Focus on your questions

The starting point is being clear on what you want to model. This is about focusing on those questions which will help explain how customers currently behave, and that you can use in your campaigns. Examples I’d be looking at include:

• Which customers are most likely to respond?
• Which are most likely to become high value?
• Which are most likely to stop buying?
• Which are most likely to buy product x?

Once you’ve got your initial list, you can then look to build a model for each. It really can be that simple, particularly with many marketing technologies having ‘lightweight’ marketing-friendly modelling functionality included.

Just go for it

Despite this accessibility, too few marketers seem to have embraced modelling. So, what’s holding people back? I think it’s a confidence thing, with many marketers fearful of taking a risk. Just remember, we aren’t trying to build a model to help us to land on Mars – we are just trying to better predict who to target in our marketing activity. Our model is just helping to identify those who are likely to respond/buy/stay and those that aren’t. It won’t be exact; it doesn’t need to be. It’s all about using lots of data to increase our odds of targeting the right people.

Test and Learn

Still unsure? Well, that’s where testing comes in. You can generate your scores, create deciles all the way from your top 10% to your worst 10% and then run tests to how each decile performs. You’ll soon see how effective your model is at predicting behaviour, and how much extra ROI you could be delivering for your business.

Good now is better than perfect never

Some marketers also have a ‘paralysis by analysis’ tendency - they are keeping trying to refine a marketing model until it’s perfect. That might be appropriate for ‘strategic models’, but as marketers, all we are trying to do is improve our odds of targeting an audience. Your models are likely to perform better than your hand-cranked and tired current campaign selections anyway – don’t get sucked into seeking perfection. You will get a greater uplift in ROI by deploying a model for every campaign than you will be creating one supposedly ‘perfect’ model.

It really works

Still sceptical? Well, I’ll share with you what happened when I turned up-on site at one of our sports retail clients to talk about modelling. Within 10 minutes we’d built a model to predict likely ‘golf buyers’, and 10 minutes later we’d added a sample of these high scorers into a campaign alongside their current ‘golf’ buyers selection. We then clicked send, and just waited to see response roll in. When our client shared his response stats, it was the ‘high model scorers’ group who outperformed their regular selection by over 19%. This boost was hidden in the data all along, just waiting to be found.

To that retailer, surrounded by “too much” data to profitably make sense of it all, the answer was to turn that problem into a solution with modelling.