Tuesday, October 03, 2006

Yes, No,or Something In Between

When is an average wrong?

Most events marketing is concerned about are binary -- that is they happen or they don't. Buy, cancel, respond, churn, request, register, and click are all examples of a behavior that either occurs or it doesn't. Given that it is probably impossible and certainly impractical to categorize people into one of the two camps for future campaigns we've come up with some ways of estimating the probability that a customer or prospect will do something. Now we have values that look continuous from 0 to 1.

However, there are risks in interpretation when doing numeric gymnastics.

When reporting this likelihood with an assigned value to it, e.g. estimated sales revenue, strange numbers can occur. For instance, consider an email campaign where responders will spend $100 and non-responders will spend $0. The average may be $50 but the expected value isn't that at all if conversion rates are anywhere close to typical. The likely contribution for a 1,000 mailing with a 2% conversion rate is $2,000 (1,000 * 2% * $100) not $50,000 (1,000 * $50).

And yes, these simple errors are made. And they do impact decision making.

A widely publicized example comes from government circles. In 2000, the USGS published a tome about the status of known oil fields and the likelihood of finding more. Using the math similar to the email campaign it was concluded that there is a lot of oil to be found in Greenland.
  • 5% chance of finding 112 billion barrels
  • 95% chance of finding 1 barrel
  • Average reserve is reported as over 47 billion barrels

With that figure, which is twice the size of the estimated reserves of the US before the recent find in the Gulf, a lot of attention has been paid to an area that has produced little to date.

Using the email logic, the expected value should probably be somewhere in the 5 billion range, or the size of the United Kingdom. Or less than a year's supply.

No comments: