Chaos theory

The aim of most companies is sustainable long-term growth. Unfortunately, in order to meet shareholder expectations they must strive to continually match, if not beat, the market: an objective that is rendered difficult by declining new product supply, which puts pressure on your current product resources to work harder.

In such a situation, effective management of promotional investment provides the next level of competitive advantage, one that some companies find surprisingly difficult to grasp.

One of the roots of this problem was all too apparent at a recent pharmaceutical conference which focused on return on investment (RoI). A number of companies discussed and presented their RoI measurement results and, throughout, the general common theme was the lack of agreement on how to measure RoI and a distinct absence of any underlying theory.

Given the current preoccupation with, and reliance on, measuring RoI, this is worrying.

This apparent ad hoc approach to RoI measurement has two important implications for pharma companies. Firstly, it means that cross comparison between studies, even within the same company, is difficult. Secondly, and more importantly, without theory as to how the causal relations between variables operate, it is almost impossible to interpret these results accurately.

All this exists despite the obvious importance of gaining a proper understanding of RoI measurement, and not just because effective promotional management gives you a competitive advantage and average expenditure on promotion is 28-30 per cent of sales.

The key problem is that maintaining the status quo and using last year's budget as a template makes your strategy visible and easy for your competitors to read. However, in the face of this fundamental strategic error, only a handful of companies regularly practice zero budgeting or evaluate the effectiveness of their campaigns. The reason: lack of a simple effective RoI yardstick.

As an industry, pharma actively measures efficiency against various benchmarks; call rate, coverage and frequency. In contrast, it actively fails to measure effectiveness: which activities are the right ones to do and will actually build sales growth? Pharma operates a black box model, which is a problem.

The aim should be to develop a clear promotional model that allows you to test the effect of alternative strategies. Armed with this, you can isolate effective combinations of activities, and identify and eliminate wastage. In this way, cash may either fall straight on to the bottom line or be deployed elsewhere and more effectively.

Such a model should possess certain characteristics. Firstly, it must allow the calculation of in versus out, not proxies for return. Secondly, the model must be underpinned by strong theory that provides the means to interpret, understand and use results. Thirdly, such models are no good in isolation, as a snapshot of a moving target; like Hercules' river (Hercules adapted the flow of two rivers - Alpheus and Peneus - to help him complete one of his 12 Labours), the market moves on and it is important for your strategy to remain flexible and adapt to changing conditions.

It is important to recognise, however, that the purpose of RoI calculation is not to provide a simple cost-cutting exercise, although it is a very useful way of reaching an informed decision about the success of a given activity and better than relying upon random cuts based on gut feel.

The primary purpose of calculating RoI is to maximise profitable sales growth. There are two synergistic strategies here: you can identify the promotional effect of each marketing activity, or individual sales team, and concentrate spend on high-return activities and cut low-return activities. The result: increased sales, more profit and less wastage.

You can also identify the different realised strategies within your salesforce and measure the effect of each one, then you can move your low performers onto a higher pay-off strategy.

What you intended
Most companies have a very clear intended strategy that they expect their salesforce to deliver. In reality, however, each rep is an individual who applies their skills and interprets instructions differently. Often, the tacit skills built from experience lead to different interpretations and mark out the difference between success and failure.

Recent research at Aston University has identified that most salesforces actually deploy six or seven different strategies, not just one corporate strategy. Unsurprisingly, this leads to several realised strategies - markedly different from the intended strategy - implemented by reps employing a variety of approaches.

More importantly, the productivity of these different realised strategies can also be very marked - on average 50 per cent of sales or more. This represents the low-hanging fruit of salesforce productivity, where a few changes to encourage reps to adopt the winning strategy can impact on your return on salesforce investment significantly.

Knowing the different ways in which your salesforce is deploying its energies allows you to shift emphasis and capture, on average, a 30 per cent increase on your sales line.

The potential rewards are great, so why do companies consistently fail to grasp this most important lever of competitive advantage? To some extent the answer is habit; they simply follow what they have seen work before. The problem here is that the market is becoming increasingly challenging and next year's success will accrue to those companies which predict accurately and ambush the future, not play safe and watch things happen - or worse, wonder what the hell happened.

Technological breakdown
Technology is yet another fly in the ointment of accurate and effective RoI measurement. The ready availability of statistical software and myriad different methods serve only to make the issue worse.

This problem was made obvious at a recent presentation I attended, during which the speaker presented his company's RoI results at length, yet displaying statistics that were simply wrong.

Most statistical methods are subject to strong assumptions that underpin them: ignore them at your peril. In this particular case, there was clear multicollinearity in the data; ie, strong cross-correlations between variables and the residuals from the regression (what the analysis did not explain in the data) violated the assumption of equality of variance.

In short, the calculation was incorrect because the assumptions underpinning the statistical test were violated and, therefore, the conclusions drawn from it were also wrong.

