Credit card panel data vs. app foot traffic panels for retail comps

Two analysts can run the same ticker through two different alt-data feeds and walk away with opposite comp calls. That's a panel problem. It shows up most often when you're trying to reconcile a credit card spend read against an app-based foot traffic read in the weeks before a print.
Both are samples dressed up as populations. Worth knowing where each one comes from before you put a number next to it in a model.
What each panel measures
Credit card spend panel data comes from an aggregator that licenses transaction records from one or more issuers, then extrapolates that sample up to a category or ticker-level spend estimate. The panel is whoever happens to hold that issuer's card. The resulting read reflects whatever demographic and geographic mix that issuer's cardholder base happens to have, a mix that's rarely disclosed in a way you can audit. Add in cash, gift cards, BNPL, and store credit transactions the panel never sees, and you've got a spend proxy with a structural tilt baked in before any seasonal adjustment touches it.
App-based foot traffic panels work the same way on the mobility side. A vendor builds its count from a panel of phones that opted into location sharing through one or more apps, then scales that panel's visits up to an estimated store-level count. The opt-in population skews toward whoever uses those specific apps, younger and more urban than the chain's customer base, especially for a retailer with a rural or older footprint. Panel size also varies store to store. A flagship location in a dense metro carries hundreds of panelists a month. A strip-mall location in a thin-panel market carries a handful, and a handful doesn't hold up once you're trying to isolate a same-store comp.
Where the coverage gaps show up
Both panels get reweighted when their underlying sample changes. A card panel adds or drops an issuer relationship and the historical series gets restated, often with no date attached to the change. An app panel loses distribution when an OS changes a permission default, or a popular app in the panel gets deprecated, and the underlying sample shifts under you without a line in the methodology notes. You find out when last quarter's trend line quietly changes shape in this quarter's pull. Track how much a panel's own revisions have moved in past quarters, and put that number in the model note alongside the current read.
Transaction panel coverage gaps compound this. A single-issuer card panel misses any cross-shopping on a card outside that issuer's book, so a chain that skews toward a customer base holding a different card mix than the panel's issuer will read softer or stronger than its real comp, independent of demand. Match that against an app panel with its own geographic skew, and the two feeds can tell you directly contradictory stories about the same quarter. Each is built on a different slice of the population, and the slices tend to drift further apart the longer the quarter runs.
A count instead of a sample
This is the gap a daily vehicle count run across every parking lot in the chain, store by store is built to close. It counts every vehicle that shows up in the lot on a given day, from satellite imagery, indexed and seasonally adjusted into a time series you can line up against both panel reads to see which one is tracking the real traffic.
If the spend panel and the app panel are giving you different answers on the same name this quarter, a store-level traffic count across the full base is worth putting next to both before you write the number down.