Urban Legends in Marketing Analytics: Predicting the Future


In this age of big data, big claims and big hopes about data-driven marketing, predictive customer intelligence plays a big role.  Despite all the attention it receives, there are some enduring misconceptions, or “urban legends,” about what predictive analytics can and cannot do to boost results from marketing campaigns.  Here is #1.

Myth #1: Predictive Analytics Predicts Behavior

Customer #1106 was predicted to buy Product A, but did not buy. Is there something wrong with the analytics??  Slow down — no person or system can predict the future.  The key is to understand that predictive analytics = probabilities or other measures of propensity. The goal is to improve your “batting average” by working the probabilities over time and at scale. If the chance of getting a six when you roll a die is 1/6, you probably will not see exactly one six if roll just 6 times, or exactly two if you roll 12 times, or three if you roll 18 times. But you will get closer and closer to that 1/6 hit rate the more times you roll.

To illustrate in marketing terms, if you recommend a product to a customer that has a 10% probability of purchasing that product, there is a far greater chance they will not buy it. However, if 100 customers have a collective 5% chance of buying your “best selling” product, but you can recommend the specific product to each one where they individually have a 10% chance, you have just improved your odds of getting a sale from each customer by 2x. This does not mean you will get 2x the revenue every time. Outcomes of specific campaigns may vary. But keep doing it over time and it will converge on the probabilities. It’s just math. It works.

Accurate Probabilities

The key is to get accurate probabilities. You can find out quickly how accurate your method is by doing a back test. Analyze only the data you had up to a point in time in the past, say 30 days ago. Then predict probabilities for what happened in that 30 days. Since you already have the purchase data for that last 30 days, you can see how accurate you were. Did all the people with a 10% probability to purchase in that last 30 days actually purchase at a 10% rate?

Leveraging the purchasing probabilities that exist among your customers to get even a little lift makes a big difference over time. For some of our retail and e-commerce customers, averaging a few pennies more in converted revenue per email can result in millions of additional revenue dollars per year. Sometimes campaign results are improved dramatically, other times only a little, but averaging 10% or even 20% over time is quite typical.

Create marketing programs that tap into and roll with the purchasing probabilities that exist in your customer base over time and at scale (hopefully reaching your whole customer base), and you’ll be using the natural power of predictive customer analytics to maximize orders from existing customers.

Is it Worth it?

One client story, an e-commerce retailer, shows how this works:

  • About 750,000 emails were sent to both target and control group in each of six campaigns over a six week period.
  • Target group received personalized product recommendations, control group did not.
  • Target group averaged $0.43 more spending per email (about 14% lift over control group).
  • Revenue lift of $314,700 for the period; tracking to $2.6 million for the year.

And that’s not out of the ordinary. Our clients know it’s worth it.  And that’s no myth.

Peter Moloney is CEO of Loyalty Builders, whose Marketing Lift Service offers a simple, cloud-based predictive analytics service enabling marketers to get revenue lift from more relevant communications to their customers. Request a Free Customer and Revenue Analysis here.

Tags :