Marketing technology has become so efficient at automating marketing campaigns that it’s backfiring. Customers are less and less receptive now. Most e-commerce companies know they send too many emails and other communications to customers that are ignored and increasingly unwelcome. The resulting “annoyance factor” is raising the barrier to communication. The way forward now is to get smarter by being more relevant, targeted, and personalized. For that, marketers need predictive customer intelligence.
Predictive customer intelligence is a set of indicators, usually in the form of scores or metrics, on the likelihood of a customer taking certain actions in the future. Accuracy of predictions at the individual customer level translates directly into effectiveness. So what are the most useful metrics for getting more revenue from marketing campaigns?
1. Timing Scores
Predicting when a customer will be ready to buy allows marketers to target customers with offers at the right time. Why do that if they are already expected to buy? Remember, a predictive analysis does not predict the future. Nothing can do that. It is not whether a specific customer is, or is not, going to buy. The “prediction” should be represent the probability of each customer purchasing within a certain timeframe, usually based on the active window for expected responses to a campaign. Except in certain circumstances, the actual probability of any given customer purchasing is low, because most customers usually will not buy during the campaign window.
Among customers with a 10% chance of buying, 9 out of 10 are not expected to buy. This is how you validate the accuracy of your predictions: Among the customers with a 10% chance of buying, how many actually buy?
If your predictions are accurate, a group of customers with a 10% chance of buying will buy at a rate that is more than five times that of a group of customers with a 2% chance of buying. If a group of 10% customers are the most likely buyers, relatively speaking, in a given timeframe, it’s a good idea to target them with offers. There are two reasons for that.
First, since the chances are much greater they will not buy, you want to encourage them. Second, while the probabilities are low, there are 5 times more potentially interested customers in the 10% group than the 2% group, or a group of random customers. Offers targeted at the higher propensity group will be more effective at less cost than targeting everyone. However, another strategy often used is making less rich offers to higher-propensity customers than those in lower propensity brackets who need more incentive to buy now.
2. Product Scores
Knowing the probability of a customer buying specific products in a given timeframe can be extremely valuable for lifting revenue. Again, the actual probabilities are probably very low, but the products with the highest probability for each buyer become great product recommendations for that buyer, much better than offering “best sellers” to everyone. Consider this example:
In this real example, customers have been bucketed by probability to buy product #1517 into 10 ranked groups as predicted three months ago (Group 9-10 being the most likely to buy that product). Then, the customers who actually purchased product #1517 in the subsequent 3 months were counted. It turns out that 291 or 0.041% of all 706,521 customers actually did buy the product. The predictions were reasonably accurate in that most buyers were found: the analysis predicted 257 customers out of the 291 customers who actually bought to have meaningful probability to buy. Groups from highest to lowest propensity did in fact buy at a decreasing rate.One interesting part is that the scoring was able to identify about 134,000 customers with a more than 2x probability of buying versus the average customer. Promoting this particular product only to this group of customers would result in more buyers at lower cost and avoid annoying disinterested shoppers.
Another interesting development is that the scoring identified 2,887 customers with a propensity to buy this product 12.6x better than average. Even though most did not buy, and only 15 did, you can be sure there is much higher concentration of opportunity within this group overall. A well-executed promotion to this special group would undoubtedly have aroused more interest and pushed more fence-sitters to buy this product.
Going a bit further, if all available products can be scored such that you know the top 5, 10, or more unique products most likely to be purchased by each customer individually, you improve the odds of a response even further. This is because every customer receives the specific recommendations with the highest probability for that customer, increasing the odds 5x, 10x, or more above average. A personalized campaign that does that can really perform.
Finally, this example also shows how predictive customer intelligence is most effective at scale. Trickling offers out to one customer at a time, given that most customers will not respond, does not fully leverage the improved odds. But if you can reach 1,000 or 1 million customers in a campaign where the odds of conversion have been doubled from, say, the typical 1% to 2%, that will result in a significant lift in revenue. If you cast a die 1000 times, a one will come up roughly one-sixth of the time. In other words, if you accurately calculate the odds and use them enough times, the results will gravitate towards those odds.
3. Expected Spending
Metrics that anticipate the amount a given customer will spend in a certain timeframe can be used to improve profitability of campaigns, by making offers that encourage “stretch spending.” For example, a mailer or catalog that costs $1 to send to a customer can be limited to a group of customers expected to spend, on average, at least that. On the other hand, customers expected to spend $500 might qualify for incentives if they spend at least $600.
4. Risk Scores
Targeting customers who are at risk of not buying anymore, or churning, can improve retention rates. Risk in this case is the probability of a customer reaching the end of the “active period” without making any purchases. The active period is the length of time where, if customers go this long without buying, there is a 95% or 99% chance that they will not buy again. It may be too late for high-risk customers, but intervention tends to be time-sensitive. Targeting formerly good customers whose risk scores are increasing or targeting customers crossing certain thresholds with timely messages and incentives can mean the difference between saving and losing a customer forever.
5. Long-Term Value & Lifecycle Scores
The above scores can be used separately or in combination for short-term impact. However, each customer should also be viewed in the context of their long-term potential to be a valuable customer and the lifecycle stage they are in compared to other customers (e.g., first-time buyer, highly loyal, fading, etc.). This bigger picture, the long-term buying context, can and should influence decisions about how to treat customers for broader programs, such as loyalty programs, despite expected activity during near-term campaign timeframes.
There are lots of ways to lift revenue by using the right sort of predictive customer intelligence. If your goal is more revenue, it’s hard to translate customer personas or segments into that. Try looking for predictive metrics that relate directly to revenue, such as expected time of purchase, products to be bought, and amount of spend by each customer, as well as risk of not buying and the long-term buying context. There many ways to get this intelligence, though weighing accuracy and affordability for different needs will have to be the subject of another article.
Loyalty Builders offers cloud-based predictive analytics service designed to enable marketers to improve revenue through customer messaging.