Fluid Dynamics And Its Application To Marketing

  • Uploaded by: mark brown
  • 0
  • 0
  • December 2019
  • PDF

This document was uploaded by user and they confirmed that they have the permission to share it. If you are author or own the copyright of this book, please report to us by using this DMCA report form. Report DMCA


Overview

Download & View Fluid Dynamics And Its Application To Marketing as PDF for free.

More details

  • Words: 2,319
  • Pages: 4
edmonds_brown.qxd

12/12/01

18:59

Page 1

STRAP LINE

The smoking chimney: a new approach to modelling Doug Edmonds,Tortilla 71 and Mark Brown,The Marketing Store, describe the development and implications of a new modelling technique

H

AVINGTAKEN a new approach to

modelling based on what we have experienced by observing nature, we have arrived at an interesting approach to modelling that produces easily actionable and interpretable results. The trouble with articles about models If modelling is seen as irrelevant and boring by the marketing and research community at large, how do they perceive an article about modelling? We won’t answer that question here, but we hope we can offer a fresh perspective on the modelling debate. In particular, we hope to spur modellers on to bring their wares out of the back room and on to the boardroom table.

The trouble with models Modelling is the furthest activity from most brand managers’ minds. To be a good brand manager does not require one to be good at numerical analysis. What makes a good foot soldier is different from what makes a good general. Consequently, the downfall of most sales models is that they cannot easily be understood by the numerical layman. Herein lies the problem. Most models are developed for the modelling community. They approach the whole process of analysing effectiveness from the point of view of someone, the modeller, who has excellent numerical and analytical skills. In itself this is a fundamental problem. As Schulz and Meer wrote in September’s Admap (1), there are some principles every modeller should follow in order to be able to communicate to a broader audience: ● Transparency. ● Simplicity.

March 2001

● Causal factors are not abstract. ● Intelligibility.

We have always been interested in models but not in the minutiae of them. We aim to demonstrate a new model that has been developed with brand management in mind. We begin our story, as many inventors have done before, with a random walk. A stroll that is possibly more interesting than the model. Our random walk marries the unlikely bedfellows of smoking chimneys and marketing. EXHIBIT 1

The smoking chimney

Chimneys We have spent many hours watching smoking chimneys spewing out their fumes. As students of gas and fluid mechanics in earlier incarnations, we realised that there might be a valid link to marketing and research. When a column of hot gas exits a chimney or smokestack, and enters a static volume of air, several interesting things happen. At the edge of the rising column of smoke the air surrounding it is still.The difference in speed between the moving smoke and the still air starts a physical process. (Students of fluid

Admap © World Advertising Research Center 2001

mechanics would call this process entrainment, but the name isn’t important.) The rising smoke causes the still air to move, by the process of friction. As the air starts to rise it is drawn into the smoke. As it is drawn into the smoke it mixes with it and cools down. The cooled smoke then slows down. At the same time it begins to roll outwards away from the centre of the column, causing it to thicken (Exhibit 1). This sounds nice but what does it mean to you? To answer this required a lateral jump. The lateral jump from chimneys to marketing For us, the lateral jump came by putting this effect witnessed in nature in the context of a brand’s sales curve. From years of developing and attempting to evaluate various forms of marketing activity, one thing is certain: whether it is advertising, new products, sales promotion, price-cutting, off-shelf display or any other interaction we can create between brands and consumers, the results are unpredictable.This is not to say that we cannot get close to estimating a volumetric effect, but this is usually based on past experience. Quite what happens and why may be a mystery. This is because there is not a linear relationship between marketing activity and consumer response. This led us to think that there may be a ‘critical mass effect’, whereby activity needs to be of a certain scale to jump to a new level of effectiveness. The trouble is that this is not true. Some small-scale activity can be highly effective. So we began to think of the effect as a series of small eddies, which sometimes come together to create bigger ones and sometimes cancel each other out. This

1

edmonds_brown.qxd

12/12/01

18:59

Page 2

STRAP LINE

EXHIBIT 2

Raw sales data for an fmcg brand in its seventh year

seems to make sense, because it may help to explain the effect of ‘word of mouth’.A particularly relevant price promotion in a particular store may get talked about among friends, which could have the unexpected effect of increasing store traffic. An ad campaign for a beer with a couple of frogs in it may get talked about in the pub or the playground, preconditioning TV viewers for their next viewing, thereby increasing awareness scores among the target market and unexpectedly creating a playground fad. If we take this principle and assume that the proximity of the consumer to the activity affects the warmth of the relationship to the brand, we can picture whirlpools with strong brand advocates in the middle, slowing to brand acceptors at the edges. Like all good analogies ours is less than perfect. It does, however, provide a springboard for defining what gives a sales curve (for an fmcg brand) its shape. When a brand is launched it attracts a group of users.These stay as hot users – or, if their demand for the product is cooled, they become less warm to the brand. In the extreme instance they may even stop buying the brand. Moving this analogy on helped us refine how to construct the conceptual model.

