Don Marti

Sat 13 Dec 2014 07:29:11 AM PST

Look who's beating the advertising business at the BS game.

I read Bob Hoffman's blog, and, fine, I have to agree that advertising has a certain amount of bullshit in it. But the sad news is that old-fashioned brand bullshit is losing out to web-scale Big Data bullshit. Seriously, ad people, you're getting beat by a bunch of computer programmers. That's weak. Our idea of bullshitting is stuff like Look at the the ROI to the company if you buy me a faster computer! We're just tech people, no formal training in any of this stuff. We shouldn't be able to out-bullshit anybody. But I guess that as soon as you throw TECHNOLOGY and STATISTICS into the mix, ad people are all, whatever you say!

Bwah ha ha.

How about a simple example of the kind of thing that gets through?

I'll start a used car lot, and hire a statistician. She stands around with a clipboard and watches the people who walk in. 20% of the people kick at least one tire. Out of the tire-kickers, 10% end up buying a car. Out of the rest of the people, only 1% end up buying a car. So, out of every 1000 visitors:

20: kick a tire and buy a car.

180: kick a tire and don't buy a car.

8: don't kick a tire, buy a car anyway.

792: neither kick a tire nor buy a car.

What do I do with this information besides sell 28 cars? Maybe, not much. But let's say I need to hire my nephew. So he comes in to work and starts handing a live rat to everyone who kicks a tire. Now, half of the people who get a rat just run away.

100: kick a tire, get a rat, run away.

10: kick a tire, get a rat, buy a car.

90: kick a tire, get a rat, don't run away but don't buy a car.

8: don't kick a tire, buy a car anyway.

792: neither kick a tire nor buy a car.

Now, are the rats a good idea? If you want to go by common sense, probably not. I'm selling 18 cars instead of 28. But let's say the nephew and the statistician work together to justify the rats. The statistican can do multi-touch attribution on car sales. How does that work?

Simply speaking, channels that appear more often in converting paths than to no-converting paths receive a higher weight, which in turn allows them to claim more conversion credits and thus revenue.

By multi-touch attribution, the rat plan is a huge win. There are 18 converting paths and there's a rat on 10 of them.

So, did I convince you that we should be handing out rats to more customers? Probably not. But use real-world messy data, dress it up with a few more graphs and some more mathematical-sounding language, and make the rats digital? Hell yeah.