The following is a summary of the first keynote of the 2013 MRA Corporate Research Conference in Dallas.
Title: Predictive Analytics for Marketing Researchers: Learning from Data to Predict Consumer Behavior
Speaker: Eric Siegel, Ph.D, Founder Predictive Analytics World and Text Analytics World
Q: How do you put credibility on being able to predict the future.
Predictive Analytics:
- Nate Silver: predicted outcome correctly in each state, but this is not predictive analytics but forecasting for each state.
- Obama campaign analytics team used predictive analytics to make predictions for each voter — used persuasion modelling, to see if the vote preferences of individual voters would be changed by actions such as phoning them or knocking on their door.
Definition of predictive analytics: “Technology that learns from experience (Data) to predict the future behavior of individuals.”
In order to drive better decisions a predictive model is used to determine an individual’s likelihood score to do something. The higher the number the more likely.
Predictive analytics does not have the precision of traditional market research, but it can be profitable to use it.
Example:
- If sending a direct mail piece to 1,000,000, assuming 1% response rate, $220 revenue who responds and a $2 cost for each mailing – profit $200,000
- Targeting by use of a model may mean that a company could reduce mail out to 250,000 that are more likely to respond. In a hypothetical case if 3% respond the profit would jump to $1,150,000 because of saved marketing costs.
The big task of predictive analytics is getting all the information into a database (data preparation)
Examples:
- An online data site getting a user from moving from a free membership to a paid account can be predicted by email address: EarthLink, SBC, Comcast highest, Yahoo & Hotmail lowest.
- A website ran a test of seeing whether ad click-throughs would change if individuals were targeted in what ads came on-screen. There was a 25% increase in click-through rates, leading to a 3.6-5% revenue increase for the site, translating to an extra $1.5 million revenue every 19 months.
Things to keep in mind:
- Most organizations don’t need to buy third-party data, they usually have enough themselves
- Causality and correlations don’t always hold together: when more ice cream is sold more shark attacks — why? likely increase of sales a sign of better weather — people more likely to be outside and on beach
Three places were companies often use predictive analytics:
- Response modeling for acquisition
- Churn modeling for retention — target customers who are about to leave , give them a good deal (heavily subsidized phone etc) cost to profit, so cannot offer to everyone. Decision tree created, with customized variables a PhD tool (push here dummy).
Point of predictive models is the actionability, and sometimes knowing what not to do:
- Offering a free phone to renew a contract for a potential churner may not be a good idea, because it may remind them they no longer are under contract “Sleeping dog customers” (don’t wake them up).
- Obama team determined there are individual voters if they knock on the door they would be more likely to vote for the other voter. Did polls of people who were visited by campaign and see if they had shifted their vote, applied results across all voters across the country.
Organizations use predictive analytics to predict who to:
- advertise, recommend or discount
- loan or advise
- assist or educate
- investigate or incarcerate
3) Which recommendations, ads, content to show
Ties with social media
Predicting based on who you know:
- Elements such as buying behavior and political views likely to be quite similar.
- A major national telecom company found that people are 700% more likely to cancel service with a specific mobile company if a person in their network does.
Predicting based on social activity:
- Banks are looking at using type of social activity to help them in decision-making process for credit products.
- IRS using this to determine who to audit.
- Efforts to look at anxiety level on social media to predict outcome of stock market.
Predicting based on what you like:
- Only 1/4 of millions of posts need to be scanned to determine what goes live.
Take-aways
- Leverage big data and learn from it to predict
- Predictions for the individual (customer, voter)
- Significant differences on predictions fortifies and bolsters all kinds of organizational operation
