Planogram optimization is the decision of how much shelf space each product gets, and where, so that a category earns the most from a fixture that cannot grow. Unlike a share-of-sales rule, an allocation model puts facings and shelf positions where demand responds, within what the fixture and the delivery cycle can hold. That response is real but small. A meta-analysis of 1,268 estimates (Eisend, 2014) puts average shelf space elasticity at 0.17, and field experiments by Drèze, Hoch and Purk (1994) found that location had a large impact on sales, while extra facings mattered much less above a minimum threshold. Most lifts published since were estimated, not tested against control stores, which is why planogram optimization is worth exactly as much as the way you measure it.
What Planogram Optimization Decides: Facings, Position, Capacity
Planogram optimization is the allocation of a fixed amount of shelf space among the products of a category, by number of facings and by shelf position, so that expected sales or profit is highest under the physical limits of the fixture and the merchandising rules the retailer sets.
That means three decisions per item:
- how many facings it gets, a facing being one visible unit at the shelf front;
- which shelf it sits on, and where inside its block;
- how many units the slot holds between deliveries.
Corstjens and Doyle opened their 1981 space allocation model: "The allocation of scarce shelf space among competing products is a central problem in retailing." A 2021 review by Bianchi-Aguiar and colleagues dates the first empirical shelf studies to the 1960s and the first models to the 1970s.
Which products belong in the range at all is a separate decision with its own model, covered in deciding which products earn a place in the range; here the assortment is an input, and the question is how much space each listed item gets and where. Whether stores then build the shelf as planned is an execution question, answered by checking the shelf against the plan with image recognition; this article stops at the plan.
Space Elasticity: How Much Demand Responds to Shelf Space
Space elasticity tells you how much more a product sells when it gets more room on the shelf. Eisend's meta-analysis in the Journal of Retailing defines it as "the ratio of additional sales to additional space allocated in retail settings." Across 1,268 published estimates the average was 0.17, lowest for commodities, higher for staples and highest for impulse buys. Price is held fixed here, which is what separates space elasticity from price elasticity.
At that average, ten percent more space goes with a little under two percent more sales: an illustration of a cross-study average, not a forecast for any SKU. Eisend also found that adding space produced higher estimates than taking it away. Our reading: removing a facing from a slow seller usually costs less than its neighbor gains, provided the slow seller stays above its availability threshold.
The response flattens as space grows; the Lowe's team behind the OPTIMUS preprint uses logarithmic curves "to model diminishing returns with respect to space expansion." Neighbors matter too. Corstjens and Doyle argued that "both main and cross space elasticities have to be considered," and Schaal and Hübner (2018) note that the cross effect "is complex and costly to measure." In planogram optimization we start from each item's own elasticity and add cross effects only where substitution is obvious.
Shelf Position Often Matters More Than Facing Count
The most useful single finding for planogram optimization concerns where an item sits. Drèze, Hoch and Purk ran a series of field experiments on shelf management and space elasticity, and their abstract is direct: "We found that location had a large impact on sales, whereas changes in the number of facings allocated to a brand had much less impact as long as a minimum threshold (to avoid out-of-stocks) was maintained." They reported "modest gains (4%) in sales and profits from increased customization of shelf sets" and changes of 5 to 6 percent from shelf reorganization.
Van Nierop, Fok and Franses (2008) show "important dependencies between the layout of the shelf and stock-keeping unit (SKU) sales and marketing effectiveness," and add a warning for anyone fitting a model: "We also take into account potential endogeneity of facings." Best sellers get many facings because they sell, so a naive regression of sales on facings overstates the elasticity.
The order we use follows: first the facing threshold that protects availability, then position, then any extra facings.
From Share-of-Sales Rules to Allocation Models
Many shelves are still planned by rule: space in proportion to share of category sales, or room for a fixed number of days of stock. Drèze and his coauthors put it bluntly: "Shelf management is a difficult task in which rules of thumb rather than good theory and hard evidence tend to guide practice." Planogram optimization replaces the rule with a model of how demand and cost respond to space.
Early models covered both sides. Corstjens and Doyle combined space elasticities with "procurement, carrying and out-of-stock costs." Bultez and Naert's SH.A.R.P. model (1988) started from "Shelf space is the retailer's scarcest resource," and was validated on data from a Dutch supermarket chain.
Software lagged. A 2012 review by Hübner and Kuhn found that "Scientific decision models either focus on space-dependent demand or substitution effects, whereas software applications use simplistic rules of thumb." That was the market in 2012, not a verdict on tools sold today.
Planning in sequence also costs profit. Hübner and Schaal (2017) note that assortment and space "are mutually dependent if space is scarce," and in numerical experiments their integrated model beat assortment-first planning by up to 18% in profit.
What AI Adds to Planogram Optimization, and What Was Actually Measured
AI planogram optimization has two parts, and neither arranges a shelf by itself. A learned demand model estimates the response to space and position per category and store cluster; an optimizer, typically integer programming, then divides the space under the fixture's constraints.
