The German fashion and lifestyle retailer is using AI across its online business. Personalised product rankings. Virtual try-on in its app. 

At Big Data & AI World Frankfurt, Iris Hollick and Dr Richa Sharma from Breuninger’s data science team explained how the retailer tries to turn AI projects into measurable business value. Here’s three things they learned along the way: 

 

1. Ownership Comes First 

Breuninger’s earlier AI work was often spread across different teams, with unclear ownership and similar problems being tackled in different ways. Generative AI helped the team move faster, but it didn’t solve those problems.  

“If you have a lot of speed and you do not have ownership, that only means faster confusion”, Hollick said during the session. 

That’s why the retailer now defines responsibilities at the start of a project. A data scientist is responsible for building and validating the model. A business owner is accountable for the final outcome. Product and engineering teams are consulted when the solution needs to be integrated. 

“Models don’t create value, but owners do”, Hollick said. 

2. The Right KPI Depends on the Project 

Breuninger’s personalised product rankings are one example. The model uses customer behaviour, including purchases, product views and brand preferences, to decide which products are shown higher up for each shopper.  

The project increased revenue per session by 7%. During the Q&A, the team also said one tested variant increased conversion by around 2–3%.  

“What really defines success is that we have the right KPIs for the right use case,” Dr Sharma said. 

For personalised rankings, that means looking at measures such as conversion rate, click-through rate and revenue per session. A sales forecasting model would need to be judged differently.  

3. Not Every AI Project Should Make It to Scale 

Once an idea is technically possible, the team asks whether the likely business impact justifies the cost and effort. If it does not, the project can stop there. 

During the Q&A, Hollick gave the example of a recent A/B test for a product that didn’t work out as hoped. It involved real-time forecasting and would have been expensive to keep running.  

“The innovation was super cool, but we needed to stop it”, Hollick said. She added that the work could still be reused for a different product in future. 

 

The takeaway? “AI creates value when ownership, governance and measurement work together”, Hollick said. 

And one thing to remember: “The biggest risk is not failing but failing too late.” 

Find out more about the session, Beyond the Model: Turning AI Ownership & Governance into Measurable Value, and meet speakers Iris Hollick and Dr Richa Sharma on the Tech Show Frankfurt website.