Data Mining and Predictive Modeling: Andrew Grenville, Kevin Dang #Netgain7 #MRX


Netgain 7 MRIA
… Live blogging from downtown Toronto…

Predictive Modeling

Andrew Grenville, Chief Research Officer and Kevin Dang, Senior Research Manager, Vision Critical

Data Mining and Predictive Modeling to Drive Panel Management Strategies

  • Analytical panel management has key factors
    • data set-up
    • what business challenge are we trying to solve – e.g., valuable panelists, churn
    • methodology and analysis – which model, which variables, how to fine-tune, which statistic works better
    • evaluation and performance – is the model stable, can you move it through time
    • application – if we know who will churn, how can we act in time
  • static reports, drill down, ad hoc reports, forecasting, predictive modeling, optimization – researchers tend to stop before predictive modeling
  • Want fast turn around time, waiting a couple weeks for analysis isn’t going to work
  • Needs to incorporate longitudinal data, learn from the past
  • Needs to be flexible to incorporate both survey data and panel data
  • Needs to have as little ongoing IT support as possible
  • Identified all the interaction points along the panelist life cycle – disqualifies, over quota, incentives paid, other touch points
  • First cleaned out poor data, e.g., 115 year old people
  • Churn model – urban area or high income were more likely to churn, 18 to 30 more likely to churn, low response rates more likely to churn
  • Conducted exploratory and confirmatory analysis – predictive accuracy was 76% [awesome!]
  • Challenges with model include the quality of the data, lots of time was spent harmonizing the data for modeling
  • Challenge was making model useful and actionable – it had to be as simple as possible so that it was actionable

2 responses

  1. What traditional MR companies have to accept is that making money on the mark up of data collection costs is an old business model. The traditional MR business model generally had a lot of profit and revenue associated with data collection with a relatively small amount of the revenue associated with analytics or modeling. It is clear that MR can’t use this model anymore, or at least it can’t use it all the while. Analytics is the new frontier, whatever we think of the quality of data, we can get the data. What to do with it is the next question. I would suggest that in the long term MR professionals should forget about criticizing survey design and concentrate more on honing their analytical skills.

  2. Dang! Kevin! (pun intended) That was one cool predictive model and you’ve just made CHURN one of the words of the day!

%d bloggers like this: