Our approach towards accuracy
updated about 2 years ago
We focus hard to balance user friendliness and advertiser impact. Balancing requires to make continuous improvements on different levels. We outline below some of the most important efforts we took.
User friendliness is about...
Low battery consumption: we keep consumption below 2% on average days for the average user. That is why we don’t track every signal every second, but work hard to improve our SDK tracking technology.
Advertiser impact requires us to strive for optimal quality on multiple levels:
Locations: we have a POI database of 150.000+ locations with meta data such as category, opening hours, popular hours, brand, … of which most accompanied with its building shape to allow for polygon matchings (i.o. radial).
Trackings: we optimize tracking algorithms to minimize obsolete trackings (at home, non matchable, while driving...) and optimize to grow the number of relevant trackings (near POIs, between opening hours…). This way we collect between an average 50 and 400 location points per user per day.
Algorithm: our algorithms take as much metadata into account as possible and will improve over time via self-learning algorithms (machine learning, AI). E.g. for matchings we use distance, accuracy, the number of stores (alternative matches) nearby as well as info about opening hours, popular hour and popular days for the POI, even weather impact amongst others to alter the propensity of a match. On top of that, historic behaviour is constantly updated and used for improving insights (earlier visits to H&M and never Inno, means chances are highest the user visited H&M again instead of Inno).
Impact is the result of high quality on all three levels combined. This goes for matching, segmenting, predicting as well as for attribution score calculations.