From the marketing mix to the modeling of marketing effects

This is bad news for the many ad technology providers that rely on cookie-based data, but good news for the marketing industry. After ten years of constant hype, attribution modeling, including multi-touch attribution, has never proven to be the surefire way to increase the marketing effectiveness that it promises to be. It was this growing awareness that rolled the stone in the first place. The tightening of GDPR cookie consent regulations and Google’s announcement of its intention to remove third-party cookies by 2020 are only accelerating their decline. A few months ago, Adidas said its attribution models tend to produce the wrong results, for example suggesting that performance advertising is the main driver of online sales. Through econometrics, Adidas eventually discovered that the role of video and other brand-centric activities had been vastly underestimated.

Ironically, what was once supposed to be the main benefit of attribution models – tracking the individual customer journey – is now turning out to be their trap. Aside from legitimate privacy issues (damaging the industry’s reputation in the process), the use of person-level data can be a deceptive basis for making marketing decisions. This is especially true for industries like automotive and consumer staples, where decision making is complex, brand value plays a key role, and offline channels account for the majority of sales.

Tracking a single user can distract from the big picture

On the other hand, individual career paths have never played a role in econometric methods such as the modeling of the marketing mix, which are now experiencing a comeback. They all involve producing generalizable answers from generalized data and creating a time series that clearly relates marketing activities to sales. Done well, marketing mix modeling provides an accurate estimate of the real impact of past and future marketing activities.

Of course, marketing mix modeling faces its own limitations. First of all, it is expensive and time consuming. Also, to be precise, it takes at least two years of historical data. This explains why even many large companies can only afford to perform marketing mix modeling in two to three year intervals. Perhaps most disadvantageously, however, traditional marketing mix models have no real-time relevance. For example, marketers may discover through marketing mix modeling that a channel is underperforming, so they will reduce investments and shift budgets to the best performing channels. But the actual effects of these actions are obscure until the next marketing mix modeling cycle is complete, which may be several years later. Thus, companies are not able to constantly change their course of action, making the method unsuitable for management purposes. Additionally, the effect of activations and channels with comparatively low investment – which disproportionately includes digital activations – is very difficult to measure. In summary, traditional marketing mix models are not suited to take advantage of the real-time availability of digital data and the newly acquired ability to optimize short-term activations.

The best of both worlds: towards modeling marketing effects

Marketing mix modeling as we know it has been around for at least 30 years. Innovative thinking and contemporary technology can solve many of its problems. Based on this premise, we created the marketing effects modeling approach, which we see as the next step to combine the big picture of marketing mix models with the real-time and behavior-based nature of digital data. The idea is quite simple: to the standard linear regression analysis, which establishes a historical baseline, we add the live metrics component.

First, the data available daily and derived from the company’s KPIs is added to the mix. Pretty much anything that fits the bill can be integrated, from Google searches to store visits or even the number of calls arriving at a call center. The model then uses machine learning algorithms to fill existing data gaps by improving daily values ​​with predicted hourly values. This way we get live measurements which have proven to be very accurate in our work with major FMCG customers. As a result, we essentially get a real-time continuous system that offers all the benefits of marketing mix modeling without its drawbacks. Additionally, digital metrics allow us to measure organic consumer interest in brands, products, and campaigns, factors that are often overlooked when analyzing media effectiveness. Ultimately, paid media investments are only a small part of the equation, and digital metrics can improve our understanding of many other contributing factors, including brand strength and external consumer trends.

For the first time, marketers can track their marketing ROI in real time and across all channels without having to rely on personal data. The ability to continuously monitor a channel’s performance saves on non-value ads and maximizes the impact of invested budgets.

Exciting times for the industry

Considering the benefits, we expect to see a proliferation of similar approaches relying on classic marketing mix modeling with new technologies in the years to come. Over time, they will also become more affordable for small and medium businesses. It will be interesting to see how not only media planners, but also strategists and creative planners exploit these new opportunities.

Econometrics is also accompanied by a shift in focus towards marketing theory, the good old 4P. As reporters Drum McCarthy and Blustein point out, we are going to see “a return to brand awareness and direct response campaigns” as well as “to the traditional cornerstones of advertising where every move and action of the consumer doesn ‘is not attributed to a single d. “This is good news, both for the marketing industry, for large advertisers and for consumers.

Niklas Stog, associate partner at TD Reply.