Foot traffic data has no single accuracy rate. The reliability of a visit estimate depends on the location signals, place boundaries, visit rules, sample coverage, and adjustments behind it. The useful test is whether the metric measures what your decision requires and holds up against a suitable independent reference.
Imagine two reports for the same shopping center and month. One shows 84,000 visits; the other shows 71,000. These are illustrative numbers, but the buyer’s question is practical: which figure should inform the next lease, staffing plan, or market comparison?
Start by checking whether the reports count the same thing. One might estimate visits to the whole property, including parking. Another might measure visits to selected tenants. They may also use different samples and rules for identifying a visit.
At Factori, we see this as part of the context a business needs alongside its data. To use a foot traffic number well, an analyst or AI agent needs to know how it was produced and what it can support.
First establish what the number counts
Foot traffic measurement can use door sensors, manual counts, or estimates derived from mobile location data, among other methods. This article focuses on aggregated mobility data: movement observations processed into summaries about places and time periods.
These measures answer different questions:
| Measure | What it represents | What it does not establish by itself |
| Location observations | Recorded signals associated with devices and times | A count of visits or people |
| Observed visits | Sessions that meet the provider’s visit rules within its sample | All visits to the property |
| Unique observed visitors | Distinct identifiers counted over a stated period | An exact count of distinct human beings |
| Estimated total visits | A modeled expansion or estimate of visitation | A direct door-counter measurement |
| Foot traffic index | Activity relative to a denominator or baseline | An absolute visit count |
A person can visit more than once. An identifier can change. One person may use multiple devices, while others are absent from the sample. The dataset’s definitions determine how these situations are handled.
When traffic rises, growth in observed visits could reflect more repeat trips, more observed visitors, or a change in the sample. It does not automatically mean more customers.
How mobility signals become foot traffic data
A simplified workflow is: filter location observations, associate activity with places, classify visits, resolve repeated observations, adjust for the sample, and aggregate the result. Providers may combine or reorder these stages. The following six decisions explain why the output can differ.
Which observations are usable
Location observations vary in positional accuracy, timing, and frequency. A point near a storefront may be compatible with several places. Sparse observations can miss a short stop or make its duration difficult to estimate.
Quality checks can identify duplicate records, implausible travel, inaccurate coordinates, or observations that lack enough context. Stricter filters may remove false visits, but they can also remove genuine activity unevenly across places or groups. More filtering is not automatically more accurate.
Which place receives the activity
A good location signal can still be assigned to the wrong business. A mall may contain dozens of tenants sharing a building footprint, entrances, and parking. Coordinates alone may not distinguish adjacent stores or different floors.
The point-of-interest data needs to represent the relevant place: a tenant, building, shopping center, or broader destination. Boundaries that include a road or parking area can produce a different measure from boundaries limited to the premises.
What qualifies as a visit
A stop near a place is not necessarily a shopping visit. It could be a delivery, a roadside pause, or time spent at work. Visit classification may use dwell time, repeated observations, movement patterns, and recurring activity to distinguish these situations. Those rules infer behavior; they do not directly reveal intent.
A long minimum dwell time might miss legitimate convenience-store visits. A short threshold may admit nearby pass-through activity. The tradeoff depends on the category and what the metric is intended to measure.
Where one visit ends and another begins
Several signals can belong to one visit. A return after a short absence may count as the same session or a new one. The rule affects both the total and the apparent frequency of repeat visits.
Unique visitor counts also depend on the reporting window. Adding daily unique visitors produces a sum of daily counts, not a monthly count of distinct visitors: the same identifier can appear on several days.
How the sample is adjusted
Observed devices are a sample of activity. Coverage can differ by market and change between periods. Normalization can help make a comparison more meaningful, but its denominator determines what the result means.
Consider this simplified example for one location. All values are hypothetical; the panel measure is the number of active devices in the same defined market and period.
| Measure | Month 1 | Month 2 |
| Observed visits to the location | 1,000 | 1,200 |
| Active devices in the market panel | 10,000 | 15,000 |
| Observed visits per 1,000 active panel devices | 100 | 80 |
Observed visits increased by 20%, while visits relative to the panel denominator fell by 20%. Neither calculation establishes the change in total real-world visitation. They answer different questions, and the second still depends on panel composition and observation quality.
