Wednesday, December 19, 2012

big data bollocks

A couple of things I've learned from speaking at ad industry events over the past few years are these.

The first thing is to try and make your 15-20 mins as entertaining as you can. A story to back up your points is more important than graphs and charts, and the stories will be the thing that audience members will take away, more often than not.

The second thing is to be mindful that most events will have a hashtag connected via which delegates will tweet the bits and pieces that resonate.

A controversial or otherwise interesting 'blanket' statement about this or that will often get tweeted so it's always a good idea to structure a few of your points to be tweetable.

Also I've found that it is unlikely you will have the whole crowd nodding in agreement with you and quite often there will be significant disagreement. Don't worry about that, trying to appeal to every point of view inevitably ends up in appealing to no-one.

To that last point, I've had a few bits of feedback from delegates at last weeks AIMIA Future of Digital bash.

In my final section I proposed that 2013 may be the year in which we see the bubble burst in the whole big data situation.

There was equal parts agreement and dismay among those present.

The thought was thus; the value of big data is vastly overrated.

This is not to say that there is no value but rather that the value is derived from the processing and analysis of said data and it's conversion into important information.

For those familiar with the DIKW model, that information requires further distillation in order to come out the other end as Wisdom.

In adland parlance we would call wisdom 'insight'.

The data in itself may indeed be the new 'oil' however it is crude oil at best.

The other thing is that more data does not necessarily mean better.

In fact the more data one has will often make it harder to find the patterns that become the required information to distill into insight.

Better means better.

Having dealt with many businesses over the years who cannot even make sense of their own opted-in customer database, more data is not going to help them in any shape or form.

The other point is that any data set needs human beings to interpret it.

And knowing how as humans we are subject to no small amount of foibles and biases is testament to the difficulty of this task.

To illustrate I quoted this oft repeated psychology experiment (to add, we have conducted one of these ourselves and achieved remarkably similar results to those experiments of a similar nature from academia).

We asked two groups of financial services employees to assess their likelyhood to approve a credit card application from a recent graduate.

The applicant had creditworthy history, and was gainfully employed with an above average salary etc.

However, with the first group we gave them one extra data point to consider.

The applicant had an outstanding student loan of circa $5000.

With the second group we gave two data points.

The applicant had an outstanding student loan of between $5000 and $12500.

This second group were given an extra option in there assessment process. Either approve or decline the application. Or await further information about the extent of the outstanding debt.

Not surprisingly the majority of group two asked for further information.

We then revealed that the debt was actually very close to the $5000 number.

In group one around 70% declined the application for credit.

In group two only around 30% declined.

This is despite both groups having nearly identical data in the end.

By firstly anchoring group two on the $12500 number, the $5000 debt didn't feel so bad.

The point being that humans have clear difficulty with making consistent assessments when faced with only two pieces of data.

Good data has long been the lifeblood of marketing (ask any direct marketer) but at this point in time perhaps we don't necessarily need more data but better data, and there's a criminal shortage in the advertising industry today of the actual human skills needed to interpret, distill and convert into insight.

blog comments powered by Disqus