Sign in with Google 
Andreas Castillo Kristensen is head of digital media at Danish home furnishing retailer Jysk. His role involves leading all global search activities, including SEO and optimising performance across the retailer’s 29 markets.
At Jysk — a home retailer with Scandinavian roots and global reach — we invest heavily in digital media. Our Shopping ads are a big part of this, and we know that to truly succeed in search, we must continuously optimise our product data feeds.
We had a strong foundation for our search activities, delivering consistent results across markets. But as our ambitions grew, so did our understanding of what was possible.
This wasn’t a question of resources or capabilities — our teams had both.
Instead, it was about unlocking agility in how we worked with our central product data. The goal was to find a way to rapidly enrich our product feeds to better match how consumers search — and ultimately shop — on via search ads.
We needed our paid search specialists to test, learn, and act on insights at the pace the market demands.
The data deficit: Why AI performance needs high-quality feeds
To do this, we needed to create the right conditions for continuous learning, faster experimentation, and smarter optimisation of our Shopping ads campaigns.
The future of search rewards rich, structured, and high-quality data. We were already using sophisticated, AI-driven tools like Smart Bidding that thrive on detailed information. But providing these tools with basic, unoptimised data is like owning a race car but only having access to regular fuel. You’ll still move, but you’ll never reach top speed.
“When we first identified the opportunity within Jysk’s product feed, we knew we needed a dedicated environment for rapid experimentation,” says Mathias Hillerup Larsen, co-founder of s360. “We introduced our proprietary tool, Quantum, to act as that agile layer. It gave us the power to ideate and execute complex data enrichments — like AI-generated titles and visual optimisations — completely independently of core systems.”
We were excited when s360 suggested Quantum as a way to work on and enrich our data without interfering with the company’s master data systems.
“The challenge for large organisations is often the same: product data is built for stability, not necessarily for the rapid testing cycles required by modern search.”
Creating the digital playground for experimentation
The breakthrough came together with our agency, s360, and a solution that acted as an agile layer between our master product feed and our advertising channels. Called Quantum, this was a proprietary technology built on Google Cloud.
“The challenge for large organisations is often the same: product data is built for stability, not necessarily for the rapid testing cycles required by modern search,” explains Mathias Hillerup Larsen, co-founder of s360. “We built Quantum to act as a bridge, allowing marketing teams to enrich and test data in real-time without affecting the integrity of the core internal systems or central data.”
Quantum takes our existing data, allows us to add new or missing attributes, and sends that optimised version directly to our Shopping ads.
We started by looking at data gaps that directly impacted our ads’ Quality Score. This is a rank of how relevant or useful your ad or landing pages are to someone searching for your keyword.
A key early win involved product reviews. While we had customer reviews available internally, the data wasn’t structured correctly for search results. Using Quantum, we mapped this existing data into a format search engines could recognise.
This small change — making non-visible data visible — allowed shoppers to see star ratings and build trust right into the ad.
To ensure we were making the right moves, we applied a rigorous testing framework. “We split Jysk’s products into a control and a variant group. We’d test a change on the variant group, for example adding a review, to see if the new ad got more clicks. We’d then switch the groups (adding the review to the control group) halfway through to remove any statistical noise,” adds Hillerup Larsen. “This ensures that when we call something a winner — often with 98% statistical significance — we are certain of its impact.”
We found that around ~60% of our tests were winners and the new process allowed us to fail fast and recover instantly.
This setup gave us what we now call our “playground” — a dedicated space for experimentation with zero internal risk.
Using Google Gemini’s Veo and Nano Banana for visual excellence
Data enrichment isn’t just about text; it’s about visuals, too. As the AI era advances, the quality of ad assets is paramount. We began investigating tools like Gemini’s Nano Banana for image generation and Veo by Google Gemini for video generation.
“Where we had only standard pack shots, we could create product images that fit the platform’s exact requirements with Nano Banana, or use Veo to optimise our display and video assets. All of this enhances the brand experience,” says Hillerup Larsen.
The unstoppable results and our improvement loop
These fast, incremental changes had a disproportionately large impact due to our scale.
By extrapolating the results from our tests, we estimate this change led to an increase of approximately 20 million clicks on our Shopping ads. This success was further validated when we won the Global Search Award for “Best Global PPC Campaign: Large Winner”.
The true victory is the strategic shift this project enabled. It has changed how our team operates and collaborates with the rest of the business.
Our process is now a continuous loop. We test an idea and if it succeeds, we no longer approach the wider organisation with a hypothetical request; we come to the table with a proven winner. This shifts the conversation from a marketing “belief” to a business certainty.
We aren’t stopping here. We have an extensive roadmap of upcoming tests. For instance, we are currently testing our title strategy to remove “fluff” descriptors like “optimal comfort” or “elegant design” to see if cleaner, attribute-heavy titles improve matching.
This project has given us the tools to test quickly, find the winners, and loop the successes back into the main strategy. We haven’t just fixed a data feed, but fundamentally influenced the future of our marketing’s role in data governance.
Social Module
Share