over a long period, it heats more slowly. The smoke column will draw in air at a lesser rate – and cool at a lesser rate. It does not take an enormous leap of imagination to see these that effects relate to promotions and advertising, and their subsequent effects on a brand’s sales. Increasing demand for a brand quickly is the effect of a price promotion or a viral campaign. Slowly increasing the demand for a brand over a longer period is the effect of a piece of communications – what used to be called advertising. An epiphany? Not an epiphany as such. We realised, though, that if we looked at consumers in the context of this modelling approach we could account for the effect of promotions in the same way that we could

account for the effect of advertising. In short, because both effects modulate consumers’ behaviour in a similar fashion, but on a different timescale, we could isolate one effect from the other with relative ease. It all sounds good, but does it work? What is this model about and what does it hope to do? The conceptual model This model is based on behaviours. In essence, the modelling process begins with identifying a brand’s individual buyer’s behaviour. Once a sales curve has been converted into a series of different buyer curves, the second phase of modelling begins. Arguably, this is the more complex and time-consuming phase, as it involves modelling the changes in the buyer groups’ size in the face of communications and promotions. To repeat ourselves, we can overlay the effect of both communications and promotions on to the model because we are dealing with people. It might sound odd, but people are easier to model than abstract concepts like advertising and brand awareness.This modelling process is based on the assumption that distribution remains constant. If this is not the case, then things become a little more complex, but still achievable. The benefits of the conceptual model Like other modellers, we would love to be able to predict sales curves of brands with the minimum error. However, to

EXHIBIT 3

Creating a numerical model from the sales data

Turning on the taps Imagine we can apply heat to our smoke plume at any point, with any intensity we wish. Heat it very intensely over a short period and it becomes more turbulent, and draws in more still air, causing it to cool quickly. Similarly, if we heat it gently

2

Admap © World Advertising Research Center 2001

March 2001

edmonds_brown.qxd

12/12/01

18:59

Page 3

STRAP LINE

EXHIBIT 4

Converting the sales data curves into popualtions

achieve this with any model means being able to incorporate changes in sales that are dramatically removed from the underlying base sales. It is these spikes and troughs that cause most models to fall down when trying to predict them. Whether our model can achieve such accuracy in every case remains to be proven. We believe that there is merit and consumer insight to be drawn from first being able to create a model for the underlying sales.The immediate benefits of a model that can get close to replicating base sales are three-fold: ● Broad consumer behaviour. The model paints a picture of the broad behaviour of the buyers of a brand. This can be as informative to the brand owner as being able to determine the relative effectiveness of promotions vs other communications. ● Indication of the effect of long-term strategy. Using this technique it is possible to evaluate the efficacy of a strategy in terms of changing the profile of a brand’s buyer groups over a period of time. ● Project, over time, the base-level sales. Once we have created a model for baselevel sales founded on groups of people, it is then possible to use the curves representing the populations to predict behaviour and sales in the future. In generating and creating this approach to modelling we have come to the belief that the downfall of many models is the urge to try and replicate real life to two decimal places.While we hope that one day our model will be able

March 2001

to do this, we are developing it primarily to provide, at least, transparent results to the marketing team. At this stage we need to bring the model to life with an example.This is a little more complicated, so we have skipped some technical details to give a taster, rather than a complete meal. An example To demonstrate our model in practice we have chosen to show how we would create a model for an fmcg brand in its seventh year of existence. Exhibit 2 shows the raw sales data that we were given to construct the model. As outlined above, the first step is to

take the sales curve and create a numerical model for it. At this point we are not interested in awareness, media spends, promotional activity or distribution.We just want to be able to create a curve that has a similar shape to the sales curve, using a known set of equations. The results from this are shown in Exhibit 3. To move into the next phase of modelling we make our first assumption: the different curves we use to create the model replicate the value contributions of heavy, medium and light brand buyers. Using buyer data we can convert these curves to populations. In doing this we arrive at the bedrock of the model. No longer are we operating in abstract realms of awareness and saliency but in the realms of people.This calculation gives us the curves shown in Exhibit 4. These consumer groups are based on a modelled sales curve that is actually relatively poor. To improve it we must cater for the effects of promotions and advertising. Because we are dealing with people, we can have a much firmer idea of how different groups will react to promotions and advertising.We won’t bore you with proprietary details, but we end up with a much more accurate model. By incorporating the fluctuations of the group sizes in the face of promotions, we increase the correlation between the actual sales and modelled sales, r2, from 0.66 for the basic form of the model up to 0.75 (Exhibit 5).

EXHIBIT 5

Improved correlation model (achieved by allowing for population fluctuations)

Admap © World Advertising Research Center 2001

3

edmonds_brown.qxd

12/12/01

18:59

Page 4

STRAP LINE

Where next? This is not the end of the process. We believe the model could become more advanced. Having developed it in its basic form, we now want to make it able to model the fluctuations between different user groups in a dynamic sense. To achieve a more sophisticated product, we need to test it on more data.This process has already started and we are fine-tuning our approach with each new set of data we gain access to. Conclusions The essence of our model is that, in an age where data are becoming increasingly more abundant, and the misuse and misunderstanding of data has become rife, we have created a way of modelling that people can understand. The reason for this is that our model aims to model people’s behaviour – and people, we find, understand people a lot more easily than black boxes. The chief benefit of the model is that it is possible to relate findings to real life. It allows refinement of brand, communications and promotions strategies with greater ease, and with greater insight.We hope that even in its infancy the reader can see how our approach can help demystify the complex arena of modelling. 1. K Schulz and D Meer: ‘How to make econometrics more valuable’. Admap, Sep 2001.

Doug Edmonds worked in the world of fluid mechanics prior to joining the world of advertising. He now runs the quantitative sales analysis consultancy,Tortilla 71,as well as being head of numbers for 2cv:research.

Mark Brown is director of planning at the Marketing Store, London.

4

‘The chief benefit of the model is that it is possible to relate findings to real life’

Related Documents


More Documents from ""