Attribution is a term that means all things to all people; but its complexity is often misunderstood. In this feature, ExchangeWire speak with Paolo Gaudiano (pictured below), founder and CEO, Infomous, and attribution expert in his capacity as board member of attribution specialist, Concentric. Gaudiano explains how a 'concentric' approach to attribution is much more effective in understanding the individual and how that understanding can be scaled into marketing strategies.

Attribution is an important topic, but one whose definition is not entirely clear. Having studied and applied human behaviour for more than two decades, I want to describe a novel approach to attribution – and, more generally, to marketing analytics – that places the consumer squarely at the centre of the problem, analysing the impact of marketing in a 'concentric' fashion, from the inside out. This approach is the heart of a marketing analytics firm that I helped to launch five years ago, aptly named 'Concentric'.

Let’s begin with a definition: from a marketer’s perspective, attribution generally refers to understanding which marketing activities and touchpoints contribute to the ultimate decision of a consumer to purchase a product.

But this simple description conceals the complex chain of events that shape the consumer’s behaviour. When you consider the vast number of activities, touchpoints, and metrics available to today’s marketers, attribution can be analysed at many different levels: one might ask how much a Facebook campaign contributes to spontaneous brand recall; how word of mouth interacts with other channels to influence brand perception; how product packaging influences in-store purchases; and so on.

Clearly, many complex interactions determine how an ad ultimately influences a consumer’s behaviour. It is exactly the complexity of these interactions that motivated the concentric approach to marketing analytics.

Your market as a complex ecosystem

The term 'complex' refers to any system with two characteristics: 1) the system consists of many elements interacting with each other and with their environment, and 2) the behaviour of the system as a whole emerges from the individual behaviours and interactions in ways that are not always obvious.

Examples of complex systems exhibiting emergent behaviours are common in nature: flocking birds, schools of fish, termite nests...

In contrast, the behaviour of many human-made systems can be predicted from the behaviour of their elements: cars, airplanes, buildings, computers... in these examples, each part is doing a specific task, and its impact on the entire system is clear. If a car stops running, you can usually identify and replace the defective part.

But there are many human-made systems that are, in fact, highly complex 'ecosystems', whose behaviours emerge in unpredictable ways, such as traffic jams, stock-market crashes and stampedes. And we don’t need to look at such extreme examples: I would argue that all businesses are complex systems, whose emergent behaviours – such as market share, revenues, and customer satisfaction – depend, in complex and often unexpected ways, on their employees, clients, competitors, and on the business environment in which they operate.

And this is the heart of the problem: the business management tools developed over the last century are based on the assumption that business problems can be solved by analysing and improving each of the components of the business in isolation – what is known as a reductionist approach.

But the accelerating pace of technology development, increasing globalisation, and the availability of instantaneous communications, have lead to an exponential increase in complexity, which makes a business more like a school of fish than a car. As a result, reductionist approaches become inadequate for business management.

The inside-out approach to managing complexity

The concentric approach takes a completely different view of the problem, focusing on individuals – in this case, consumers – and their interactions.

On any given day, a consumer may see a product, use it, talk about it with friends, see ads, and so on. Each of these interactions has a slight impact on how the consumer perceives the product. When it comes time to purchase, the accumulation of all these experiences shapes the decision-making process, ultimately leading the consumer to purchase product A instead of B.

The concentric approach begins with a model of an individual consumer, including the behaviours and influences we just described. Each simulated consumer is based on existing theories and empirical studies on consumer behaviour, such as interpersonal influence, price elasticity and advertising effectiveness. We then build a software simulation in which individual consumers go about their daily activities. We create a marketplace with thousands of consumers whose individual characteristics match the characteristics of the real population.

We then add all relevant, competing products, and we overlay communications channels populated with messaging, based on actual media activities for all competitors in the marketplace.

When simulated  consumers 'see' an ad, their perception of the brand changes slightly. Their perception also changes when they use the product or talk about it with friends. When the time comes to buy the product, each simulated consumer decides based on its accumulated perception about all available brands, with some randomness to reflect the sometimes irrational choices made by consumers.

When we let this simulated marketplace run for a full year, we are essentially replicating what happens in the real world. We track metrics such as brand-awareness or market-share across the simulated population. And, we calibrate the simulation, until the emergent market metrics match those seen in the real world.

But, unlike the real world, we can look inside the brains of our simulated consumers to see exactly how each interaction with a brand contributed to the final outcome. For instance, we can tell exactly how many times a consumer saw an ad, see how it influenced their perception, and how this ultimately influenced purchase decisions. Similarly, we can evaluate the contribution of word-of-mouth, product quality, competing products, or any other factor in the simulation.

Using this approach, we always exceed 90% accuracy – meaning that we can start the simulation as if it were, let’s say, January of 2015, simulate an entire year in terms of what products existed and how much advertising was done on what channels, and replicate with, better than 90% accuracy, how sales, market share, and consumer sentiment evolved by the end of the year.

Once the model is accurately calibrated, we have a high degree of confidence that we have properly captured the causal relationships in the model. We can then test entirely new scenarios with a high degree of confidence that the results will be very close to what would happen in the real world.

For instance, we can test different media plans, competitive strategies, or virtually any initiative available to a marketer. We can even test new product launches, a thorny problem for most analytics platforms. Think about it: if you see an entirely new product in a store, do you freak out because you have never seen it before and need to retrain yourself? Of course not! In our simulation, the causal relationships allow us to estimate the reaction to any product, old, or new.

It may seem difficult to believe that this approach actually works. My company, Icosystem, has successfully applied this inside-out thinking to a wide variety of business problems, helping global organisations for the past 16 years. We first applied this approach to marketing in 2005, and after several successful projects, with some leading agencies and brands, we created Concentric. Since that time, Concentric has worked with amazing clients – often having to overcome significant skepticism. In the end, however, we have proven that this approach is successful, and that is a very powerful way of addressing attribution and a wide range of additional marketing issues.