What is predictive analytics?
“Predictive analytics is information technology that produces a predictive score for each customer or other organizational element. Assigning these predictive scores is the job of a predictive model which has, in turn, been trained over your data, learning from the experience of your organization”. Predictive Analytics World (follow them on LinkedIn)
Predictive analytics for marketing
This blog post discusses the efficacy and efficiency of using predictive modeling as part of a marketing program to segment a database for the purposes of marketing directly to a targeted group of individuals.
The process begins with collecting and integrating data from various sources, combining it with your marketing and customer data, and using it to build a predictive model to make predictions such as which marketing actions have a higher chance of succeeding and which do not. These insights can then be used to inform strategic planning and optimize content, campaigns, or promotions.
Let’s use a historical health club database example to clarify the process. In this example, there are eight predictor variables (perceived value rating, satisfaction rating, service quality rating, age, weight, length of membership, frequency of attendance, and participation in programs) and a response variable (renewed membership, did not renew membership). A statistical profile of the renewals is generated and compared against a non-renewal profile. Profiling is used to better understand the characteristics and attributes of the membership. However, these profiles are limited in predicting who will/will not renew their memberships – as every variable is treated with equal weight.
Predictive Modeling vs Profiling
In order to make predictions about who will/will not renew their memberships, we need to know the relative importance of each variable. Hence, we would develop a predictive model – as a predictive model is multi-dimensional, with every variable weighted differently in importance. Predictive modeling can be used to determine which of these variables are statistically significant and which carry the most weight in terms of predicting the response variable. A predictive score can be calculated and members can be segmented into high-scoring members (primarily renewals) and low-scoring members (primarily non-renewals). The final model is an algorithm and can be used to score the member database.
Predict Response Behavior
Each variable (individual characteristic) is evaluated to determine which are predictors of response behavior. Each is then statistically weighted either positively or negatively. Each member is then screened for the statistically significant variables and the weights are assigned accordingly. All weights are summed up to generate a total score for each member. Low-scoring members have either negative-scored attributes or little-to-no positive-scored attributes for renewing their membership. In contrast, high-scoring members have many positively-scored attributes for renewing. As total scores increase, the probability of renewing increases. For example, the results may show that only 1 in 10 subscribers (10%) with a score of 19 or less are expected to renew, while 3 of 4 subscribers (75%) with a score of 79 or higher are expected to renew.
In this scenario I would suggest that more marketing efforts be focused on low-scoring members because they are less likely to renew (their membership) than are high-scoring members. The renewal/non-renew model scores can be used to decide who to send renewal notices to. Since high-scoring members are likely to renew, it may be more cost-efficient to reduce the number of renewal notices to members in the the top-scoring ranges and instead send those notices to members in the lowest-scoring ranges. The number of notices sent to members in the middle score ranges can remain unchanged.
Run Smarter Marketing Automation
Predictive marketing technology can also used to run smarter marketing automation by triggering your sales and marketing platforms based on past member behavior. For example, predictions about the probability of a particular member to renew their club membership can be imported into your marketing automation tool. The tool would then trigger follow-up emails to members that are less likely to respond to your membership renewal promotion.
A Chiropractic Example: Targeting those most likely to respond
Suppose you want to run a new patient special that includes an exam, consultation, and X Rays. You could send an email to everyone in your database. However, you would risk increasing your unsubscribe rate as many recipients might see your emails as irrelevant or untimely.
A more efficient and effective option is lead-generation modeling. This predictive modeling technique targets prospective patients that “look-like” your profitable patients. A predictive analytics tool is used to analyze all historical (and preferably real-time) data to model the significant characteristics of your current and past patients. This model is then applied to your current database to filter out the prospects with a high likelihood of responding to your discount promotion. You can then target filtered leads with a discount promotion that is relevant and timely. Note that in this case we are targeting those prospects that are more likely to respond to our promotion.
Other benefits of predictive marketing technology
Predictive modeling systems are highly efficient for new customer acquisition or cross-selling to your current customer base. The technology can help:
- Save marketing dollars and improve marketing ROI
- Increase revenues
- Predict buying propensity and prioritize leads
- Categorize your customers’ and speculate about their needs
- Know your customers’ wish lists
- Guess your customers’ next actions
- Categorize your customers as loyal, seasonal, or wandering
- Guide segmentation and persona building to facilitate personalized communication
- Increase alignment between sales and marketing
- Identify customers at risk of churn
- Increase customer response rates
- Improve conversion and click-through rates
- Uncover cross-sell and up-sell opportunities
- Empower your sales team with intelligence regarding which accounts to pursue and which not to pursue
- Expose gaps in digital content
- Choose the right marketing mix
- Support a product bundling strategy
Lastly, building a predictive model is an ongoing process – not a set-it-and-forget-it solution. Current market conditions demand that you update your model periodically (with new customer information) to counteract the model’s tendency to to decay over time. Preserve your competitive edge and keep the process relevant by updating your model periodically.
To learn more about how predictive marketing can dramatically improve your business or practice contact us for a free consultation.