--- title: "Analog Modeling: Look-Alike Site Selection" description: "Let your own results set the criteria: profile the locations where your stores perform, then screen a market for look-alike locations you haven't entered." slug: "/docs/guides/analog-modeling" last_updated: "2026-08-31" status: "published" industry: ["retail", "real-estate"] location_types: ["place", "area"] tags: ["analog", "site-selection", "look-alike", "expansion", "screening"] is_case_study: false locked: false related_metrics: ["foot-traffic", "popular-times", "visitor-demographics", "visitor-profile", "trade-area"] related_datasets: [] example_count: 1 example_tools: ["python"] --- # Analog Modeling: Look-Alike Site Selection Let your own results set the criteria: profile the locations where your stores perform, then screen a market for look-alike locations you haven't entered. ## Metadata - Industry: retail, real-estate - Location types: place, area - Tags: analog, site-selection, look-alike, expansion, screening The most convincing evidence about where to open next is the set of locations already working. Analog modeling, or look-alike site selection, turns that evidence into a method: profile the locations where your stores perform, then screen a market for locations that match the profile and lack your presence. Ranking sites on generally desirable attributes finds good locations in general. Analogs find good locations for you. That is why the method travels well into real-estate committees: its conclusions rest on results the committee already trusts. --- ## 1. The two halves of an analog An analog joins two things that live in different systems: - **Performance data**: sales, margin, memberships, whatever defines a win in your business. Usually this is your own, and it is the half that makes the method yours: it decides which locations are the ones to imitate. Competitor stores serve the same role when their format is comparable and you know which of their locations perform, knowledge real-estate teams at larger retailers routinely hold, with the caveat that the definition of a win is then theirs. - **Measured location attributes**: what the winning locations look like in the world. How much activity surrounds them, who it is, what else operates there, how far people travel to them. This is the half our location metrics supply, and it makes the profile transferable: the same attributes can be measured at places you have never operated. Neither half works alone. Performance without attributes says which stores win but not what their situations share. Attributes without performance is a generic desirability ranking, a legitimate but different method ([area identification](/docs/guides/area-identification), where you choose the criteria instead of letting your results choose them). ## 2. The dimensions of a profile What "looks like" means is a choice of dimensions. The ones that recur: - **Activity scale and rhythm**: how much measured traffic the surroundings carry, and when. A location that lives on weekday lunches and one that lives on weekend afternoons are different situations at equal volume. - **Audience**: the demographics and lifestyle profiles of the people who visit the area, not only those who live nearby. Visitor-based profiles capture what a place attracts; resident statistics only what it neighbors. - **Context and co-tenancy**: what else operates there. The categories present, the anchors that pull traffic, the depth of neighboring tenants. For in-center and high-street formats this is often the strongest dimension. - **Catchment shape**: how far and from where people travel to locations like yours, from [trade areas](/docs/concepts/metrics/places_trade_area). Two sites with equal surroundings can serve very differently shaped catchments. Each dimension can be measured for every candidate in a market, which is what turns a profile into a screen. ## 3. From network to profile With a small network, and most expanding networks are small, keep the model simple. Classify: group the locations your stores could occupy into a handful of transparent classes (by scale, audience, context), place your stores in them, and let performance per class name the winner. A few coarse classes beat many fine ones: spread across twenty segments, a small network proves nothing; across four classes it starts to. Read class results directionally, treat a class with a single store as a data point rather than a benchmark, and give the most weight to dimensions that order consistently across classes. Each class should be explainable in a sentence: transparent rules over clustering. When a candidate is accepted or rejected, the reason should be nameable. ## 4. From profile to candidates The last step is a filter: every location in the market that matches the winning profile and has no presence from you. What remains is where judgment comes back in. The screen cannot know lease availability, rents, formats that do not fit your operation, or what your pipeline already holds. An analog shortlist is the input to that judgment, not its replacement. From here, each candidate deserves a [store profile](/docs/guides/store-profiling) of its neighbors and, where the question is what the surrounding area lacks, a [void analysis](/docs/guides/void-analysis). A worked run at shopping-center grain, classifying a state's malls, joining explicitly fictional internal sales, and shortlisting the matching centers, is in [Finding candidate malls for an in-mall retailer](/docs/guides/analog-modeling/us-candidate-malls-colorado). ## 5. Where analog modeling sits Three site-selection methods share the same screening machinery and differ in where the criteria come from: | Method | The question | Criteria come from | | --- | --- | --- | | [Area identification](/docs/guides/area-identification) | Which areas score best on what I care about? | Chosen explicitly by you | | **Analog modeling** | Which locations look like the ones where I win? | Derived from your own performance | | [Void analysis](/docs/guides/void-analysis) | What does this area lack that its demand could support? | A reference: analog areas, competitor networks, your network | They compose: an analog profile narrows a market to candidates, area identification ranks territory where you have no network to learn from yet, and void analysis tests what a specific candidate's area is missing. ## Examples ### Finding candidate malls for an in-mall retailer _(Python ยท USA)_ Classify a state's malls into a handful of transparent classes, join explicitly fictional internal sales, and shortlist the look-alike centers with no store yet, in Python on calendar-2025 data. Full worked example: [/docs/guides/analog-modeling/us-candidate-malls-colorado.md](/docs/guides/analog-modeling/us-candidate-malls-colorado.md) ## Related metrics - Foot Traffic (place) - Popular Times (place) - Visitor Demographics (place) - Visitor Profile (place) - Trade Area (place)