Predictive analytics can leverage a variety of statistical tools including regression, probability models, and machine learning. But you don’t need a Ph.D. in statistics – or an expensive statistical package – to begin getting value from predictive analytics. Here are a few predictive techniques that can be carried out in Excel and can begin driving value immediately for the marketing organization.
In cases where there is limited historical information, businesses may choose to forecast the success of a new strategy – or the behavior of a new customer segment – based on “lookalikes” from their current dataset. For example, say a retailer just invested in a new affiliate network. They know these customers are likely to be different from their "average" historical customer - but using business judgment, they determine that this segment is likely to be similar to customers acquired through their paid social channel. Although this method still relies on historical metrics, it uses a predictive premise (“this new segment will be similar to another”) to arrive at more robust forecast than simply using the overall historical baseline.
If a company is looking to forecast the engagement (revenue, profit, or repeat purchases) of a given customer segment, it can extrapolate from limited data based on the evolution of past cohorts. (See our course on Cohort Analysis for more background.) For example, say that a cohort from a new acquisition channel spent, on average, $100 in their first two weeks. The retailer knows that spend in the first two weeks generally accounts for 50% of a cohort's two-year revenue. Based on these inputs, they predict that this cohort will spend $200 in its first two years. This technique enables the company to build a “forecasting curve” that acknowledges that all customer segments are different – but that they evolve in similar and predictable ways. Our Cohort Analysis course contains a handy Excel workbook that simulates this technique.
The pioneers of direct marketing realized that customer lifetime value is driven primarily by three dimensions: how recently a customer has transacted; how frequently that customer transacts; and the average monetary value (average order size) of that customer's purchases. Because these are the three core dimensions of lifetime value, customers with a similar RFM profile tend to be quite similar in terms of predicted spend. How can this help the marketing team make better predictions about the future spend of its newest customers? A marketer can divide the customer base into quantiles on each of these dimensions - for example, the top third, middle third, and bottom third based on transaction frequency - and find the historical CLV for each RFM "cell." For example, what's the average two-year spend of a customer who's in the top third by recency, the middle third by frequency, and the bottom third by average order size? By then situating new customers within each of these RFM cells based on their behavior to date, the marketer can come up with much more accurate CLV predictions.
You can try out your own simple Recency/Frequency/Monetary Value (RFM) Analysis with our RFM Spend Calculator worksheet that you can download here.