The past few years have shown us a shift in the dynamic of marketing led by the strengthening of television. We have also seen the rise of digital and as an overall pie the market for media has grown and hence print has also grown with it.
But something else has also happened; with all this growth we have also noticed the reduction of unbiased performance results of particular mediums, and specific vehicles and channels. This has been driven by the fact that over the last few years, the specific channel is now the one telling you whether it has worked or not, within metrics that they themselves have defined and using methodologies that they have governance over.
To give you a small case, when you advertise on either Facebook or YouTube you are essentially advertising inside a black box while having to rely on metrics that they have given, and these metrics are more with respect to reach, frequency and engagement. What these metrics fail to capture is the effect on sales.
Similarly, the challenge with TV measurement is well documented, despite the best efforts of BARC, with its noble intentions and a multifarious panel of participants, including Corporates, agencies as well as media companies.
Even with the limited reporting, the challenge still remains that television reports’ measurements of reach and frequency do not reflect the impact or the outcome on sales and fundamentally what marketers want is to be able to decipher the impact of their marketing activities on revenues. So while we look at reach, visibility, and frequency numbers, the true test or question being asked is, did marketing really create an impact on sales?
Now, this article is not to necessarily challenge the governing bodies or suggest better frameworks for these bodies and I’d like to iterate that I am not an expert on the subject. However, what I do understand is that with clouded black-box measurement mechanisms or rather, measurement mechanisms that are too limited in terms of their sample size, large irregularities and errors can be committed when forecasting the impact of marketing on sales, especially using the current traditional methods.
Fundamentally marketing is a growth driver. Marketing generates great revenue, margins and cash flow. It is the combination of an art and science that has been around for the last 400 years and has stood the test of time because of its ability to relentlessly prove itself. It is time; I believe that marketing takes a step up in terms of its own credibility and its ability to prove itself. For this, I propose that all companies move to a framework of reporting where while we look at reach numbers and impression numbers and try to decipher trends on that, we also move towards more mathematically proven models that showcase how marketing affects sales. Marketing needs to speak a financial language in the boardroom and apart from this, marketing also needs to be able to truly mathematically quantify its output.
The ability to quantify effects mathematically has been around for a few years now and while it’s a relatively new concept to the marketing world, the methodology used to quantify marketing’s impact on sales, that is the methodology of deep regression has been around for eons when it comes to human existence. It has been used to forecast climate change, for food distribution, and in financial markets extensively and only found its way into marketing relatively recently.
Regression Analysis was available to humankind in 1805, gained prominence in 1809 and was first used in marketing in 1949 so a good hundred and fifty years after the establishment of the science. To put it simply, what multivariate regression does is that it recreates the past so you have a dependent variable and multiple independent variables, that is in layman terms, you have an outcome and you have multiple activities that impact that outcome. So if sales is one outcome then all the various marketing inputs are the activities that define the outcome. Essentially what the mathematical equation does is recreate the past and redraw the line towards the past with the least possible deviation towards error. And in doing so, it delivers multiple equations while redrawing that past and gives you the most accurate outcome which has the smallest error term. This is precisely why it becomes, in a sense, the most unbiased impact of marketing on sales. It is not governed by a body, nor any organization which has vested interests, and not the media provider who is offering the data that delivers its outcomes. It is a science which is more than two hundred years old and been used in various fields by humanity.
Now the reason Multi Variate regression in marketing did not really make a significant dent across organizations historically is because of its inability to deliver speed and its dependency on highly sophisticated academic talent. Any good Marketing Mix Model takes anywhere between 12-18 weeks to deliver first outcomes and is usually done by PhDs in statistics. On last count on LinkedIn there were about 300 PhDs in Regression modelling and 3000 PhDs globally so clearly it’s a very small talent pool which makes the modelling fairly expensive.
Additionally, in today’s VUCA environment where marketing investments are shifted fairly quickly and new mediums evolve very swiftly; the infusion of new data creates the need to construct a new model and therefore also ensures that you have to recalculate the model and further wait for another 4 months before new cuts or runs come in. And what marketers are saying today is that that is too late for them to make agile decisions.
But thankfully that is a thing of the past because with the increase of the computing power of machines and with a focus on areas where machine assisted learning has gained prominence, MMM can now be done by computers with fairly high accuracies, almost as accurate as a PhD level candidate. Also, the speed of that modelling with a computer takes a matter of minutes vs. four months that a human usually takes to calculate this.
So this combination of man and machine can actually deliver significant returns to companies over the long run because companies can essentially use this mathematically led deep regression led MMM to be able to decipher really which elements of the marketing mix are having an impact on sales.
Essentially any good MMM model gives you a lot of data but fundamentally it focuses on giving you the multiplier which is 1 Rupee in on a marketing activity is how many Rupees out? The time to impact which is how long does the 1 Rupee take to deliver its full 100% return. The volatility – so when 1 Rupee in gives 17 Rupees out, what’s the volatility or the accuracy of that 17 Rupees coming in every time? Is it sometimes 12, sometimes 24, or is it always hovering between the 16, 17, or 18 Rupee mark?
And these points of data are highly valuable to a marketing manager and a media manager because depending on the climate they are in, these decisions help them take wiser investment decisions with their money.
To summarize I am making two large points. Firstly, mathematics need to find a stronger place in marketing because mathematics is significantly more unbiased than the human mind and it will help companies make wiser decisions without getting clouded with the politics of decision making that often moves money in media as a whole.
The second is that we need to move towards unbiased measurement mechanisms. So it will suit companies who market really well and rely on measurement mechanisms that have a fiduciary responsibility towards and answer only to those who are deploying the money.
By Harshil Karia, Founder – Schbang