Using Machine Learning to Isolate the Impact of TV Work

We had to create, from other bits of the u k, a control that looked like birmingham. And you'd select somewhere else, that would be your control for your birmingham test. We're still spending on t v now. So maybe the headline in terms of the impact of the test was that t v work, which is why we're still doing it. The way we stitched all those body parts together was through machine learning. At any one time during that eight week test, we had 200 thousand of those models running to mimic what we observed in the greater birmingham area.