CRE Co-Visit Affinity: A Shortlist for Tenant Mix

A busy supermarket may share visitors with most retailers nearby. That makes raw overlap a poor reason to put a particular tenant next door. The useful question is whether a candidate attracts the center's observed visitors more often than its local popularity would suggest. Co-visit affinity helps you build that shortlist. It does not tell you what rent the tenant can afford.

Define the Leasing Question Before Counting Visits

Choose the job first: filling a vacant unit, testing a category gap, or reviewing an existing tenant mix. For a vacancy, compare candidate categories already operating in relevant nearby centers. Name those comparison sites before seeing their scores. A commuter-oriented center and a destination shopping center may serve different trips even when they sit in the same market.

Define co-visitation as a device observed at both the subject property and a candidate during the chosen window. This method does not require those visits to occur on the same trip or in a particular order. If your leasing question depends on an anchor sending visitors directly to an adjacent shop, window-level overlap cannot answer it. That needs a separately specified sequence analysis.

Keep the unit of analysis explicit. Here it is an observed device within a window, not an identified shopper, household, or purchase. Record the eligible geography, observation requirements, visit rules, and exclusions in the analysis brief. For wider investment context, place the brief beside your commercial real estate data requirements.

Separate Shared Reach From Affinity

Let U be a fixed eligible panel for the market and window. Let A be devices in U observed at the subject property, and B devices in U observed at the candidate retailer. Count a device once in each set. Retain eligible devices visiting neither site in U; defining the universe as visitors to A or B changes the baseline you are trying to measure.

These ratios adapt the confidence and lift definitions in the mlxtend association-rule documentation. The adaptation treats a device's observed sites during the window as its set of items. It is an analysis design, not a claim that mobility observations are shopping transactions. If A or B is empty, leave lift undefined rather than manufacture a score.

Directional overlap can differ when you reverse A and B. Lift is symmetric under this fixed-universe definition, so it cannot tell you which tenant sends traffic to the other. Michael Hahsler's reference on association measures defines lift relative to independence and warns about low observed frequencies and sensitivity to the comparison population. Read a high score with those limitations attached.

Before calculating the ratios, reconcile the sets back to the delivery. Separate repeated pings, repeated visits, and repeated appearances of the same device across candidate sites. Keep an audit table showing which rule removed a record and which rule changed its site assignment. Do not add candidate rows together to estimate the center's audience: the same device can appear in several rows. Preserve the denominator definition with the export so a leasing colleague cannot compare a device count from one run with an overlap share from another. Missing observations should remain a documented limitation of the panel, not evidence of absent demand.

Use affinity to narrow a leasing investigation. The final decision also needs property and tenant evidence outside the visit panel.

Stress-Test the Shortlist Before Interpreting It

Start with geometry. Inspect adjoining units, shared parking, and multilevel properties. If the visit method cannot distinguish the candidate's unit from the center around it, label the result at center level. Do not present it as tenant-specific affinity. Our foot traffic accuracy guide explains why site assignment belongs in the evaluation.

Ask for the underlying shared-device support alongside the score, subject to permitted reporting rules. A striking ratio based on sparse observations deserves an insufficient-evidence label. Set support and stability requirements before ranking candidates; there is no universal cutoff in this method. Recompute the shortlist in a separate window and review changes in panel eligibility, store openings, closures, and seasonal trading before interpreting movement.

Then vary the comparison geography as a sensitivity check, keeping the other definitions fixed where feasible. If the preferred category changes, explain which population each result describes. Choose the geography that fits the leasing question, not the one that produces a more persuasive chart. Record exclusions as well, and apply the project's approved location-data restrictions before analysis; use the site's sensitive-location buyer guidance for that review.

Turn the Result Into a Leasing Packet

Give leasing a candidate table with the category, comparison sites, shared-device support, directional overlap, lift, window, and stability notes. Add a separate property-fit column for unit size, access, parking, opening hours, and lease restrictions. A category can share an audience with your center and still be physically or commercially unsuitable for the vacancy.

For buyers assembling the visit input, GSDSI's Location and Mobility Data Feed is a relevant dataset to evaluate against the brief. Ask which observations, site mappings, and permitted aggregate outputs are available for the actual trade area. Treat the affinity worksheet as your proposed analysis, not as a promised field in a delivery.

The practical shortcut is a smaller set of leasing questions: which category warrants investigation, which comparison site deserves a visit, and what evidence is still missing. Keep sales potential and incremental traffic as separate hypotheses. To scope the input data for that work, request a coverage sample for your candidate properties.