How Decathlon uses AI for sports goods range planning across Europe

Decathlon, Europe's largest sporting goods retailer with over 1,700 stores across the continent, has spent the past two years embedding artificial intelligence into the core of its range planning process. The French company now uses machine learning to forecast demand at store level, optimise assortment depth and breadth, and balance commercial targets against sustainability constraints including waste reduction and circular design. The shift addresses a longstanding tension in multi-country retail: how to offer locally relevant ranges without sacrificing the buying power and brand coherence that come from centralised planning.

The stakes are high. Apparel and footwear assortment decisions lock in costs months before products reach shelves, and mistakes lead to markdowns that erode margin or unsold inventory that must be written off. For Decathlon, which sells private-label products across more than 80 sports, poor range planning also risks disappointing customers who expect specialist knowledge in each category. At the same time, European Union regulations on textile waste and ecodesign are pushing retailers to plan for longevity, repairability and end-of-life take-back, adding new constraints to an already complex process.

The company's AI tools now sit at the intersection of merchandising, supply chain and environmental strategy. By autumn 2026, the approach has begun to yield measurable results in stock turn and waste reduction, and offers a template for other multi-category retailers navigating similar pressures.

What business problem does Decathlon's AI system solve?

Decathlon's assortment challenge stems from its scale and structure. The retailer designs and manufactures most of its products in-house through dedicated sport brands, such as Kalenji for running and Quechua for mountain sports. Centralised design and procurement teams work 12 to 18 months ahead, setting global ranges. Local merchandisers then adapt those ranges for national or regional markets, choosing which products to stock, in which sizes and colours, and in what quantities for each store.

Traditional planning relied on merchandisers' intuition, last year's sales and manual spreadsheets. That approach struggled to account for micro-local variation, such as the popularity of trail running in Alpine towns or the surge in cycling during urban car restrictions. It also led to safety stock buffers that inflated inventory and increased the risk of obsolescence. With average fashion and sports apparel stock turns of four to six times per year, even small improvements in forecasting accuracy can free significant working capital.

AI-driven assortment planning automates much of the forecasting and optimisation. Machine learning models ingest historical sales, foot traffic, local demographics, weather patterns and events calendars to predict demand at SKU and store level. Optimisation algorithms then propose ranges that maximise predicted revenue or margin within constraints such as shelf space, minimum order quantities and sustainability targets. Merchandisers review and adjust the recommendations before committing orders.

How does the system work in practice?

Decathlon's AI assortment platform combines three layers. The first is demand sensing, which uses time-series forecasting models to predict sales for each product, size and location. The models incorporate external signals including local weather forecasts, school holiday calendars and nearby sports events. For example, running shoe sales typically spike in the weeks before a marathon, and ski apparel demand correlates tightly with early-season snowfall in nearby resorts.

The second layer is assortment optimisation. Given forecast demand and inventory constraints, the system recommends which products to stock in each store and in what depth. The algorithms balance three objectives: maximising sales, maintaining minimum brand presence in each sport category and meeting sustainability criteria such as the share of durable or recyclable products. Merchandisers can adjust the weightings or impose hard constraints, for instance requiring that at least 30 per cent of textile SKUs meet defined ecodesign standards.

The third layer is continuous learning. The system tracks actual sales against forecasts and feeds forecast errors back into the models. It also runs counterfactual simulations, estimating what sales would have been had the store stocked a different assortment. Over time, the models learn which local signals matter most and which product attributes drive substitution when a preferred SKU is out of stock.

What results has Decathlon reported?

Decathlon has shared selective performance data in sustainability reports and at retail conferences during 2025 and 2026. In its 2025 sustainability report, published in March 2026, the company stated that AI-supported assortment planning contributed to a reduction in unsold seasonal inventory, though it did not publish an aggregate figure. Separately, in a presentation at a European retail technology conference in June 2026, a Decathlon supply chain director noted that machine learning forecasts had improved accuracy by several percentage points in categories with high seasonal volatility, such as winter sports and outdoor camping.

The company has also highlighted environmental benefits. Decathlon reported that in 2025 it increased the share of products designed for durability and repairability in its European assortment, supported by algorithm constraints that prioritise such items when forecast demand and margin are comparable. The retailer operates a take-back and resale programme, Trocathlon, and has stated that better demand forecasting reduces the volume of unsold goods that would otherwise enter discount channels or be destroyed.

Quantitative impacts remain hard to verify independently. Decathlon is a private company and does not publish detailed financial results or inventory metrics. Trade press coverage of the AI rollout has largely relied on company statements rather than third-party audits. Nonetheless, industry analysts note that even modest improvements in forecast accuracy, on the order of two to five percentage points, can materially reduce safety stock and markdowns in businesses with thousands of SKUs and locations.

How does Decathlon balance local relevance and central control?

One of the thorniest questions in multi-market retail is where to draw the line between central efficiency and local autonomy. Decathlon's model preserves central control over product design, sourcing and brand architecture, while giving country and regional teams influence over assortment composition and stock depth. The AI system supports that division of labour by providing local merchandisers with quantitative decision support rather than imposing top-down mandates.

In practice, central teams define the global product range and set parameters for the optimisation algorithms, including sustainability constraints and minimum representation for each sport category. Country merchandisers then use the AI recommendations as a starting point, adjusting based on local knowledge, promotional calendars and relationships with sports federations or clubs. Store managers have limited discretion to reorder fast-selling items within allocated budgets, but cannot introduce SKUs outside the approved national assortment.

