Assortment, Micro-Demand & Store-Level Forecasting
One supermarket has five shelves of coconut water that move slowly; three kilometres away, the same product is sold out by Sunday evening. The average demand report says both stores are “healthy” - the customer on the floor says something very different.
- Assortment is the set of SKUs a store chooses to carry; depth is how much stock it holds for each SKU.
- Micro-demand means demand varies by store, catchment, daypart, weather, festival, price and local customer mission.
- Store-level forecasting estimates demand at SKU-store-time level, not just category or city level.
- The core trade-off is availability vs working capital: too narrow causes lost sales; too broad creates dead stock, waste and markdowns.
- Use the flow: segment stores - sense local demand - forecast SKU-store - set assortment and depth - learn from exceptions.
- Track WAPE, forecast bias, service level, stockout rate, inventory turns and waste or markdown rate.
- The interview trap: using one national forecast for every store and ignoring local substitution, lost sales and shelf constraints.
Big Picture - Why Store A and Store B Should Not Look Identical
Assortment planning becomes powerful when you stop treating stores as clones. A city average hides the real operating problem: each store serves a different shopper mission, local income profile, commute pattern, weather exposure and competitive set.
Core Explanation - The Assortment Forecasting Logic
Assortment, micro-demand and store-level forecasting sit at the intersection of merchandising, analytics and supply chain. Merchandising asks, “What should this store sell?” Forecasting asks, “How much will this store sell?” Supply chain asks, “How do we replenish without choking cash or shelf space?”
The unit of thinking is usually SKU-store-time: one item, in one store or dark store, for one time bucket such as day, week or daypart. That granularity matters because demand for the same SKU may behave differently in an office district, a residential society cluster, a tourist area and a student neighbourhood.
The Five Decisions in Assortment Planning
The last step connects directly to inventory policy. Once you have a store-level forecast, you still need reorder points, review frequency and safety stock rules; that is where setting inventory policy for a multi-product business becomes the natural next skill.
The 2x2 Matrix - Which SKUs Deserve Space?
Every SKU is fighting for shelf space, working capital and planner attention. A good category manager does not forecast all SKUs with the same intensity. Core staples need tight availability. Long-tail items may be better online, in a nearby hub or ordered only when demand appears.
Micro-Demand Signals You Should Name in an Interview
Micro-demand is not magic. It is simply better signal collection. A strong answer names the real drivers that make one store behave differently from another.
A Blinkit-style dark store in a student-heavy Bengaluru catchment and one near a family-dense Gurgaon society cluster should not carry the same depth across snacks, breakfast, baby care and household staples. The primary driver is local mission density - what customers nearby repeatedly need within minutes - supported by search data, repeat purchase behaviour, substitution patterns and replenishment speed. The strategic lesson: hyperlocal assortment wins only when forecasting, fulfilment and replenishment work together.
Metrics That Prove the Forecast Is Working
Interviewers like this topic because it quickly reveals whether you can connect analytics to operations. Use 4-6 concrete measures, not vague statements like “accuracy improved”.
Worked Example - Calculating Forecast Quality
Suppose a store forecasts weekly demand for a milk SKU as 90, 130 and 100 units. Actual demand is 100, 120 and 80 units.
WAPE = 40 / 300 = 13.3%. Bias = (320 - 300) / 300 = 6.7%, so the model is slightly over-forecasting. In an interview, the important move is not just calculating the error; it is saying what you would do next - check whether the over-forecast came from promotion assumptions, stockout correction, weather, pack-size rounding or one abnormal week.
Definitions You Can Say Cleanly
- Assortment planning: deciding which SKUs a store carries, and in what breadth, depth and space allocation.
- Micro-demand: demand variation at granular levels such as store, SKU, daypart, catchment, weather or local event.
- Store-level forecasting: estimating future demand for each SKU at each store or fulfilment node over a defined time period.
- Breadth: the number of categories or product families offered.
- Depth: the number of SKUs, variants or units carried within a category.
- Lost sales: demand that existed but was not captured because the product was unavailable or not listed.
Zara: Store Feedback as a Demand Radar
Zara shows how fast fashion uses store-level demand signals to adjust assortment, replenishment and product decisions instead of relying only on long seasonal forecasts.
Fashion retail is a brutal forecasting problem. Demand changes by city, store location, size curve, season, social trend and even weather. If a retailer locks the whole season’s assortment too early, it can miss emerging demand and get stuck with markdown-heavy stock.
Zara’s parent company Inditex describes an integrated model where stores, digital channels, design, sourcing and logistics are closely connected through data and operating routines (Inditex annual reports). The important point for this topic is not “Zara is fast”; it is why the system works.
The interview takeaway: do not explain Zara as “good forecasting” alone. The primary driver is the demand-sensing-to-response loop, supported by product cycle speed, integrated inventory visibility, store feedback and flexible fulfilment.
How AI Changes Assortment, Micro-Demand & Store-Level Forecasting
AI matters here because the data is too granular for manual planning alone. A national weekly category forecast may have hundreds of rows; SKU-store-day forecasting can have millions.
- AI improves granular forecasts. Machine learning models can combine sales history with weather, festivals, local events, price changes, search behaviour and stockout flags to forecast SKU-store-day demand.
- AI recommends assortment actions, not just forecasts. A model can flag “add”, “delist”, “increase facing”, “reduce depth” or “transfer stock” recommendations by store cluster, while planners apply business judgment.
- AI makes exception management practical. Instead of reviewing every SKU-store combination, planners can focus on the few exceptions with high revenue risk, high waste risk or repeated bias.
A practical student workflow: load a retailer’s annual report, store format notes and your own topic notes into NotebookLM, then ask it to generate “10 store-level assortment interview questions with expected metrics, constraints and follow-up probes.” For a deeper operations angle, revise using AI for inventory optimisation and replenishment after this lesson.
Interview Relevance
“You are managing assortment for 100 convenience stores across a city. Some stores face stockouts while others carry slow-moving inventory. How would you improve store-level forecasting and assortment?”
Always separate forecasting error from execution error. A good forecast can still fail if replenishment is late, pack size is wrong, shelf space is unavailable or store staff do not update stock accurately.
Common Mistake
The costly mistake is using city-level or national average demand to decide every store’s assortment. It hides local missions, lost sales and substitution effects, so the candidate sounds analytical but not operational. The one-line fix: segment stores first, then forecast and decide assortment at SKU-store-time level within real shelf and replenishment constraints.