AI for merchandisers and planners: a practical guide
How AI supports merchandise financial planning, forecasting, allocation, replenishment and markdowns in fashion, what data it needs, what planners still decide and how to start.
KEY TAKEAWAYS Summary by the editors
- For merchandisers and planners, AI is most useful in demand forecasting, allocation and replenishment, markdown and price optimisation, and in explaining plan versus actual variances.
- Most planning AI is predictive machine learning running on sales, stock and product attribute data, not generative AI, although language models increasingly act as an interface to it.
- A 2026 Fraunhofer IWU project with a German textile manufacturer showed that a neural network forecast could explain 82.7 percent of sales fluctuations while planners kept the ability to review and adjust it.
- In January 2024 Marks and Spencer announced an integrated planning platform for Clothing and Home covering financial planning, assortment, forecasting, allocation, replenishment and markdowns, to be rolled out over three years.
- Planners remain responsible for targets, trade-offs between margin and availability, and for overriding forecasts when they know something the data does not.
AI helps merchandisers and planners by producing faster, more granular forecasts and by proposing allocation, replenishment and markdown decisions that would take days to calculate by hand. The planner's role moves from building numbers to setting targets, checking model output and making trade-offs, and the quality of the result depends far more on clean data than on the algorithm.
What do merchandisers and planners actually do with AI?
Merchandise planning in fashion translates strategy into numbers: sales, margin and stock plans by category and channel, open-to-buy budgets, option counts, size and colour depth, and in-season decisions on allocation, replenishment and markdowns. The work is highly quantitative and repetitive, which makes it a natural fit for predictive models. Language models add a second layer: they can explain variances in plain language, answer questions about a plan and draft commentary for trading meetings.
Retailers are consolidating these steps. In January 2024 Marks and Spencer announced that it would move its Clothing and Home business onto one integrated planning platform covering merchandise financial planning, assortment planning, forecasting, allocation, replenishment, promotion and markdown optimisation and supplier collaboration, rolled out over three years, according to Just Style.
Where does AI help in merchandise planning today?
| Task | What AI does | Data needed | Maturity |
|---|---|---|---|
| Demand forecasting | Forecasts sales by style, size and location from history, seasonality and drivers | Sales and stock history, prices, promotions, calendar | Established |
| Allocation and replenishment | Proposes initial allocation and top-ups per store or channel | Store clusters, sell-through, stock positions, lead times | Established |
| Markdown and price optimisation | Recommends timing and depth of markdowns to clear stock at best margin | Price history, elasticity, stock and sell-through | Established |
| Open-to-buy monitoring | Updates remaining budget as sales and orders change and flags risks | Planned sales, markdowns, stock, open orders | Emerging |
| Variance explanation | Explains plan versus actual differences in natural language | Plan and actual data, product and store attributes | Emerging |
| New product forecasting | Estimates demand for styles without history using similar items and attributes | Rich product attributes, images, comparable sales | Emerging |
| Autonomous replanning agents | Adjusts plans and orders automatically within rules | Integrated planning data and approval rules | Experimental |
What results can planners realistically expect?
Published results are scarce and often come from vendors, so planners should be cautious. One independent example: in June 2026 Fraunhofer IWU reported on an AI forecasting tool built for the German textile manufacturer frottana (MÖVE brand). Using neural networks on four years of sales history, the model explained 82.7 percent of sales fluctuations and deviated from actual monthly sales by about 9 percent on average. Fraunhofer stressed that employees can review, adjust and enrich the forecasts with their own expertise.
The lesson is that useful accuracy is possible even with limited history, but only when the data is consistent and the planners stay involved. Value usually shows up as less time spent on manual calculation, earlier warnings and fewer extreme errors, rather than as a perfect forecast.
What data do planning models need?
- Clean sales and stock history at SKU, size and location level, with stock-outs flagged so that lost demand is not read as low demand.
- Product attributes that are consistent across seasons, so that new styles can be linked to comparable past styles.
- Price and promotion history, including markdown dates and depths.
- Plan data: targets, budgets and open-to-buy, which Shopify's guide defines as planned sales plus planned markdowns plus planned end-of-period stock minus planned beginning stock.
- Supply constraints: lead times, minimum order quantities and delivery reliability.
Gartner reported in February 2025 that 63 percent of organisations either lack or are unsure whether they have the right data management practices for AI. In planning, the typical gaps are inconsistent attributes, missing stock-out flags and separate systems for wholesale, retail and e-commerce.
What stays human in merchandising?
Planners set the targets the models optimise for, and those targets involve trade-offs: availability against markdown risk, newness against proven sellers, margin against market share. Models also do not know about a competitor's store opening, a delayed shipment, a marketing push or a change in brand direction unless someone tells them. Planners should own the override process, record why they changed a recommendation, and review overrides after the season to see whether people or the model were right.
Planners also act as translators between the model and the rest of the business. Trading teams, buyers and finance need to understand why a recommendation changed, so planners who can explain the main drivers of a forecast in plain language build the trust that adoption depends on.
Which skills should merchandisers and planners build?
- Understanding forecast error measures and confidence ranges, not only point forecasts.
- Setting and testing business rules and constraints that models must respect.
- Using AI assistants to query data and draft trading commentary, and checking every figure.
- Running controlled tests, for example AI-assisted allocation in some stores and the usual process in others.
- Working closely with data and IT teams on attribute quality and data definitions.
How can a planning team start in 30 days, and what are the risks?
The main risks are false precision, models learning from distorted history (stock-outs, past mistakes), black-box recommendations that planners cannot explain to trading teams, and integration effort that is larger than expected. Language model summaries can also misstate numbers. Starting with one decision, a control group and transparent override rules keeps these risks manageable.
Frequently asked questions
How is AI used in merchandise planning?
AI is used to forecast demand, allocate and replenish stock, optimise markdowns and prices, monitor open-to-buy and explain variances. Most of this is predictive machine learning on sales, stock and product data, with language models increasingly used as an interface for questions and commentary.
Will AI replace merchandise planners?
AI automates much of the calculation in planning, but planners still set targets, handle trade-offs and bring knowledge the data does not contain. The role shifts towards supervising models, managing exceptions and working with data teams.
What data is needed for AI demand forecasting in fashion?
At minimum, consistent sales and stock history by SKU and location, stock-out information, price and promotion history and reliable product attributes. For new styles, attributes and comparable past items matter most. Missing or inconsistent data is the most common reason forecasting projects disappoint.
How do you measure whether AI planning works?
Compare AI-assisted decisions with a control group or with back-tests on past seasons. Useful measures include forecast error, sell-through, availability, markdown rate and time saved. Agree the measures with finance before the pilot starts.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
