feed received · Retention: weekly feed open rate and listings saved · Quality: investor-rated relevance of the top five. |
| Constraints | Listing-data access and terms · Compute per investor per week at consumer price points · Regulation data freshness. |
| Open questions | Whether the screener belongs inside Studio 3Sixty5 as a buyer-side module (it since has been extended there) · Which further strategies — long-term rental, house-hack — justify their own formula set. |
Strategy & bets
The whole product is a bet on the studio’s lifecycle running fast:
- Validate — test whether investors would act on a curated shortlist before building the agent that produces one.
- Prototype — once the shortlists were being acted on, an agent that reproduced the manual process: ingest, screen, rank, explain.
- Ship — as a newsletter, because the shortlist is read on a Sunday night and an email needs no install, no onboarding and no app-store review. A mobile surface came after, as a reader for the same feed.
- Measure — open rate, saves, and self-reported relevance of the top five.
- Scale or stop — published thresholds. If weekly open rate and saves fell below them for a sustained period, the product would be sunset or folded into a stronger one.
Deliberate non-goals kept it small: no underwriting, no offer flow, no property management. Each would have tripled scope and none was in the “stop searching” job.
System design: the evaluation workflow
The workflow does what a careful STR investor does for one listing — for every listing near a park entrance, every week.
- Locate. Start from the demand anchor — the park entrance — and define the search radius. Proximity is the single biggest driver of ADR in these markets.
- Find listings. Pull active listings in the radius: price, property type, beds, lot.
- Financing. Apply the investor’s financing assumptions and compute the debt-service coverage ratio (DSCR) — the number a lender will use, so the number the workflow uses too.
- Condition. Evaluate the listing photos for condition: what’s turnkey, what’s dated, what’s a project. This is a vision-model step, and it’s the one that separates a listing that’s cheap from one that’s cheap for a reason.
- Setup cost. From condition and property size, apply the setup formulas: furnishing, renovation, permitting — what it costs to turn the listing into an operating unit.
- Revenue. Pull nearby Airbnb comparables for existing ADR and occupancy, adjusted for proximity and size.
- P&L. Put it together — how much one can make — and rank against the investor’s thresholds, with the assumptions shown per listing.
The same workflow, a second strategy: fix and flip. The seven steps are a general underwriting loop, and swapping the revenue model turns it into a flip screener: locate → find listings → financing (short-term, hard-money terms instead of DSCR) → condition from photos → renovation scope and cost → after-repair value from nearby sold comps instead of ADR → profit and timeline instead of monthly P&L. The step that carries the flip case is condition: the vision model reads the listing photos room by room — kitchen, baths, flooring, roof and exterior where visible — and the renovation cost is estimated from what it sees, scoped by room and by level of work (cosmetic, update, gut), rather than from a flat per-square-foot guess. For a flip that estimate is the deal, so it ships with the photo evidence behind each line item. Same ingestion, same explainability; the formulas in steps 3, 6 and 7 change and step 5 gets deeper. Both strategies depend on the same computer-vision step: condition from photos is what makes setup cost real for an STR and renovation cost real for a flip. Build that once, well, and the underwriting formulas around it are cheap to swap. That’s the test of whether a workflow is a product or a one-off — the second strategy cost a fraction of the first.
Two choices mattered. Explainability over accuracy. Investors didn’t trust a ranked list without the underwriting behind it, so each listing ships with its DSCR, setup cost and ADR assumptions — a hedge against being wrong that also makes being wrong survivable. Weekly batch, not real-time. Real-time alerts were the obvious feature and would have multiplied compute cost; the interviews said Sunday night, so the pipeline runs on a schedule and the economics work at a consumer price.
Metrics & experimentation
- The validation metric was whether investors acted on a shortlisted listing, not whether they liked the idea.
- In production: newsletter open rate, click-through to listings, and a one-tap relevance rating on the top five. Those three numbers are the stop criteria.
- The explainability layer was tested as a variant; feeds with rationale outperformed feeds with rank alone on saves.
Outcomes
- Validated demand before engineering; shipped as a weekly newsletter, with a companion mobile reader on iOS and Android.
- The STR/MTR screening agent proved reusable: it has since been extended into Studio 3Sixty5 as a buyer-side deal-sourcing module, turning a standalone consumer product into a capability inside the larger platform.
- Perch runs against published stop criteria. That discipline — knowing in advance what “not working” looks like — is the reason the studio can carry five products without any one of them quietly consuming the roadmap.
Retro — what I’d do differently
- Treat it as a module earlier. The screening agent’s best home turned out to be inside the agent-facing platform; a standalone app was the right way to validate it, but I’d have planned the fold-in from the start.
- Charge during validation. Willingness to act is a good signal; willingness to pay is a better one.