Attribution Modeling

Attribution Modeling

Published B&T – July 2016

One of the three main industry publications recently published 10 key lessons to come out of a “high level conversations with industry professionals” on attribution (one of the new buzz terms).

In principal it is bedrock information that every advertiser would want to know – as the old saying goes “I know that half of my advertising works, I just don’t know which half.” To be able to apply some form of accurate quantitative measure to each of the wide range of communications options available today, with the aim of optimising the millions spent, is an elusive question that has dogged the industry since advertising first began.

Attribution is not new. TW Anderson published the first widely used text book “An Introduction to Multi-Variate Statistical Analysis” in 1957, though the maths it is based on was developed in the early 1800s. In simplistic terms, attribution (some form of multi-variate analysis, or more precisely, Factor Analysis may be more appropriate) is used in science, medicine and more recently in business, to try quantify the influence each variable has on a particular outcome. It is difficult to give “folksy analogies”, but imagine looking at the how rainfall influences a plant’s growth. Too little or too much will kill it. But rainfall is not the only variable that impacts on the plant’s growth. Drainage, sunlight, temperature, different nutrients – there are a huge number of variables that influence the ultimate outcome. The goal is to quantify what percentage each variable contributes to an optimal outcome and identify which variables contribute the most.

But plant growth is very simple compared to human behaviour. (A generalisation. We have all met people who make plants seem intelligent and well-adjusted). Not only are there so many more variables at play, the variables and the impact of each is different for each individual.

I started my career as a mathematician (BSc Pure and Statistical Maths), initially consulting to engineers and scientists conducting research studies, before joining a market research company in 1980. We were doing attribution modelling back then, though the available data limited its effectiveness.

There are two aspects to attribution:

  1. Having a working understanding of why it is being done and most importantly how to use the results.

  2. Doing the calculations.

It was interesting to read some of the quotes from the participants. They ranged from wise, experienced commentary “….external factors such as climate, economy, sales offers, cannot be take into account within an attribution model. The best head space would be to understand that attribution isn’t a silver bullet to understand how channel performs, but a tool to adapt and improve to hopefully help in delivering insights into the impact of our marketing efforts.”

He is spot on. There are so many variables at play it is literally impossible to develop a “one size fits all model”. As well as those he mentioned, think of all the factors that can influence a purchase. The “first click/last click” model is useless at best; it assumes that the purchaser is operating in a vacuum. You don’t need to be a mathematician to understand this – experience is the currency that matters.

Then there were comments such as “This is a specific skill set with its own language. People talk about rim weighted, linear regression, normalise models and a number of other mathematical models that don’t make sense to clients. This is a skill set they don’t teach at university.”

He is correct. They don’t teach linear regression at university. It is elementary statistics, now taught in high school. You don’t just pick up these skill sets. But to understand and use the results, you don’t have to.

The most important lesson of all is that unless you have a mathematician employed at your agency, outsource the attribution modelling. Statistics is applied mathematics and if you have not studied pure maths disciplines such as multi-dimensional calculus and linear algebra (no, this is not algebra about straight lines), then you don’t have a hope in hell of understanding multivariate analysis, let alone trying to develop your own modelling. It takes years of specialised study. And it is not a matter of using some algorithm and popping in the data. A computer only does as it is told (programmed). Firstly, you must develop the equation (and a different one would be required for each scenario), then you have to find a programmer who can write code for complex mathematical analysis.

This is not being derogatory to the people at this summit, but I would bet London to a brick that none are qualified to conduct attribution modelling. I would not expect them to be. Insurance professionals do not have to be actuaries. It is a highly specialised area. The actuary calculates and prices risk, but this is not the whole of the insurance business. Investment banks engage specialised experts to develop investment products.

It is only in advertising that we feel we have the skill sets to do this in-house. Which is why management consulting firms, investment banks and large accounting firms are moving into our areas of business and starting to eat our lunch.

Advertising has no official professional body that sets minimum standards such as The Institute of Chartered Accountants, The AMA or The Bar. There are plenty of webinars/seminars, articles, papers etc. but they are not objective and peer reviewed. Their purpose is to sell something, as opposed to educate without any bias.

We find ourselves in the situation of having 3 major industry magazines, attempting to fill this gaping hole. Commercial operations, beholden to the industry for their very existence. It is not their fault, but they are caught is a massive conflict of interest. Unlike official publications of professional bodies, they are not going to write anything critical.

Agencies are fragmenting into so many diverse areas, the majority of which are not core to their original charter. If we want to remain relevant as an industry, we need to stick to our competency of advertising/communications. Each day you can read the business media and see reports done by PwC, KPMG, UBS, Macquarie Bank on topics that should be our bread and butter.

Attribution modelling is just one example. We are not geared to compete with the McKinseys of this world to conduct these studies. But we know a shit load more than they do about what the results mean and how to implement the findings.

We have been made to not only look foolish, but also potentially corrupt for using “Views” as a metric for online video. Bloody hell, any fool could see the massive holes in the logic of this metric. But media companies are now separate from the creative and online pays 2 to 3 times as much as traditional, so why rock the boat? Media companies still primarily earn their money from commissions. To the boards of directors who are used to dealing with the companies mentioned, that is seen as being a broker. Providers of objective advice do not take commissions. Maybe it is time to look at different remuneration structures and to bring creative and media back together again. After all, Advertising is Media + Message. They should not be discrete entities who compete with each other. They should be a package where the sum of the parts is greater than the whole.

If advertising agencies want to exist in 10 or 15 years, they need to get back to what they do best – not just create ideas, but create ideas for communications/advertising that ultimately plays a significant role in generating profitable sales. (Any idiot can write a headline like “50% OFF”) the fragmented media landscape presents us with far more opportunities than ever before, but instead we are running off trying to develop new services/products that are not our core business.

And of course, there are the purchasing departments of clients who have been cutting budgets for so long now and you may just give them another reason to cut them even further.

DIVERSITY – Driven by Ignorance, Resulting in Division

DIVERSITY – Driven by Ignorance, Resulting in Division

What the Hell is Content?

What the Hell is Content?