[MUSIC] A lot of big data and analytics is used to make conclusions about cause and effect. That is why I've given you a check list to determine whether you can trust these conclusions. I have called this telling apart good analytics from bad analytics. Now suppose that you apply the checklist, and you find that some analytics fails the checklist. Does this mean that the analytics is automatically useless? Actually, no, because analytics does not have to be about cause and effect to be useful. Let me give you an example. I recently saw a presentation by a major retailer. The presenter showed the graph like this. He had segmented consumers into retail-only shoppers, online-only shoppers, and omni channel shoppers, meaning that they shop both in retail stores and online. Omni channel shoppers, the right column, were much more valuable to the retailer than shoppers who only use one channel. The causal interpretation of this graph is this. If I could get retail only shoppers to also embrace the online channel and therefore become omni-channel shoppers, this would double the revenues that are received from them. Makes sense? Now let's apply the checklist. Could there be differences other than what channels they use between retail only shoppers and omni-channel shoppers? Pause the video. Take a minute to think about this and then come back. I asked you think about whether there could be differences, other than what channels they use, between retail-only shoppers and omni-channel shoppers. Sure there could. For example, we know that online penetration among poor consumers is lower. So retail only shoppers are likely to have less income than omni channel shoppers. So if I have somehow managed to get retail-only shoppers to also embrace the online channel, should I get to double revenues from these consumers? No, of course not. Because part of the revenue difference is due to income differences. So clearly this piece of analytics has no causal interpretation. But is it useless? No, it's not. For example, I might use these results to conclude that my most valuable customers are those who like to interact with me on multiple channels. If this is so, perhaps I should create a seamless experience for them regardless of the channel they want to use. Pricing should be the same, promotions should apply online and offline at the same time. Consumers should be able to return online purchases in stores and so on. Not causal but useful. In summary the checklist prevents you from making incorrect conclusions about cause and effect. But remember that analytics need not always be about cause and effect to be useful. [MUSIC]