Imagine a retail group reviewing the same homeware range for a Dubai destination mall, a smaller Sharjah branch and a Riyadh neighbourhood store. Head office ranks products by sales and proposes one cleaner catalogue. The store teams object: different shelf space, different basket sizes and different reasons for visiting. Retail assortment planning is the decision about which customer needs each location will serve, and which products deserve the space and cash required to serve them.
Adding an AI recommendation engine does not settle that decision. A model can rank what sold. Leadership must decide what the range is meant to achieve, which constraints are real and what evidence would justify removing an item. I would begin with one category and four decisions: define its job, group comparable stores, test the proposed mix and control the change.
Retail assortment planning starts with the category's job
Write down why the category exists. Is it bringing customers into the store, completing a wider purchase, creating margin or meeting a recurring everyday need? Different roles justify different ranges. A low-volume accessory may complete a profitable purchase; another may simply occupy a shelf because nobody has challenged it.
Choose the planning period and the unit of decision. For homeware, that might be individual products and pack sizes; for fashion, style, colour and size availability matter differently. Separate the number of choices from the quantity held behind each choice. More options and deeper stock are different uses of the same budget.
Oracle's assortment planning documentation separates assortment strategy, product selection for store clusters and, for short-life products, item flow. That is a useful distinction even without that software. Approving a range should lead to an executable buying and replenishment plan, not just an attractive selection board.
Put the constraint beside the objective: available shelf or fixture space, working capital, minimum supplier orders, delivery frequency and category commitments. If management demands more choice, less inventory and perfect availability, ask which trade-off it will accept. The optimiser needs a decision, not three incompatible slogans.
Group stores by evidence, not the country label
Begin with a small number of store groups that the merchandising team can operate. Compare category demand, selling space, price mix and replenishment conditions. A large store and a small store in the same city may need different ranges. Two stores in different cities may share a useful core.
Oracle's location clustering guidance explicitly balances more targeted assortments against the work of managing additional items, suppliers and planograms. Treat that workload as part of the business case. Fifty locally tailored ranges are not progress if the buying team can reliably maintain only five.
Use a common core where the evidence supports it, a local extension for demonstrated differences and a bounded space for experiments. Document store exceptions with a reason and review date. Avoid letting an exception become permanent because a senior manager once requested it.
Check the calendar behind the comparison. Ramadan and Eid trading, tourism peaks, school periods, promotions and store openings can distort a simple year-on-year view. For a UAE and Saudi operation, align the relevant trading conditions explicitly. Do not interpret a temporary demand pattern as a permanent difference between markets.
Ask what the sales history could not observe
A product with weak recorded sales may have been unavailable, poorly displayed or introduced too late. Before delisting it, inspect stock availability, ranging dates, price changes, promotions and returns. A zero in the sales file does not tell you whether customers rejected the product or never had a fair chance to buy it.
This is related to retail demand forecasting, but the executive decision is different. Forecasting helps estimate quantities under stated conditions. Assortment planning decides which alternatives will be offered together. Changing those alternatives can change where demand goes.
Oracle's demand transference documentation models how assortment changes redistribute demand and describes why promotions complicate that analysis. The practical question is whether a removed item's customers will accept another product, buy elsewhere or abandon the purchase. Historical sales alone should not be treated as proof of complete substitution.
Ask the analyst to show the assumption for each proposed removal. Is there an equivalent pack size, price point and use case? Does an apparently similar item serve a different need? Where evidence is weak, compare conservative and optimistic substitution scenarios. A proposal that only works when every customer switches deserves a smaller test.
Approve a range using a short decision sheet
For each store group, record the customer need, retained core, additions, removals and expected operational effect. Show contribution after relevant handling and inventory costs, with space and cash requirements visible. Keep supplier funding separate enough that a temporary payment cannot disguise a range customers do not want.
Require an explanation for protected items. Some may support a deliberate entry price, a complete size run or an essential companion purchase. Protecting them can be sensible. Hiding them from the decision makes it impossible to distinguish strategy from sentiment.
Keep newness affordable. Set an agreed share of space or purchasing budget for unproven products, state what would earn them a continuing place and schedule the review. Those limits are management choices, not universal industry benchmarks. They make experimentation possible without quietly expanding the permanent range.
The final sheet should fit the buying meeting: proposed mix, binding constraints, uncertain assumptions, named owner and stop conditions. If the model cannot explain why a removal improves the category after likely substitution, it is providing a ranking rather than a defensible decision.
Test the category result and the store workload
Pilot the change in comparable locations while keeping a comparison group where practical. Record the old assortment and the date the new range actually reached the shelf. Plan the evaluation period around the category's purchase cycle, and note price, promotion or availability changes that could explain the result.
Measure category contribution, availability, stock investment, basket effects and customer requests for missing items. Include transfers, markdowns and store labour caused by the change. Sales of the replacement item may rise while total category value falls. That is not a successful substitution.
Make the operational handoff explicit: purchasing stops or reduces old orders, stores receive a clear shelf plan, ecommerce reflects the approved local availability and remaining stock has a destination. This is where business automation helps carry an approved merchandising decision across systems. It should not make an uncertain range decision irreversible.
Retail assortment planning earns its place when every addition and removal has a commercial reason the store can execute. Approve the smallest range change that can prove its value. A shorter product list is an administrative result. A better category is the business result.