Predictive analytics improves email marketing results. That’s hard to dispute, but it’s not quite right. Relevant messaging and attractive offers improve email marketing results. Predictive analytics has the “potential” to help.
It’s only potential because there is so much involved in doing it right.
Most predictive technologies are based on the principle of finding customer attributes or behaviors that predict outcomes. For example, what is it about a customer that seems to indicate a likely interest in buying Product A? For this approach to work best, you want to gather and evaluate as many customer attributes and behaviors as possible.
To do this, your data scientists will start asking for information. It may start as a little bit, but may grow in the quest for greater accuracy, including such data as demographics, website click streams, social media interactions, etc. There’s always more data that “could” be useful. That’s the promise of Big Data, after all.
Next thing you know, your predictive analytics project is being fed by a much larger data sourcing, acquisition, and ongoing maintenance project involving lots of resources.
When you finally have a complete set of data, it’s time to start the predictive modeling, usually the rarified domain of
statisticians and analytics specialists – today’s data scientists. The job is to derive formulas and rules that “score” or group customers, based on their attributes, against
your objective, those interested in Product A, for example. There are various techniques, most involving a lot of testing to verify and refine accuracy. Then if you also want to know who might be
interested in Product B, it’s a whole new exercise (if you have hundreds or thousands of products, you’ll have to decide which ones are worth this effort).
What if you can’t afford this time or cost, just to get going?
Fortunately, there are technologies that squeeze the most predictive power from relatively little information. It is possible to accurately predict the products each of your customers is most likely interested to buy from just a few, readily available fields of transaction data. This approach makes its predictions from a much narrower range of information but requires lots transaction history. The advantages include automating the analytics and eliminating the need for expensive data preparation projects because the few data fields needed, can easily be extracted from a single database.
But what if you don’t have the necessary skills or cannot afford all the software?
Again, you might be in luck. By constraining the problem domain to specific objectives, a new breed of solutions automates the analytics. No one has to do any data modeling to predict the products every customer is most interested to buy.
Ultimately, working with data is tough. It has to be interpreted and its imperfections accommodated. In the end, a predictive modeling exercise produces “insights” – insights about who your customers are, their intent, or what they might be interested in, usually associated with a score or probability. It’s useful stuff, but you still have to translate these insights into action.
One thing that helps is to define and refine a complete process for very specific, actionable purposes. A process or service purposed to deliver an online recommendation engine or generate mailing lists with personalized email offers leaves little doubt about how to leverage the results from your predictive analytics. That makes it accessible to a lot more marketers.