Lufthansa Life ChangingPlaces
Micro site using image recognition and machine learning for one amazing way to discover the world.
One of my first projects at MediaMonks, and one I'm still incredibly proud of to have pulled off seen its complexity. Together with DDB° Hamburg, we (MediaMonks) created a Machine Learning (image recognition) powered tool to discover the unexpected places the world has to offer. By snapping a picture of your surroundings, the image is being analyzed and then connected to an inspiring place on mother earth you probably didn't see coming. I'll talk more abount the details down here as to how it works, but the premise is: Take a picture of you surrounding via this microsite, and you'll get a great destination to discover in return. For example: Take a picture of a fire extinguisher and it might lead you to the eternal burning fires in Absheron Peninsula, Azerbaijan. Or, a picture of a cat and you might be suggested Ao Island (an island full of cute cats). Just to name some of the nearly infinite options of this tool.
This micro site experience really needed to work smoothly and deliver the best possible and most thorough responses to the input images - it needed to be relevant. This was the main thread throughout development: Making sure that it really worked. Which sounds fairly obvious, but meant we often had to take the harder route to ensure you'd always get something interesting. For this, we needed to create a formula that would filter the most relevant result(s) out of the input image, and connect it to a fitting destination. That formula was key to the magic of the experience.
It's a little over simplified, but came down to this: see what items were recognized on the input image, see what category it would fit in (manual input, which was constantly updated in the custom built CMS for this campaign), and connect the selected category to a fitting destination category.
MediaMonk's skilled UX team really shined here in the project, as they nailed down this intricate formula that we needed. What made the creation of the right formula so tricky was that the user needed to be able to read & understand why the tool came with the result that it did. So the output needed to have two pieces of information: 1. A lead that repeats what has been picked out by the object recognition so that the user understands where the result comes from and 2. The result/destination that fits that object so well.
One of the unpredictable elements of this campaign was that it was impossible to know what kinds of inputs we could expect. With privacy being an important pillar for Lufthansa, the user data collected was at an absolute minimum. All we knew was how often different objects were recognized. But that was all we needed to be able to adapt the contents and responses possible. Within the custom CMS, DDB° could quickly and easily add new responses to the system depending on the inputs that didn't have a fitting response yet. This made the tool scalable and adaptable.