Retail / Fashion, Applied AI · 2018–2019
Global fashion brand: AI in merchandising
Header image: sector-appropriate stock photography — no dedicated photography exists in the vault for this engagement (§5 of the build spec).
This is the same UK fashion retailer covered in the Global fashion brand case study above. By 2018 the business had a common data foundation in place across every function, thanks to the enterprise Data Governance Board set up during the earlier transformation programme, but its store assortment and markdown decisions were still made on judgment rather than data.
The situation
The brand's stores were carrying more stock than their footprint and footfall justified, while ecommerce availability was constrained by inventory sitting in the wrong place. Separately, in the run-up to the AW18 sale, markdown and pricing decisions were made by experienced traders using judgment rather than data, because no product on the market addressed the brand's specific markdown problem at the level needed to beat that judgment. Both problems needed solving, and neither had an off-the-shelf answer.
The challenge
The market did not have what the brand needed, on either problem.
- No off-the-shelf machine-learning product fit the way the brand's stores actually traded, across its specific range, estate and trading rhythm.
- No market product addressed the brand's specific markdown problem at a level that could beat experienced human traders.
- Store assortment planning needed a model built on usable, owned data, not a fresh data project of its own.
- Markdown decisions in the AW18 sale period were judgment-led, with the opportunity to apply machine learning clear but unproven.
What we did
We made the build-versus-buy call twice, and chose to build both times, partnering with a specialist AI partner on both capabilities.
- Built a custom machine-learning store assortment planning capability tailored to the brand's range, store estate and trading rhythm, using the common data library from the Data Governance Board as its foundation.
- Redirected inventory freed from stores into the distribution centre to improve ecommerce availability.
- Built AI-powered markdown optimisation as a separate custom capability, tuned to the brand's pricing dynamics, and trialled it through the AW18 sale period.
- Embedded both models into the weekly trading routine, so they drove decisions rather than produced reports sitting alongside the existing process.
Results
Both models earned their place in how the brand traded.
- In-store stock fell 15% with no impact on sales.
- End-of-season recalls improved from 20% to 15%.
- Ecommerce availability improved as freed inventory was redirected to the distribution centre.
- The AI markdown optimisation trial delivered a 1% margin uplift during the AW18 sale period, described internally as a material outcome on the brand's sale volume.
What made the difference
No off-the-shelf product fitted the brand's specific trading pattern, so we built with a specialist AI partner both times, using data that already had an owner and a home through the Data Governance Board. Both models were embedded into the weekly trading routine from the outset, rather than run as pilots alongside the existing process, and became part of how the brand traded.
Role
Head of Central Merchandising
Sector
Retail / Fashion, Applied AI