How adidas uses AI: from design archives to generative content

KEY TAKEAWAYS Summary by the editors

1. Since 2022 adidas designers have used AI Archive, a tool based on diffusion models trained on the brand's sneaker collection from the 1950s onwards, to explore footwear concepts.
2. In a pilot of AI coding assistants with 500 engineers, 82% used the tool daily and 91% found it useful, adidas's Fernando Cornago reported in 2025.
3. adidas built a retrieval-augmented generation tool that analyses more than 2 million product reviews for over 50 decision-makers in design, product, marketing and service.
4. adidas has not publicly described the AI models it uses for demand planning, so claims about AI-driven forecasting at adidas are not supported by its own disclosures.
5. Most published adidas AI figures come from conference talks, vendor case studies or award entries rather than audited company reporting.

adidas uses AI across design, engineering and marketing: a generative tool trained on its sneaker archive for concept creation, coding assistants for its engineers, a language-model tool that analyses customer reviews and generative personalisation in campaigns. It has built a large digital and data foundation since 2018, but it has not publicly described the AI models it uses in demand planning.

What has adidas built?

The documented applications span the value chain:

• AI Archive: a web application, deployed in 2022 and described by adidas researchers at SIGGRAPH 2023, that uses diffusion models trained on adidas sneakers from the 1950s onwards so designers can explore hundreds of concepts quickly.
• Engineering assistants: a pilot of generative AI coding tools with 500 engineers, presented by adidas's Fernando Cornago in 2025.
• Review analysis: a retrieval-augmented generation (RAG) chatbot that searches more than 2 million product reviews and serves over 50 decision-makers in design, product, marketing and customer service.
• Generative campaigns: a Campus sneaker campaign in which consumers generated personalised designs from uploaded images, documented by WARC in December 2025.

Beneath these sits a broader digital programme. The Business of Fashion reported in 2021 that adidas planned to invest more than 1 billion US dollars in digital transformation by mid-decade, hire more than 1,000 data and technology staff that year and grow its membership programme from 220 million members towards 500 million by 2025.

Why did adidas invest in AI?

The motives differ by use case. In design, the aim is to explore more concepts faster while staying true to the brand's design vocabulary, which is why the model was trained on adidas's own archive. In engineering, the aim is productivity: Cornago described generative AI tools as already a commodity for engineers that could not be taken away. In customer insight, millions of reviews contain product feedback that teams cannot read manually. In marketing, the goal is personalisation at a cost that makes it feasible beyond a few hero campaigns.

Read also

How Zalando uses AI: the assistant, personalisation and size advice

How does it work (data, models, process)?

The common pattern is proprietary data plus general-purpose models. AI Archive applies diffusion models to adidas's own product history, giving designers control over the output. The review tool converts reviews into vector embeddings, retrieves the most relevant ones for a question and passes them to a language model to summarise; according to the case study, reducing the input to about 3,000 tokens per query cut response time from 15.5 to 6 seconds. The coding assistants were rolled out as a measured pilot before wider use.

The review tool also illustrates a governance pattern: the case study describes a central catalogue for data and model governance and experiment tracking, so that the teams using the answers can trace which data and model version produced them. Data foundations matter elsewhere too. According to The Business of Fashion, adidas already generated more than 5 billion euros of sales from products created with 3D design, which gives AI design tools digital assets to work with.

What results has adidas reported?

Published figures exist, but almost none come from audited company reports:

adidas has not disclosed the effect of AI on revenue, margin, inventory or product development cost at group level, nor how AI is used in demand planning.

What are the limits and open questions?

The following is editorial analysis. The biggest gap is planning. Demand planning is where AI arguably has the largest financial lever for a sportswear company, yet adidas has not described its approach publicly, so any claim that adidas runs AI-driven forecasting should be treated as unverified. Second, the evidence base is uneven: vendor stories and award entries are selected to show success and rarely report baselines or failures. Third, archive-trained design tools raise questions about how much they reinforce existing aesthetics instead of producing new ones. Fourth, engineering productivity gains from coding assistants are self-reported and hard to translate into business value.

Read also

How Mango uses AI: generative design, virtual models and Mango Stylist

What can other fashion companies learn?

1. Train or ground generative tools on your own archive and data; that is where brand differentiation comes from.
2. Use RAG to unlock unstructured feedback such as reviews, and keep the context small to control cost and speed.
3. Pilot internal productivity tools with a defined group and measure adoption before scaling.
4. Separate marketing experiments from operational AI; a successful campaign says little about planning maturity.
5. Ask vendors and partners for baselines, not only uplift percentages.

Analysis: adidas illustrates a common pattern in large brands. Generative AI spreads quickly where the risk is low and the data is owned, as in design exploration, internal coding and review analysis, while AI in core planning moves more slowly and is discussed less openly.

Frequently asked questions

Does adidas use AI to design shoes?

Yes, for concept exploration. adidas researchers described AI Archive, a tool deployed in 2022 that uses diffusion models trained on the brand's sneaker collection from the 1950s onwards. Designers use it to explore concepts; development and final decisions remain with design teams.

Does adidas use AI for demand planning?

adidas has not publicly described the AI or machine learning models it uses in demand planning. Its documented AI applications concern design, engineering, customer reviews and marketing.

How does adidas use generative AI in marketing?

One documented example is a Campus sneaker campaign in which consumers generated personalised designs from images they uploaded. According to an award entry summarised by WARC, it was associated with a 12% increase in e-commerce sales.

What did adidas learn from rolling out AI coding assistants?

In a pilot with 500 engineers, 82% used the tool daily and 91% found it useful, and two-thirds showed measurable productivity gains, according to adidas's Fernando Cornago. He concluded that such tools had become a commodity that engineers expect.

GuideThe complete guide to AI in fashion e-commerce, marketing and retailRead the complete guide