Attribution Part 2 - The History of Attribution

It is probably not a good idea to start out a post about the history of attribution modeling with a story about myself but when you are old enough to be part of history…

Early in my career I worked in the Customer and Marketing Information Center for a midsize regional bank in the US. One of my quarterly tasks was to compare the amount of revenue claimed by each of the marketing campaigns run during the quarter with the amount of revenue posted in our quarterly earnings report. On average each dollar of revenue was claimed by seven different marketing campaigns. These were the days of just two kinds of marketing, above the line and below the line or put another way, branding and direct.

Above the line (TV, radio, outdoor) would use post campaign market research to measure the recall of the advert. Based on the percentage of people  that recalled the ad they would attribute that percentage of new business to the campaign. Below the line was a little more accurate in that they knew who had received the DM piece or the outbound call and they could look at which of the target audience took up the offer. This was less than ideal but it was all we had at the campaign level.

We were a little more sophisticated when measuring the effectiveness of the overall marketing spend. By using a Marketing Mix Model (MMM), we were able to correlate sales data with changes in the marketing mix. This allowed us to look at all channels at the same time. To understand how changing the ratio of spend across the different channels impacted sales however, we had to run very controlled experiments and controlled experiments are not always compatible with hitting sales targets.

This MMM approach had some other drawbacks. First, it only looks at the marketing mix without any regard to the messaging or targeting. You could never really be sure what was driving the change; the mix or the message. Another problem was that channels with a small reach got caught up in the noise. The two biggest problems however were that MMM overweights the media towards the bottom of the funnel and it could take weeks or months to get the results in depending on how long the sales cycle was for the product on offer.

Attribution modelling has seemingly moved on since then. There are a lot more models but that has not necessarily translated into a better life for marketers. Arguably due to a spate of M&A in the attribution space in 2014, marketers have entered what Gartiner calls the trough of disillusionment in their Hype Cycle. Marketers report that the models they are using are too inaccurate, too hard to implement and use day to day, the vendors are too expensive for the value they deliver. In late 2018, the Mobile Marketing Association reported an average NPS score for attribution vendors at a shocking -29.

So if attribution modelling is so horrible, why do we even bother? Come back for part three of the series where we get to the bottom of why we need attribution.