From the very beginning we decided to compile the semantic core manually in order to prevent non-targeted traffic, which could appear if we had approached automatic setting based on the feed. That is why we requested the client to differentiate the tires according to their model, size, season, radius or brand.
We compiled the semantic nucleus, divided up the words into groups. It turned out to be more than 106 000 key words in total. Direct search advertising worked in the most effective way, though dynamic smart banners did well, too.