This distinction appears in actual dataset documentation. Advan’s Weekly Patterns+ documentation advises interpreting its trade-area values as ratios or scaled indicators, rather than absolute numbers, because its trade-area and visitation panels differ. That guidance applies to those fields; it is not a rule for every provider.
What aggregation and suppression remove
Monthly totals can help compare month-to-month patterns but conceal a Saturday-evening surge. A shopping-center total may be useful for destination analysis while offering too little detail for an individual tenant.
Aggregation can smooth some random variation. It does not repair a biased sample, a misclassified visit, or a changed boundary. Small counts may also be withheld under data-quality or privacy rules. A missing or suppressed value must not silently become zero.
How to validate foot traffic data before using it
Use reference data that matches the quantity you want to evaluate. Door counters can help assess entrance counts, once their own rules and operating periods are checked. A store’s transaction records measure purchases, not all visits. Another provider’s modeled estimate is a useful comparison, but agreement between two models is not independent proof of accuracy.
A practical evaluation has four parts:
- Match the scope. Use the same premises, entrances, dates, opening hours, and definitions. Resolve differences such as re-entry, staff inclusion, and missing days before scoring error.
- Test representative locations. Include the formats and settings that matter to your business, such as standalone stores, shared buildings, and dense retail corridors. Report results by type so an average does not hide weak locations.
- Separate totals from direction of change. Check both the error in absolute counts and whether the series captures peaks, troughs, and changes. A dataset can capture the direction of change while systematically overestimating its level.
- Keep evaluation data separate. Assess performance on locations or periods not used to tune the method. Review the largest errors, coverage gaps, and any supplied uncertainty ranges, as well as the average result.
Research supports the need for this scrutiny. A Communications Physics study compared seven mobility sources across 145 countries and found substantial differences in their resulting mobility networks. It was not a retail footfall accuracy benchmark, but it shows why mobility sources should not be treated as interchangeable.
What to give an AI agent alongside a traffic number
An agent receiving “1,200 visits” has less useful context than one receiving the metric definition, location scope, reporting period, panel denominator, and quality flags.
Ask it to identify whether the field is observed, estimated, or indexed before calculating changes. Preserve missing and suppressed values. Require it to flag breaks in methodology and distinguish a traffic change from a conclusion about demand.
Better real-world context gives the system evidence to interpret the number, instead of relying on the column name alone.
Evaluating mobility data with Factori
Factori Mobility Data provides movement and visitation context for location analysis, forecasting, and planning. Teams can use it alongside place information, population context, and their own business data. Compatible AI assistants can access the Factori workspace through its MCP server.
For an evaluation, bring a small set of representative locations, the reporting period, and the decision the data must support. Review the available fields and coverage, then agree on a suitable reference and success criteria. The evidence required to compare two markets may differ from the evidence needed to estimate a store’s entrance count.
About Factori
Factori is a leading global location intelligence company that provides unmatched data insights to help businesses better understand the physical world:
Factori datasets are governed, privacy-safe, and structured to join seamlessly with your existing workflows across SQL, data warehouses, BI tools, and ML pipelines. With over 90B+ location signals collected every day across 150+ countries, Factori delivers broad market coverage and reliable location intelligence at scale. Datasets are available via APIs, raw data, the Factori platform, and MCPs to support different use cases and markets.
FAQ
1. Is foot traffic data accurate?
There is no single accuracy rate. Reliability depends on the signals, place boundaries, visit rules, sample coverage, and adjustments behind the number, and on validating it against a matching reference.
2. What is the difference between observed visits and unique visitors?
Observed visits are qualifying sessions within a sample; unique visitors are distinct identifiers over a stated window. Daily unique counts cannot be added into a monthly unique total.
3. Why do two providers report different foot traffic for the same place?
Different place boundaries, visit rules, samples, and normalization denominators produce different numbers for the same location and month.
4. Does more filtering make foot traffic data more accurate?
Not by itself. Stricter filters remove false visits but can also drop genuine activity unevenly across places or groups.
5. How do you validate foot traffic data?
Match the scope, test representative locations, separate totals from the direction of change, and keep evaluation data separate from any tuning.