Across ten runs spanning departments and store clusters, the Lowe's OPTIMUS framework reports an average sales lift of 11.8% and an average margin lift of 9.5%. The method matters as much: "For each run, we compared pre- and post-optimization outcomes using historical data, capturing the combined effect of assortment selection and space allocation without isolating individual optimization stages." The authors add that how stores are clustered changes the elasticity curves, and with them the results.
An MIT thesis built on Target data (Habel, 2024) seeks "the optimal trade-off between planogram-store personalization and standardization." For one category, spices, over a six-month horizon, it reports an estimated 46% cut in operating cost and an estimated 3% revenue improvement. Generative models are earlier still: a 2026 arXiv paper builds planograms with a diffusion model, with results from simulation only.
Published evidence on space allocation: what was measured and how.
| Study | Setting | Reported result | How it was measured |
|---|---|---|---|
| Drèze, Hoch and Purk (1994) | Supermarket field experiments | 4% gains from store-specific shelf sets; 5 to 6 percent changes from reorganization | Field experiments in stores |
| Eisend (2014) | Meta-analysis across categories | Average space elasticity of 0.17 | 1,268 published estimates |
| Hübner and Schaal (2017) | Integrated assortment and shelf-space model | Up to 18% more profit than sequential planning | Numerical experiments |
| Hübner, Düsterhöft and Ostermeier (2021) | Shelf dimensioning plus allocation | Up to 5% more profit than literature benchmarks | Numerical tests and a retailer case study |
| Habel, MIT thesis with Target data (2024) | Planogram portfolio, one category | Estimated 46% lower operating cost, estimated 3% more revenue | Model estimate, not a store test |
| Lowe's OPTIMUS preprint (2026) | Department-level space, ten runs | 11.8% average sales lift, 9.5% average margin lift | Before and after comparison on historical data |
Only the first row comes from an experiment in stores, and it reports the smallest effect. Treat every other number as a reason to run a test, not as a forecast.
The Constraints and Data an Allocation Model Needs
Planogram optimization works inside a physical box. Düsterhöft, Hübner and Schaal (2020) are explicit: "Without considering the actual shelf dimensions, results from optimization models cannot be transferred to actual shelves." Hübner, Düsterhöft and Ostermeier (2021) sized the shelves inside the model and reported up to 5% higher profits than literature benchmarks.
One vendor's implementation guide for Oracle Retail AI Foundation shows how a commercial system handles this. A sales model computes "the expected sales for a particular item i with k number of facings and at a particular elevation in the planogram," and an inventory step "simulates a (s, S) inventory model to estimate service levels as a function of number of facings." When rules contradict each other, the optimizer "does not return any solution." The guide never says elasticity; the response to facings comes from the sales and inventory models.
Merchandising rules are constraints too: brand blocks, minimum and maximum facings, items kept together.
The data planogram optimization needs is mostly about the shelf:
- sales per SKU, store and week;
- current planograms with facings and position per item;
- product and fixture dimensions;
- case pack sizes and delivery frequency;
- out-of-shelf records, so an empty slot does not read as missing demand;
- planogram change history, the only variation a response can be estimated from.
Once a model runs, track space productivity metrics per SKU, category and store, so the next reset starts from measured response.
How to Pilot It: One Category, Control Stores
A planogram optimization pilot should isolate the effect of space from everything else that moves sales. The steps we use:
- one category with a stable assortment;
- a cluster of stores with similar formats;
- control stores that keep their current planogram over the same period;
- measures agreed in advance: category sales and margin, on-shelf availability, replenishment labor;
- a check that test stores actually built the new plan.
Before and after comparisons cannot separate space from season, the weak spot in most published results. Drèze, Hoch and Purk tested in stores instead.
At DS Stream we usually start with the shelf data (planogram files, fixture dimensions, change history), because that limits the model more often than the algorithm does. It is part of our data and machine learning projects for retail chains. To try it on your shelves, bring us one category and its current planogram files.
FAQ
What is space elasticity in retail?
Space elasticity measures how much a product's sales change when its shelf space changes. A meta-analysis of 1,268 published estimates (Eisend, 2014) found an average of 0.17. Each category needs its own estimate before the number drives a shelf plan.
How many facings should a product get?
In planogram optimization the answer comes in two steps. First, enough facings to hold the stock that covers demand between deliveries. Second, an extra facing only where the model expects it to outsell the neighboring item in the same space. Drèze, Hoch and Purk found that above the threshold, facing count mattered much less than position.
How is this different from assortment optimization?
Assortment optimization chooses which products are listed; planogram optimization divides the shelf among them. When space is scarce the two interact: Hübner and Schaal (2017) found in numerical tests that solving them together beat sequential planning by up to 18% in profit.
Does every store need its own planogram?
No. Store-specific planogram optimization raises relevance and the cost of maintaining each variant. An MIT thesis using Target data framed this as a trade-off between personalization and standardization, estimating that fewer, better-chosen planograms for one category could cut operating cost. Clusters of similar stores are the usual middle ground.
