Assortment optimization refers to the process of selecting the right mix of products to offer to different customer segment in each store or channel. The primary goal is to maximize sales and profits by tailoring the product offerings based on consumer preferences, local demand, store size, and seasonal trends. This process focuses on balancing product variety and product depth that meet consumer needs while also reducing overstock and understock issues.
Assortment optimization advantages result from modeling the directions in which sales flow when SKUs are discontinued or introduced, and so every retail cluster has a spread that justifies its place.
Retail assortment optimization involves selecting the right product for each store or group of stores to stock, drawing on local shopper demand, shopping behavior, space, stock, and category goals. In sectors where retailers limited their assortment by 10% while retaining as much of the high-value merchandise in the store as possible, the sales and gross profit uplift ranged as high as 7%. The most constraining factor varies by sector: freshness (grocery), shelf space (convenience), size curves and short seasons (fashion), and fast modelling cycles (consumer electronics).
| Element |
Description |
| Definition |
Selecting the optimal product mix for a store, channel or customer segment |
| Primary goal |
Improve sales, margin and category performance |
| Key inputs |
Sales, demand, customer, inventory, space and product data |
| Key decisions |
Add, maintain, expand, reduce or delist products |
| Optimization level |
SKU, store, cluster, channel or category |
| Common outputs |
Recommended assortment, SKU rationalization and localized product mix |
They sound alike but happen at different times in the retail flow. Assortment planning is the selection of what your seasonal assortment of choices is by category; assortment optimization selects from that SKU assortment to go into each store or group.
|
Assortment planning |
Assortment optimization |
| Core question |
What range should we buy this season? |
Which SKUs belong in each store or cluster? |
| Decision level |
Department, category, channel |
SKU, store, cluster |
| Timing |
Pre-season, tied to the merchandise financial plan |
Pre-season and in-season, refreshed as demand shifts |
| Key inputs |
Sales history, open-to-buy budget, trends, vendor terms |
SKU performance, demand transfer, space, local demand |
| Output |
Range plan, buy quantities, category breadth and depth |
Store-level mix, add and delist list, localized assortment |
- Ignoring demand transfer: delisting SKUs without modeling where their sales will move.
- One range for every store: applying a chainwide mix to clusters with different shopper profiles.
- Ranking on sales alone: overlooking margin, availability, and each SKU's role in the category.
- Treating the mix as fixed: skipping cluster-level refreshes between category resets.
- Planning in isolation: changing the range without checking price gaps and the promotion calendar.
Range allocation is controlled by the planning process and, once that range has been established, assortment optimization ranks each SKU in relation to store-level demand, simulates where sales will transfer as the range adjusts, and determines the optimal mix of SKUs for store clusters.
- Collect data: sales, margin, inventory, customer, store and product.
- Segment stores and customers: cluster stores or channels by demand and shopper profile.
- Analyze SKU performance: sales, margin, velocity, availability and each product's role.
- Model demand and substitution: estimate where sales move when SKUs are added, removed or changed.
- Optimize the assortment: recommend SKUs to add, keep, expand, reduce or delist per cluster.
- Monitor and refine: track results and update the mix as demand shifts.
Regardless of the model chosen, assortment optimization drives additional sales and margin by driving increased shelf and stock presence for shoppers’ purchased SKUs. Out-of-stocks and overstocks alone cost retail an estimated $1.7 trillion drain on retail each year.
- Increase category sales: top sellers get more space and facings.
- Improve gross margin: low-margin SKUs with close substitutes exit the range.
- Reduce slow-moving inventory: fewer tail SKUs sit waiting for markdowns.
- Improve product availability: freed space goes to high-demand items.
Current models go further than forecasting SKU sales. Demand transferability models estimate how much of a delisted SKU's volume moves to substitutes and how much a new SKU cannibalizes. Attribute-based models forecast items with no sales history, and optimization solvers fit the range to shelf space and listing commitments. Agentic AI keeps that loop running: agents flag a store cluster when sell-through or margin drifts, test range changes and route recommended moves to category teams for approval. ProfitPulse, Polestar Analytics' agentic RGM suite, tests each move against pricing and promotion plans before execution.
A range change alters the price ladder shoppers see. Delisting a mid-tier SKU can push shoppers up to a premium item or down to private label. A promotion on an item whose close substitute was just delisted will also report inflated lift, because transferred volume looks like promotional uplift. Integrated assortment, pricing and promotion analytics models all three levers on the same demand base, so each range decision is checked against price gaps and the promotion calendar.
CPG manufacturers use it to win and defend distribution. Retailer-specific customer decision trees and demand transferability show which of their SKUs bring incremental category sales and which only split volume within their own range. That evidence strengthens listing proposals in joint business planning and guides rationalization of their own portfolio. Polestar Analytics' CPG analytics practice supports brands with assortment and merchandise planning.
Full resets usually follow the category review calendar. That is often once or twice a year in grocery, and at each season or launch cycle in fashion and consumer electronics. Between resets, rerun the model when a cluster's velocity or margin drifts past agreed thresholds, or once a new item has enough weeks of sales to read. Localized assortments need these cluster-level refreshes more often than a chainwide range.