Census data is the default starting point for most retail site selection demographic work. It is free, nationally consistent, and covers virtually every geography in the country at multiple levels of spatial resolution. For many of the questions a site selection team asks, it is genuinely useful. For others, it produces a misleading picture that looks authoritative because of how official it feels.
Understanding where census data is reliable and where it is not is not a criticism of the data itself. The Census Bureau collects what it is designed to collect, and it does that reasonably well. The limitations come from what census data was designed to measure, how often it is updated, and what retail consumer behavior it cannot capture at all.
Census Measures Who Lives There, Not Who Shops There
The most fundamental limitation is one that is easy to forget when staring at population and income tables. Census data describes the population that resides in a geography. It does not describe who shops in that geography, which is often a significantly different group.
Dense urban commercial corridors draw shoppers from beyond the surrounding residential population. A retail district adjacent to a major employment center will see substantial daytime traffic from workers who live elsewhere. A neighborhood near a major transit hub may have residential demographics that understate its retail potential because many potential customers pass through without living there. In tourist areas, visitor populations that are invisible in census counts may drive a substantial share of retail category spending.
The reverse is also true. Suburban residential zip codes with strong income demographics may have residents who satisfy most of their retail needs at major regional centers outside the local area. The income is there, but the shopping behavior routes to a different geography. A census-based site analysis that equates high household income with high local category spending will overestimate demand in those markets.
The Data Staleness Problem
The decennial census captures a full-count population picture once every 10 years. The American Community Survey, which provides the demographic and income data most site selection teams use, publishes five-year estimates that average data over a five-year period. For a neighborhood that has changed substantially in recent years, the ACS estimate may reflect conditions from three to seven years ago.
This matters most in rapidly changing areas: neighborhoods that have gentrified, areas that have experienced significant development or population loss, and submarkets where major employers have opened or closed. The income and population data for these areas may be meaningfully different from what current conditions look like on the ground.
A site that census data says has a median household income in the lower-middle range may now have a significantly higher income profile if the neighborhood has experienced significant development in the past four years. Conversely, a site that census data shows as a stable middle-income area may have experienced recent economic stress that the data has not caught up to. Neither situation will be visible in census tables.
Category Spending Is Not in the Census
Census data does not include consumer spending by category. You can see household income, age distribution, household composition, and some housing cost data. You cannot see what share of household spending goes to your specific retail category, how that compares to peer geographies, or whether the area's residents tend to spend on your category locally or elsewhere.
This gap is consequential for any category where spending rates vary significantly across demographic groups or where local spending is a lower share of total category spending than national averages would suggest. A high-income area does not uniformly translate into high spending in every retail category. Category affinity, which measures how much of relevant household spending actually flows to a specific retail type, is the number you need, and census does not provide it.
Consumer spending data derived from aggregated credit and debit card transactions by NAICS category at a geographic level fills this gap more reliably than census income data. It shows what people in a given area actually spend in your category, not what their income suggests they might spend. For categories with high variance in spending behavior across similar income levels, the difference between these two inputs is significant.
Where Census Data Remains Genuinely Useful
This is not an argument to ignore census data. For questions about baseline population density, broad age distribution, household composition, and housing tenure, census and ACS data is accurate, granular, and well-suited to retail site work. Knowing that a trade area has a high proportion of households with children under 12, or that an area has a strong concentration of renters versus homeowners, or that the population skews heavily toward adults 55 and older is useful context that census data provides reliably.
Where census data becomes a liability is when it is used as a proxy for things it does not directly measure: category spending, shopping behavior, consumer preferences, or current market conditions in areas that have changed recently. Using it as a proxy for those things without cross-referencing against behavioral data sources introduces systematic errors into the evaluation.
Combining Census With Behavioral Data
The practical approach is to treat census and ACS data as the foundation for population and demographic structure, then layer behavioral data sources on top for the questions census cannot answer. Foot traffic data shows where people actually go, not just where they live. Consumer spending panel data shows what categories they spend in, not just what their income implies they could spend. Point-of-interest data shows what retail supply already exists in the trade area.
A site evaluation that uses census data for population structure and behavioral data for spending patterns and mobility will produce more accurate demand estimates than one that uses census data for everything. That is a higher bar than many teams currently apply, but the cost of better data inputs is typically modest relative to the cost of a lease commitment made on an incorrect demand estimate.
Census data is part of the answer. It is not the whole picture, and treating it as such is one of the more common sources of avoidable error in retail location analysis.