This tiered approach mirrors practices at other European multi-format retailers. However, Decathlon's vertical integration gives it an advantage: because it controls design and manufacturing, it can adjust production runs in response to revised demand forecasts more quickly than retailers dependent on third-party brands. Lead times for replenishment orders on core items are often measured in weeks rather than months, allowing the AI system to drive in-season adjustments as well as pre-season planning.

How are sustainability goals embedded in the algorithms?

European textile regulations are tightening. The EU's Ecodesign for Sustainable Products Regulation, adopted in 2024 and entering force in phases from 2026, will eventually require that apparel and footwear meet minimum durability, repairability and recyclability standards. Extended producer responsibility schemes in France, the Netherlands and other member states already require retailers to finance the collection and recycling of unsold and end-of-life textiles. These rules create both compliance obligations and competitive risks for retailers whose assortments include large volumes of low-durability, hard-to-recycle products.

Decathlon has responded by building sustainability criteria into its assortment algorithms. Each product in the master catalogue is scored on attributes including material composition, expected lifespan, availability of spare parts and recyclability. The optimisation engine can apply minimum thresholds, for example requiring that at least a specified percentage of the recommended assortment in each category meets defined ecodesign standards. Merchandisers can also set preferences, asking the system to favour durable products when forecast demand and margin are similar.

The company has stated that this approach helps it meet regulatory requirements while aligning commercial and environmental goals. Products designed for longevity typically command higher prices and margins, offsetting the cost of better materials and construction. They also generate fewer returns and lower warranty claims. However, they may have slower stock turn, tying up working capital. The AI system attempts to balance these trade-offs by forecasting lifetime value rather than first-sale revenue alone.

What are the risks and limitations of AI assortment planning?

AI-driven planning is not a panacea. The most significant limitation is reliance on historical data. Machine learning models perform well when the future resembles the past, but struggle with structural breaks such as new sports trends, rapid demographic shifts or unexpected events. For example, the surge in home fitness during pandemic lockdowns caught most forecasting models off guard, as did the subsequent rebound in outdoor and team sports. Decathlon's system incorporates external signals and human override, but cannot fully eliminate forecast error in volatile categories.

A second risk is over-optimisation. Algorithms that maximise short-term revenue or margin may recommend assortments that neglect slower-selling but strategically important products, such as niche sports equipment that signals specialist credibility. Decathlon mitigates this by imposing minimum presence constraints for each sport category, but the balance between commercial performance and brand positioning remains a judgment call that merchandisers must make.

Data quality is another constraint. The models depend on accurate, timely feeds from point-of-sale systems, inventory databases and external sources. Missing or delayed data degrades forecast accuracy. Integration challenges are particularly acute in acquired businesses or markets where IT infrastructure lags. Decathlon has invested heavily in data platforms and governance, but acknowledges that inconsistent data definitions across countries remain an ongoing issue.

Finally, there is the question of transparency and accountability. When an algorithm recommends an assortment that underperforms, it can be difficult to diagnose whether the fault lies in the input data, the model logic or the merchandiser's adjustments. Clear ownership of decisions is essential for learning and continuous improvement, but also creates organisational friction when human intuition conflicts with machine recommendations. Decathlon's approach of treating AI as decision support rather than autonomous control helps preserve accountability, but requires ongoing training and change management.

How do competitors approach AI in assortment planning?

Decathlon is part of a broader shift among European retailers toward algorithmic merchandising. Sports and outdoor peers including JD Sports and Intersport have discussed investments in demand forecasting and inventory optimisation, though public details remain sparse. In the wider apparel sector, Inditex has built proprietary AI systems to allocate inventory across its Zara store network, using real-time sales data to redirect stock from slower to faster locations. H&M has partnered with technology vendors to improve demand forecasting and reduce markdowns, though results have been mixed.

Luxury and premium brands face different trade-offs. For them, scarcity and brand control often outweigh inventory efficiency, and assortment decisions prioritise editorial coherence over algorithmic optimisation. However, even in luxury, AI is used for demand sensing in categories such as eyewear, fragrance and accessories, where product life cycles are longer and sales patterns more predictable.

The common thread across segments is the integration of sustainability constraints. European retailers are converging on the view that ecodesign and circularity must be embedded in planning systems rather than treated as afterthoughts. This reflects both regulatory pressure and a recognition that consumers, particularly younger cohorts, increasingly consider environmental attributes in purchase decisions. AI assortment tools that ignore sustainability risk optimising for a set of preferences that are already shifting.

Frequently asked questions

What should fashion companies do now?

1. Audit current planning processes to identify where manual forecasting and assortment decisions create inefficiency, excess inventory or missed sales opportunities.
2. Invest in data infrastructure before deploying AI models, ensuring clean, consistent and timely feeds from point-of-sale, inventory and external sources such as weather and events.
3. Start with pilot projects in categories or markets with high data quality and clear success metrics, then scale based on validated results rather than vendor promises.
4. Embed sustainability constraints into optimisation algorithms from the outset, reflecting regulatory requirements and strategic commitments to ecodesign and circularity.
5. Preserve human oversight by treating AI as decision support rather than autonomous control, and train merchandisers to interpret and adjust algorithmic recommendations.
6. Monitor forecast accuracy and model performance continuously, using A/B tests and counterfactual analysis to measure incremental impact and diagnose failures.
7. Balance central efficiency with local autonomy by defining clear roles for global, country and store teams, and ensure the AI system supports rather than undermines that division of labour.