Auto-Detecting the Comp Set, How a Boutique’s Competitor List Should Build Itself.

Every legacy market-intelligence product starts onboarding the same way. A consultant gets on a call with the GM, opens a configuration form, and asks the boutique’s five comparable properties. The GM names three confidently and two by guess. The consultant types them in. The dashboard begins watching those five hotels. Twelve months later, nobody has revisited that list.

This is the wrong shape for boutique customers. Three things break specifically.

First, boutique GMs frequently have the wrong comp-set instincts. They name the hotels they walk past, the hotels they socialize with at the chamber of commerce, the hotels whose owners they like. None of these signals are reliable predictors of who is actually capturing or losing the same guest a boutique would have booked. The hotels that share a booking funnel often look nothing alike from the curb.

Second, the comp set drifts. A boutique that opens two blocks away last quarter becomes a comp set overnight. A renovation that repositions a property from midscale to lifestyle moves it into or out of the set. Configuration-based comp sets do not drift automatically; someone has to remember to revisit them, and nobody does.

Third, and most importantly, onboarding friction kills conversion. The boutique segment is the opposite of an enterprise sale. You do not get six weeks of consultant time and a kickoff meeting. You get a GM who opened a free trial at 11 p.m. and either sees value within ten minutes or never comes back. Asking that operator to populate a comp-set form on day one is the single largest churn event in the funnel.

What auto-detection actually means

A modern platform should produce a meaningful comp set from the property’s name and address. Nothing else from the operator. Here is the data layering that makes that work, and the honest tradeoffs at each layer.

Layer one: public map and business-listing data. Public mapping and business-listing sources expose structured data for nearby lodging within a configurable radius (3 km in a city, 25 km in a rural market). Each candidate gets pulled with its name, address, geo, and a rough star-rating signal. Breadth of coverage and depth of metadata vary by source, so a serious build blends more than one.

Layer two: market-segment matching. Not every nearby hotel is a comp. A budget-tier 80-room property two blocks from a 22-room boutique is not part of the same booking funnel. The platform needs to infer the property’s own segment first, lifestyle, boutique, luxury, midscale, and then filter the candidate set to the same segment. Segment inference happens automatically from amenities, price range, and rating density; the operator can override on the settings page if it gets called wrong.

Layer three: ADR-band proximity. Even within the same segment, two boutiques in different price bands are not real comps. The platform pulls public rate signals (OTA availability for the next 14 days) and filters candidates whose ADR sits within roughly ±25 percent of the property. A $180 boutique inn and a $420 design hotel one block apart are not in the same comp set even if a public map listing makes them look similar.

Layer four: presence on the OTAs we monitor. A candidate is only useful as a comp if the platform can actually watch its rates. So the final filter is presence on at least one of the major OTA channels the platform watches (Booking.com, Expedia, and the rest). A nearby property that only sells direct is removed from the auto-set, the platform cannot watch its rate, so including it is misleading.

The result

Out of the filter cascade, the platform typically arrives at a comp set of 5–9 nearby properties. That set should be visible on the map view of the product from the moment the property is created, pins clustered around the operator’s own dot, each tagged with current rates and availability the moment OTA data lands. The operator can verify with their eyes (“yes, those four are the hotels I actually compete with; this fifth one I would not have thought of, but it is genuinely similar to me”) instead of generating from scratch.

The operator can also override. A boutique that knows it actually competes with a property 7 km away because they share a wedding-market reputation can pin that property manually. A nearby property that looks comparable but is actually a corporate-extended-stay can be removed. The auto-set is the starting point, not the cage.

Why this is not a Lighthouse feature

Auto-detection is technically possible at any scale. It is not a Lighthouse feature because Lighthouse does not have the boutique-segment problem. Their customers are flagship properties with revenue managers who already know their comp set by memory, have argued about it in commercial meetings for years, and would resist any automated guess that does not match the one they want. The configuration form is not friction in that segment, it is the moment they assert authority over the dashboard.

The boutique segment is the opposite. There is no commercial meeting. There is no revenue manager. The GM wants the platform to do the work and to show them what it found. Auto-detection is not a clever feature; it is the table-stakes move that lets the product work for the audience at all.

What the platform must do honestly

Two failure modes are easy to fall into and worth flagging.

First, the auto-set should be visible and explainable, not hidden behind black-box magic. The settings page should list each detected comp with the reason it was included (“2.1 km away · same segment · $195 ADR vs your $210”) so the operator can sanity-check the inference. Black-box auto-detection that the operator cannot inspect produces the same trust problem as black- box pricing, they stop believing the rest of the intelligence built on top of it.

Second, the platform must not auto-detect when the candidate pool is too thin to be useful. A rural boutique with no comparable property within 25 km does not have a comp set, and pretending otherwise by widening the radius to 100 km produces signal that damages trust faster than no signal at all. In thin markets the right move is to show a different intelligence shape, pure demand events and search- demand signals, no comp comparison, and to say so clearly in the briefing.

Want it running on your own property? Start a free 7-day trial: no credit card, full access from day one.