How Flügger grew online revenue 13% with AI-generated color pages.
40,000 paint colors. Four markets. Mathematical pigment data turned into emotional, multilingual copy that customers - and search engines - could finally read.
Preface
In a technical world, a color is a collection of mathematical values. To people, it's something else entirely - colors add life to the world, evoke emotions, carry stories.
During the lifetime of Malgodt.dk, the team behind it (Rasmus, Lars and Kresten) did remarkable work bringing those numbers out into easy-to-discover color pages. It may have been one of the reasons the boutique paint shop was so successful. The idea was carried over to Flügger, and when ChatGPT arrived, it was time to see how far we could take it.
The problem
Flügger has a wide range of colors - different charts, collaborations, themes. With a fresh round of webshops launching, it was time to help painters and curious future customers actually discover the emotions delivered through colors they might want to own - not just scroll past hex codes.
The solution
There was huge potential to use large language models to rewrite the mathematical properties into natural language for the most desired and purchased colors. Together with Kresten Kjær Larsen and Lene Hessilt Kjær, I built a versatile prompt that produced detailed, multilingual descriptions for each color - small pieces of text that became valuable additions next to each color image.
How it was done
We took confidential data from Flügger Labs - how bright each color is, how many pigments it carries, its coverage on walls, how it tints - and created natural-language mappings to that color-tinting data. The descriptions described the colors without overtly stating the technical details.
To further protect the data, we put a zero-data-retention agreement with Azure in front of the GPT-4 endpoints. The data, paired with the custom prompt, was sent to the most powerful model available at the time. To this day, GPT-4 remains one of the best out there for this kind of task.
Results
Online color-page revenue grew approximately 13% from the first month after release to the eleventh.
Context
- The first color pages were introduced at the end of Spring 2023, giving them a generous quarter to become widely discoverable on search results.
- Statistics cover September 2023 to July 2024 (11 months). Numbers may carry small inaccuracies due to limited GDPR-compliant tracking in some cases, and have been normalized to lower ranges.
- Net revenue across flugger.dk, flugger.se, flugger.no and flugger.pl, adjusted to local currencies.
Wins
- Color pages had a clear positive impact on revenue - visible enough to act on.
- For the company, it's the return on investment that counts and that enables data-driven decisions on what to do next.
- For the customer, it's an extra layer of information that helps them discover colors - especially valuable in the noisy online search where spammy results tend to drown out the useful ones.
Comment from the team
"The distance from the technical to the emotional narrative of a colour is an artform in itself, which the usage of LLM is a good start of bridging." Peter Møllebjerg Andersen CDO, CIO & CMO - Flügger Group A/S
What does it mean for you?
Every company, from a small startup like Malgodt.dk to a well-established player like Flügger, can use AI to enhance the business.
There's often a lot of valuable data locked behind difficult-to-interpret values. That's where I'm extra passionate about the power of AI: with the right engineering, patience in testing, and small enough tasks, we can unlock entirely new perspectives. That was the great promise of the internet in 1993 - and today, we have the technology to make it real.
Credits
It was a great pleasure working with the team at Flügger.
At the time of development I was employed by Flügger. I was not sponsored to write this article, nor did I receive any compensation for it. It is analyzed and written to the best of my ability and in good faith as an independent developer with a passion for AI-applied software.