Martech.org
says that even in uncertain times, AI-driven predictive analytics can help marketers see what opportunities are coming down the pike. So why isn't this technology being used by more CMOs? At his talk at this month's MarTech conference, TrustInsights.ai Chief Data Scientist Christopher Penn showed some ways any marketing team can predictively leverage their data. And he also shared some head-scratching.
|
"About two-thirds of CMOs said that they're managing the present, they're putting out fires right now, and only about a third are looking towards the future, even in a period of time where planning and contingencies are so important, during the global pandemic. People are just not doing it, and the reason for this is a lot of the tools that you use for predictive analytics are in the marketing toolbox, but companies haven't really adopted them. Why? Because people haven't figured out how to make use of these things, how to use them to save time and plan ahead."
|
Much of the way marketers use predictive analytics is customer-focused. They want to find customers who are more likely to buy. And in an effort to be more efficient, marketers will put their efforts and budget into engaging these customers that have more intent. Another time-saving method, however, focuses on the timing of an event or opportunity that marketers can jump on. This strategy, called time series forecasting, can also help marketers avoid a bump in the road and save themselves from a headache.
|
"Seasonality includes the things happening over given periods of time that are seasonal," Penn said. Then, there's influences according to what's happening in time, and cyclicality is that cycle, that rhythm. Our data should generally be seasonal and, generally, be cyclical. If you're in B2C, you know the time between early November and January 1, you're going to be working double shifts.There's seasonality to that...But, depending on the product, the reverse might be true. That's cyclicality. So you have seasonality and cyclicality in your data."
|
Penn says you also need to have a certain amount of data to teach the algorithms what has happened in the past, and enough so that you can test your assumptions about the data.
|
"Test to make sure that your predictions are accurate, and then you can build your forecast. If you don't have enough data, predictive analytics tends not to work. Why is that the case? Well, think about it like this. If you were baking cakes right and you've only baked a cake once, you don't really know what can go wrong. And there may not have been any anomalies that day when you made that cake, so you have a very limited number of examples to say, OK, I'm pretty sure I know how to bake the cake. You'll know after a period of time what's going wrong. That's what you need with predictive analytics data. You have to have enough data to spot those times when something's gone wrong and account for it."
|