However, as most pharmaceutical data is not straightforward, it is a fundamental mistake to merely apply a few statistical tests to it. To illustrate a simple example, examine figures 1 and 2.  Figure 1 simply plots sales against activity: in this case, target calls. The box plots accompanying each axis illustrate that the data is slightly skewed, but for this example you can ignore that.

The simple hypothesis that is being tested, by placing a linear (dotted line) and non-linear (more solid line) through these points, is that the slope of the line differs significantly from zero. Visually, you can see that each line is broadly horizontal to the x axis, therefore increasing the number of targeted calls does not affect sales. The p value (probability of the result) of the linear regression line (p = 0.7771) confirms this. In short, this model does not explain your sales. Yet, without a clear theory you might accept this spurious result.

In theory, you should expect a relationship between the number of calls and sales. With one simple change to the regression model (see figure 2), the slope of the line is able to vary according to the level of targeted call activity. The situation has changed radically and the model now explains 35 per cent of sales, which is highly significant (p = 0.00001). One simple change produces a dramatic effect. Why is it, for example, a frequency effect?

Starting point
The key to developing any model is to start from a clear theoretical understanding. Think carefully about the relationship between your variables and how they may interact; synergy is, by definition, about interaction. Build a simple model to begin with and refine it systematically until you have a model that accurately reflects the theoretical relationship you are seeking to understand. It is this theoretical link that is missing so frequently from pharmaceutical RoI analyses, but how can you interpret the results without it?

This implies that pharmaceutical RoI calculations must be based upon deductive (theory-led) methods, not inductive (let the data speak) methods. In my experience, reliance on data-mining inductive methods is confusing. The simple fact is that pharma is awash with data and if you apply data-mining techniques to a large database, you will find statistically significant results.

The problem is that by varying the clustering or tree regression algorithm applied to the data, you will get different statistically significant results from the same data. This lack of prior theory is a major drawback of data-mining methods, leading to such analyses being branded `dustbowl empiricism' or `data-fishing'.

In contrast, econometric theory is based on decades of solid research and may be generally relied upon to provide strong robust analyses, which are more straightforward to interpret.

This is not to say that econometric and simpler time series analyses are without fault. I have frequently seen time series analysis applied to far too small datasets and this method carries a number of assumptions that are frequently violated by pharma data. Time series analysis can also often be wasteful: it may not make full use of the figures when applied to pharmaceutical data and the frequent use of inappropriate transformations effectively limits its usefulness still further.

A link to theory is essential for accurate interpretation and use. There are a variety of reliable, valid, reproducible techniques that provide accurate measurement of campaigns, fieldforces and promotional investments. However, it is important not to rely on one technique, but to triangulate your results using several complementary methods; each of which will shed some new light on your data, while, together, confirming your results. Try, whenever possible, to use raw data and avoid proxies.

In model-building, it is important to recognise that most people are not natural-born statisticians. We are not good at seeing patterns in noisy data, but we can be very good at picking non-existent patterns that suit our purpose.

Use of statistical theory attacks this problem in two ways. Firstly, statistical theory provides an optimal method for finding a real relationship in a noisy background. Secondly, statistical methodology provides strict checks against the over-interpretation of random patterns.

When building a model, begin with a clear link to theory. Use Occam's razor (the explanation of any phenomenon should make as few assumptions as possible; William of Ockham, philosopher and Franciscan friar) and apply extreme parsimony to your model, ie, restrict it to as few terms as possible.

Use objective tests, not subjective opinion, to judge between alternative models. Beware of models that mix qualitative data (survey results) with quantitative data. If you are looking forward, build a firm base through modelling past data, then work forward using sensitivity analysis modelling techniques.

Effective strategy is bold in execution and comprises a few key elements that work in concert. Kitchen sink strategy wastes resources, confuses customers and frequently erodes competitive advantage. Getting locked into set promotional patterns renders your strategy transparent to competitors and leads to loss of market share.

Flexible responsive strategies, based on sound information, test new opportunities and swiftly capitalise upon them, building sustainable competitive advantage. Regular RoI measurement underpins effective strategy and allows rational choice of the right mix of promotional activities to maximise sales. However, markets change and strategy is market- and segment-specific.

Recent research suggests that the average pharma firm probably wastes about 30 per cent of its promotional investment. If this were deployed elsewhere, it would undoubtedly increase sales and profits significantly. If it were merely cut and dropped to the bottom line, gross profits would rise by approximately 9 per cent.

To this opportunity can be added the extra profit that can result from a shift in salesforce strategy, from the most common realised strategy to the most effective one. The challenge is to ambush the future, unlock your hidden profits and transform the profitability of your sales and marketing operation.

The author
Graham Leask is a member of the Economics and Strategy Group at Aston University. He can be contacted at g.leask@aston.ac.uk