Visit Attribution: Five Signals Behind a Defensible Store Visit

A store-visit total hides the decisions that produced it. Start the evaluation one step earlier: ask why a particular observation became a visit to a particular place. Here, visit attribution means assigning location observations to venues. Connecting those visits to advertising exposure requires a separate measurement design. This review method is for teams evaluating mobility observations alongside POI boundaries; it does not describe a proprietary GSDSI algorithm.

Define the event before you count it

Write down the unit you intend to measure: a device-associated visit episode, a device observed at a venue during a reporting period, or another explicitly defined event. State how separate episodes are distinguished. Keep the output label tied to that unit; calling a device-based result a customer count requires additional evidence about the relationship between devices and customers.

Specify the venue level too. A shopping center, its parking area, and an individual tenant answer different questions. Freeze the place identifiers, boundary version, observation window, and rule version for the evaluation. POI polygons and geofencing data provide the place definitions for this workflow. A location and mobility feed provides the observation layer to evaluate alongside them. Ask for the fields and delivery scope needed to inspect both.

Check the five signals separately

Position quality: preserve the uncertainty

Request the reported position-quality fields, their definitions, and their missing-value treatment. Do not interpret an absent accuracy field as evidence of a good observation. GPS.gov explains that blockage and reflected signals can degrade GPS positioning. That is a reason to inspect the observation source and local conditions, not to assign a universal error allowance to a commercial feed. Location observations may originate through different collection methods; ask which method and quality indicators the supplier actually delivers.

Place geometry: inspect the assignment rule

Document how the spatial test treats boundaries and overlapping venues. For example, PostGIS ST_Covers includes a geometry's interior and boundary when testing coverage, and its documentation warns against invalid geometries. If you use that predicate, validate the inputs and record the choice. A geometric match classifies the supplied coordinates; it does not establish that the device entered a particular tenant. Keep competing place candidates when the available information cannot distinguish them.

Time evidence: distinguish observation from delivery

Request observation timestamps and delivery timestamps as separate fields. If the feed contains an event produced by another service, ask what that event's timestamp represents. Android's geofencing documentation describes configurable dwell transitions and warns that alerts can arrive late. Those are properties of that API, not a timing specification for your supplier's feed. For your evaluation, record the observation gaps and the evidence needed to support a visit episode. Widely separated observations leave the intervening activity unresolved.

Local context: make exceptions inspectable

Review the place alongside its operating context: opening hours, shared entrances, nearby roads, parking access, and changes in tenant occupancy. Treat those as checks to investigate, not proof of a person's purpose. An observation outside listed hours might reflect stale hours or a different activity. Mark the reason for uncertainty and the evidence needed to resolve it. Avoid turning a convenient business-hours filter into an undocumented definition of a customer visit.

Repeated observations: decide what counts again

Reconcile observation rows with the visit unit you defined. Ask how duplicate deliveries, repeated events, identifier changes, and a return to the same venue are handled. Keep the grouping rule and relevant timestamps with the result. Frequent recurrence can prompt a quality investigation, but recurrence alone does not establish that someone is an employee or resident. Do not attach those labels without suitable evidence and an authorized purpose.

Give unresolved candidates their own outcome

Require a reason code beside the visit decision. Accepted means the candidate meets the documented definition. Excluded means a stated rule rejects it. Unresolved means the evidence cannot support the required assignment. Keep those outcomes distinct in the evaluation report. Otherwise, changing how missing fields are handled can look like a change in demand even when the input observations are unchanged.

Protect that review trail under the agreed access and retention controls. Approve the purpose, data access, and place exclusions before matching begins. The sensitive-location buyer guide explains the diligence questions for location use. Technical filtering and aggregation do not establish permission to use a dataset; resolve that separately with the responsible reviewers.

Replay the same inputs before comparing totals

Create an evaluation set that includes the environments your analysis must handle, such as freestanding stores, shared buildings, and shopping centers. Record why each environment is included. A favorable result from easy locations does not answer whether your difficult locations are usable. The foot-traffic accuracy guide provides broader context for panel and observation limitations.

Compare the proposed definition with a plausible alternative on the same permitted input snapshot. Change a stated assumption, such as overlap treatment or the allowed observation gap, and show which candidates change status. Report the affected environments as well as the overall total. If the business conclusion changes with a reasonable rule choice, present that dependence to the decision maker before using the result to rank locations.

When suitable independent evidence is available, define what it measures before using it for validation. A door counter, transaction log, and device-observation feed describe different units. Align the period and place scope, document what can be compared, and report remaining differences. Agreement in an aggregate total alone does not validate individual venue assignments. Record an assessment as unresolved when the comparison cannot answer the question.

Make the decision trail part of the delivery

Put the evaluation requirements into the proposed delivery: field definitions, place versions, observation and processing timestamps, counting rules, exclusion reasons, and a change log. Ask how a corrected boundary or delayed observation affects a previously delivered result. Set an owner for reviewing those changes. Teams using visits in cross-channel measurement should carry the visit definition into the study documentation before joining exposure data.

A useful sample lets your team explain why a candidate was accepted and why another remained unresolved. Bring your venue formats, intended visit unit, and evidence requirements to GSDSI's team to scope a mobility and POI evaluation. That gives the licensing discussion a concrete delivery requirement and an evaluation your analysts can reproduce.