Foot-Traffic Panel Sizing: How Many Devices Do You Actually Need for a Read?

Foot-traffic buyers often ask panel size last and should ask first. Panel depth sets confidence-interval width: too-small panels flip chain rankings quarter-over-quarter; right-sized panels survive audit. This guide maps MAID depth to chain, store, and DMA questions with rules of thumb to pressure-test vendors. Start with POI quality in depth and geofencing best practices: bad geometry shrinks effective depth before panel math matters.

Why Panel Size Is the First Question

Foot-traffic is sampling: observed visits extrapolate to totals using capture rate, the kind of methodology disclosure Media Rating Council accreditation exists to scrutinize. Smaller panel vs visitors → wider CIs on counts, dwell, O/D, YoY deltas. Evaluating "100M MAIDs" without translating to CI for your chain and DMA fails procurement. A 40M panel concentrated in your geographies may beat 100M nationally maldistributed.

Document analytical grain in the charter (chain quarterly, store weekly, DMA O/D monthly) each implies different minimum depth.

Chain-Level Reads: The Easy Ask

National chain QoQ traffic aggregates hundreds of stores: noise cancels. Working rules:

• ~60-80M US MAIDs/month is the planning range we use for chain reads on 100+ US store chains. It is a working assumption, not an industry standard: ask any vendor, including Global Mobility, for its measured monthly US device count.
• 20-40M panels can work, with wider confidence intervals at quarterly frequency (our estimate: roughly 5-10% wider).
• Weekly chain reads need ~2-3× quarterly depth for same CI.
• Sub-50-store chains: aggregate to a category benchmark, because chain-level precision rarely holds.

Store-Level Reads: Roughly 10× the Panel Depth

Store #347 week-over-week removes the aggregation denominator: depth demand jumps:

1. Top-quintile urban stores: ~10-15% CI width weekly on ~60-80M panel.
2. Median stores (25-40% CI) directional only.
3. Bottom quintile: weekly store reads not defensible. Use monthly or chain roll-up.
4. Urban 100K-visit stores beat rural 50K at same nominal traffic: signal density differs.

Radius geofences inflate visits, POI polygons fix effective depth. Never size panel before geometry QA on ten hard sites.

DMA-Level Reads for Origin/Destination Work

O/D sits between chain and store demand. ~60-80M supports top-50 DMAs; DMAs 51-100 need ~2× CI width or multi-month aggregation. Below top-100, flag exploratory. Cross-check Census LEHD LODES for worker-flow structure. International visitors need country-mix disclosure, not only US depth headlines.

The Three Panel Diagnostics to Run Before Signing

Pressure-test before license:

1. Observed/expected visits at a known high-traffic big-box in a top-10 DMA: ratio far from 1.0 flags bias.
2. DMA composition vs population: adjust expectations for under-covered markets.
3. SDK mix: single-SDK concentration is regulatory and continuity risk.

Align to MRC mobility standards at procurement time. For identity joins beyond visits, see identity graphs 101. Triangulate with foot-traffic vs card panels for conversion reads.

Count Observations, Not Devices

Panel size is quoted in devices because it is the largest available number, and it is a poor predictor of whether a read will hold. What determines statistical power is how many usable observations reach the specific place and period you are measuring, and that depends on how often each device reports, not merely how many exist. A large panel of devices that report a handful of times a day can deliver fewer usable observations at a given store than a smaller panel that reports frequently.

Ask for observations per device per day alongside the device count, and ask how that rate varies across the panel rather than as an average. Panels are usually highly skewed, with a minority of devices producing most of the signal, so a mean reporting rate flatters what a typical device contributes. The number that matters for your program is expected observations at a median location in your footprint over your reporting period, and any vendor able to size a panel can compute it.

The Weakest Locations Set the Limit

Panel adequacy is usually assessed as a portfolio average, and decisions are made per location. Those two things pull in opposite directions. Rural sites, early-morning and late-night dayparts, and stores in areas with lower smartphone panel participation all receive far less coverage than a chain-level figure suggests, and they are frequently the locations where a manager most wants a read.

Size the panel against the locations you need to be right about, not the average one. Compute expected observations per location per period, then look at the bottom of that distribution rather than the middle. It is entirely normal for a panel to support confident weekly reads for most of a portfolio and only monthly reads for the tail, and stating that split honestly is more useful than a single cadence applied everywhere. Publish the minimum reporting period per location alongside the numbers so nobody reads a sparse cell as a real decline.

Do the Power Calculation Before You Sign, Not After

The question a panel has to answer is not whether it produces a number, since any panel will, but whether it can detect a change small enough to matter to your business. Work backwards from that. Name the effect size that would change a decision, a lift a campaign has to clear or a decline that would trigger action, and ask what sample is required to detect it at your chosen period and confidence.

Run that calculation against the vendor’s own observation rates during evaluation, using your locations rather than a generic example. If the required sample exceeds what the panel delivers at a typical site, the honest conclusion is that this panel cannot answer that question at that granularity, which is a far better outcome than discovering it a year later inside an argument about a result. Often the fix is not a bigger panel but a coarser question, weekly rather than daily, or a group of stores rather than one.

Panel Continuity, the Axis More Depth Cannot Fix

Depth decides how noisy a single period is. Continuity decides whether two periods are comparable at all, and it is a separate failure. When a panel gains or loses an SDK partner between quarters, the sample changed and the world did not, but the level shift lands in your trend looking exactly like demand. No sizing rule in this piece catches it, because the panel was comfortably large in both periods. That is what makes the failure durable. Every diagnostic that looks at one period in isolation comes back clean.

The detection test is a control basket. Hold a set of locations whose true traffic has no business moving together, spread across categories, markets, and store formats, and watch for common movement. When the whole basket steps the same direction in the same period, suspect the panel before you write the demand narrative. The confirming cross-check is a matched-panel read, restricting both periods to devices observed in both. If the matched series is flat while the full series jumps, you are looking at composition rather than visits, and the extrapolation factor is what moved.

Then put continuity in the contract, because it is not something you can fix downstream. Ask for advance notice of partner additions and removals, versioned deliveries so a number you already published can be reproduced later, and a written restatement policy covering what happens to history when methodology changes. Silent backfill rewrites make last quarter's board slide unreproducible, and you find out when someone asks why the figure moved. Regulatory exits produce steps rather than drift, so expect the notice to matter most in the periods that are hardest to explain. See the geo panel audit. Before sizing a panel, you can check per-chain foot-traffic coverage for major brands with the free foot-traffic tools.