Tuesday, October 07, 2008

How to Model Events

I've been thinking a lot lately about events, I guess because of the financial crisis. The events of the financial crisis could have been avoided if we'd been able to predict them better. And, now that the banks left standing are picking up the pieces, they need to start thinking about winning customers again. In banking, event detection is a poweful tool for cross-sell and customer retention.

Events are things that happen in time and are discrete. A bank collapsing. A baby being born. A president being elected. These are all events. Marketers are interested in events because they help us organize our communications around the event in question. What's nice about events is that before they happen, we sometimes can predict them with other data. And after events happen, we absolutely know they did happen. If it happened, we can tag it with a "1" and state the time it happened, who it happened to, etc.

"Event detection" marketing is pretty simple, in concept. A customer defaults on a mortage. We send them a phone call telling them to pay or else. A customer opens a new credit card account. A customer care rep calls them and thanks them and tries to upsell them on credit protection. This is what I call "reactive" event detection.

But what about "predictive" event detection? A lot of events are big, life changing things that, if we really could predict them, would give us a huge advantage. An example of this is marriage. Wouldn't it be more effective to predict a marriage 3 months ahead of schedule than to find out about it when an individual checking account is created? We could produce an integrated campaign that would get the newly married couple set up with joint checking, life insurance, and a financial consult.

I think most of the growth in marketing is going to come in this area. Using customer insight to predict events on the basis of past events, and "marketing to the prediction". To do this, though, we need much better historical data. We need to understand past events that we "can market to" and mark them in time. How would a bank collect this data? Through branch tellers in a CRM system... through their online site... but importantly, it requires a long-term view. If we have a year or so of this data, we can start to model these discrete events and understand how account behavior predicts them.

B2B marketers can do this too. Data in a B2B environment mimic those available to a customer-centric B2C marketer, in many ways. There is generally very rich contextual data around accounts. CRM systems, if used, can integrate powerful customer service data, inbound call histories, and account maps for use as independent variables. I guess what I'll finish with is the importance of "starting from the end." Don't think about the independent variables--think about what you're going to offer a customer when something happens, and work backwards from there.

2 comments:

Anonymous said...

There has been quite a bit of work on tracking events and "triggers" for marketers but it's going to be exciting to see how predictive event detection will evolve.

LEADSExplorer said...

A first approach is by integrating the website analytic system with the CRM.
A B2B web analytic system provides the company name of website visitors. The visit data by company can be used within the CRM for predicting their status or decision pattern.