Alex Steer

812 words | ~4 min

Last week I had the privilege of attending the IPA Effectiveness Awards, where we picked up a Silver for our work with Plusnet over the past five years. Our paper was called out in the judges' review for creating a culture of sustained effectiveness with an ongoing commitment to rigorous testing. The result is that for every pound spent on advertising (creative, media and agency fees), Plusnet have got over £4 back in new customer revenue.

For me it's a great example of effectiveness done right, by a team of true marketing scientists working alongside a great media planning team, a strong creative agency and a supportive client.

But it also provides a strong contrast to the way I see analytics done in a lot of marketing organisations these days.

A few years back, Harvard Business Review proclaimed data scientist to be the sexiest job of the decade. A quick job search reveals thousands of vacancies, many of them with wildly varying job specs. In fact, right now one of those vacancies is in my own team. By our definition, a data scientist is a statistical developer - someone who can turn data analysis methods into working, scalable code so they can be automated and run faster. This is a vital skill for any team that wants to be taken seriously in a world where the volume of available data to be processed, and the frequency of having to make sense of it, have both increased vastly.

But the rise of the data science star in marketing departments and agencies is not an unqualified good. It mirrors the emergence of a generation of mid-level to senior marketers many of whose training is almost entirely within either digital or direct response businesses. This is due in large part, I suspect, to the reductions in brand marketing spend and headcount in the years after the 2008 financial crash. There is a 'lost generation' of brand marketers and that cohort is now becoming relatively senior but without having had the training in many of brand marketing's craft skills, like designing a brand tracker, interpreting qual research or using econometric modelling.

The result is an assumption in many businesses that marketing analytics is largely just a data and software problem - a view often promoted by technology companies too, unsurprisingly. The result is that we as an industry have been hiring people who can do stats and write code hand-over-fist, calling them data scientists and assuming that they can figure out the complexities of predicting marketing performance.

It's an enormously dangerous tendency, the same one that the financial trading industry fell victim to a decade ago. The reason why is hard to explain if you don't do analytics for a living, but in a nutshell: pure data scientists, without marketing experience, tend to make bad assumptions about how marketing works. They bake these assumptions into their code and their models, and you end up with a badly skewed view of how your marketing works. It's very fast, the code is efficient, and it's fully automated, but it's wrong, like a driverless car accelerating smoothly towards a cliff edge. Worse, if you don't have the ability to check the assumptions (because you can't code, or don't understand the statistics), you have no way of knowing whether the assumptions in your ultrafast machine learning algorithms are brilliant or bogus.

Even when done manually rather than automated, bad assumptions can kill a brand. Often these assumptions are incredibly basic - for instance, I've seen a lot of marketing attribution models that assume that marketing only influences sales on the day (or even in the hour) that people see it, or that a marketing channel only works if it generates a direct response. This is the logic of early 2000s email marketing, being applied to large integrated brand marketing budgets with predictably hilarious but terrible results.

Data science - turning maths into code - is a vital skill, but it is not the whole game. It's time to start valuing marketing science - advanced analytical ability informed by practical experience of working with brands and advertising. It's time to start growing our own talent rather than hiring from outside; time to start training people in the harder disciplines of econometric modelling, research design and segmentation; time to recruit social scientists and economists as well as pure mathematicians and computer scientists; and time to be proud of the bits of what we do that aren't automated as well as the bits that are. Time, in other words, to insist that analytics knows a bit more about the complexities of how real people respond to brands.