The phrase Predictive Analytics is to a large extent self-explanatory; use analytics to predict future behavior. For what purpose and to what end? To anticipate customer needs or behavioral changes with the objective of gaining new customers or retaining existing ones.
The starting point of using Predictive Analytics is data; more specifically the quality and completeness of your data. Quantity does not necessary equal quality. Being able to capture pertinent data is always a challenge. Good data Matters. The old adage of GIGO, Garbage In Garbage Out still applies.
The question then becomes, “Where do we look?” or more specifically “What do we look for?”
Using Unsupervised Learning, which is a branch of Machine Learning (ML), we can now find hidden patterns or obscure pieces of insight that we can leverage to our advantage. Imagine feeding the algorithm a large data set with over 100 parameters on your existing customer base. This data may include:
- Historical Sales: Items purchased, timing, complementary purchases, quantity, amount, frequency, recency,…etc.
- Web Analytics: number of website visits, page views, which pages were viewed, time on a page, mouse hover, mouse hover length, images viewed, videos watch, duration, shopping cart abandonment,…etc.
- Firmographics: Number of employees, years in business, annual revenue (if publicly traded), location, number of offices, acquisitions,…etc.
- Demographics: Age, gender, marital status, number of kids, home owner or renter, income range, zip code, etc.
From this large data set, the ML algorithm is able to find patterns or tendencies that might provide some insight that a company can execute on. These patterns are often found in the form of clusters (i.e., customers who exhibit similar tendencies). Once a cluster is identified, a marketing plan to keep or capture can be implemented.
For example, let’s suppose you sell Insurance products. The data indicates that couples who recently acquired a pet are highly likely to want switch insurance providers.
Or, take for example, an Internet Service Provider (ISP) company who discovers that families who are expecting a child are likely to want to increase their bandwidth limit.
These are a correlations that would be nearly impossible for a person to detect on their own. In both scenarios, numerous variables were analyzed to find hidden relationships that can then be used by marketing to formulate some type of special offer and reach out to these two clusters (i.e., new pet owners and new parents).
In the case of the new pet owners, the existing insurance company can reach out with a special offer to prevent the couple from going to a new provider. Or, if a competitor gains access to this insight, they can target new pet owners with a special promotion to woo them over to them.
In the case of the ISP, they can target soon to be parents with a special offer on upgrading their existing bandwidth.
And, in both cases the promotional campaigns will be more focused and more cost effective.
Using ML algorithms to find hidden patterns is like staring at a painting that has a hidden message or messages embedded that can only be found through careful analysis and attentiveness. As the amount of data we collect on our customers continues to grow, so too does the challenge of finding these hidden message or clusters. The good news is that with advancement in processing power, calculations that would’ve taken 2 months can now be done in 2 hours without any supervision.
Victor Antonio, Co-Author of "Sales Ex Machina: How Artificial Intelligence is Changing the World of Selling"