One of the persistent assumptions in multi-store retail planning is that stores in the same city, or even in the same neighborhood, should have similar demand patterns for the same products. It's an intuitive assumption -- same brand, same market, similar customer demographics. In practice, the data tells a different story.
When we analyzed sell-through data from multi-location pilot stores across several product categories, the variation in weekly velocity for identical SKUs across stores in the same metro area was consistently larger than buyers expected. In some cases, the same SKU sold at more than 3x the rate at one location compared to another location less than 2 miles away. Aggregate forecasting doesn't see any of this variation -- it averages it out and produces an order quantity that's systematically wrong for every individual store.
What the Numbers Actually Show
Across the pilot data we collected from multi-location retailers in the Minneapolis area (2025), we found consistent patterns in demand variability. Within categories like specialty beverages, natural foods, and household cleaning products, coefficient of variation for weekly unit sales across stores in the same chain ranged from 0.35 to 0.85 -- meaning the standard deviation of weekly demand was often 35-85% of the mean. For any given SKU, this is a wide spread.
Put concretely: if the chain average for an organic oat milk SKU is 15 units per week, you might have one location selling 28 units per week consistently and another selling 8 units per week consistently. Both are real demand. Neither is an anomaly. But an order quantity calculated from the 15-unit chain average will leave the first location in a chronic stockout and the second location with chronic excess.
This variability is not random noise. It's structured -- driven by differences in neighborhood demographics, proximity to workplaces or schools, foot traffic patterns, local competition, and store layout and placement decisions. These are stable, predictable differences. They don't average out over time; they persist because the underlying causes persist.
Categories with the Highest Variability
Not all categories show the same degree of location-to-location variability. In our pilot data, the highest variability appeared in a few consistent patterns:
Specialty and premium products showed the most variability, because their customer base is more concentrated and more sensitive to neighborhood demographics. A high-end kombucha brand or an artisan granola SKU might have a very loyal following at one location where the customer base skews toward health-conscious buyers with higher discretionary spending, and minimal turnover at a location where the customer base is more price-sensitive or less familiar with the brand.
Convenience-driven categories also showed high variability tied to physical proximity to demand drivers. A store located near a gym or a yoga studio sells sports nutrition products at multiples of the chain average. The same product at a store in a different neighborhood type may barely move.
Stable staples -- basic pantry items, common cleaning products, widely consumed food items -- showed the lowest variability across locations. These are the products where chain-level forecasting works best, because demand is more uniformly distributed across the customer base regardless of location.
Why This Matters for Safety Stock Calculations
High demand variability across stores creates a specific safety stock problem. Traditional safety stock formulas are designed to buffer against demand uncertainty -- the more uncertain demand is, the higher the safety stock required to maintain a target service level. When you're calculating safety stock at the chain level, you're using the chain-level demand variability, which is the average of all the individual store variabilities.
But that average understates the variability at your high-velocity locations and overstates it at your low-velocity locations. The high-velocity location needs more safety stock than the chain calculation provides -- it's consistently running closer to zero before the reorder cycle closes. The low-velocity location ends up with more safety stock than it needs, tying up capital in inventory that turns slowly.
Store-level safety stock calculations, based on each location's own demand variability, allocate safety stock where it's actually needed rather than uniformly across all locations based on a chain average that accurately describes none of them.
The Store-Specific Knowledge Buyers Carry
Experienced multi-location buyers already know many of these patterns intuitively. They know which location runs fastest on which categories. They know the neighborhood-specific demand drivers. They may have built some of this knowledge into their par levels -- deliberately setting higher par levels for certain SKUs at certain locations based on operational experience.
The challenge is scale and freshness. Manual par level adjustments based on buyer intuition can capture the broad patterns but miss the fine-grained, current velocity data. A buyer managing 10 locations with 400 active SKUs can't maintain current, location-specific par levels for 4,000 inventory positions through manual review. They'll get the large-magnitude differences right (Location A clearly sells more oat milk than Location B) and miss the smaller but still meaningful variations that accumulate into material inventory errors over time.
How to Measure Location-Level Variability in Your Own Data
If you want to assess the demand variability across your own stores before making any process changes, a practical approach is to pull weekly unit sales by SKU by location for the last 8-12 weeks from your POS system, and calculate the coefficient of variation (standard deviation divided by mean) for each SKU across your locations.
SKUs with a coefficient of variation above 0.4 are candidates for location-specific par levels rather than chain-wide settings. Those are the products where the aggregate number is most misleading and where per-store demand sizing will have the largest impact on inventory accuracy.
The exercise usually produces a few surprises -- products where the degree of location-to-location variation is larger than the buyer expected based on intuition. Those surprises are where the biggest inventory improvements tend to be waiting.
