CREATING ADS THAT DON'T SUCK
We where trying to tackle fashion with JustVisuals machine learning visual search. We had been trying a lot of experiments one of them was a fashion app called LikeThat Style. Which allows a user to take a photo of a piece of clothing they wanted and we would find something visually similar. What we noticed in user testing was users would be very disappointed if the results where not an exact match to the photo they sent in. I was tasked with if we could use Machine learning visually search in a way that would minimize the expectation of exact match yet still be an enjoyable experience to the end user. We realized if you took the user out of a product specific search frame of mind and put visually similar results in a browse context the user found the results to be very enjoyable and useful.
We looked at bloggers and how they where monetized their content today and also who where the main players. In testing to maximize viewing the final ad was an in-image unit which fit nicely into most bloggers pages and gave us a very high click through rate. Also this ad unit saved them a lot of time compared to other units which made them hand pick the products to display. With our unit all the bloggers have to do is ad one line of code to their blog and they are done.
In testing we found our ads on fashion blogs would generate a 3x lift over other fashion related ads and even higher compared to standard run of site ads.
SPEEDING UP A BLOGGERS TIME WAS KEY
The biggest finding where a 3x lift over other fashion ads and the amount of time this saved the blogger. This allows the blogger to utilize their time more efficiently.