“We learn from history that we learn nothing from history” (George Bernard Shaw)
Hindsight is a wonderful thing, mostly because it represents the perfect viewpoint from which to determine what would have been the best course action - that course of action that should have been obvious at the time, but for some reason was not followed. Market forecasting is a complicated business, as anyone on the receiving end of hindsight will tell you. Yet there are numerous case studies available that describe and explain the outcomes of past business decisions, things like product launches, market responses to particular events, and so on. These case studies seek to educate the struggling forecaster in the lessons of the past, allowing the forecaster to recognise and act appropriately if the same trend or set of circumstances appears again in the future. By such means paradigms and principles are born. However, such thinking can lead to the view that the market can be treated as if it were some vast clockwork machine-unfathomable in its overall complexity, but nevertheless conforming to sets of rules that become obvious to the wise. Thus, it is not surprising that many forecast techniques rely on simple interpretations of past performance as indicators of future business activity.
In one sense, forecasting based on past trends has never been so easy. With the rise of sophisticated electronic sales management and accounting systems, business forecasters are now data-rich. With a bit of knowledge and training it is possible to generate reports on exact past sales performances according to each territory, each product line, or each manufacturing site. The amount of historical data potentially available is vast and usually readily downloadable to a spreadsheet program with a graphing function. Plots showing the response of various market factors can be generated with a few keystrokes, and the data pondered upon at one’s leisure.
It helps that the curves often form comfortingly familiar shapes: bell-shaped curves, exponential-growth curves and so on. A common example is the S-shaped curve, which features a slow build up, a steep middle phase, followed by a gradual levelling off. Successful products launches often show this kind of growth pattern over time, with initial slow growth showing the gradual uptake of the product or word-of-mouth spread of the product’s benefits, followed by explosive growth as the product’s popularity sky rockets, and then levelling sales as the market becomes saturated. Some sales graphs might show periodic patterns due to seasonal changes or other regular or predictable occurrences, such as sales of hay-fever remedies that coincide with spring or the appearance of particular pollens. Such occurences, although not conforming to a recognised pattern, do tend to be predictable year-on-year, with variations in the heights or lengths of the peaks or troughs creeping in over time.
A forecaster seeing these graphs might make the simple assumption that the trends will continue, fitting in with the natural order of the market. Unfortunately, real data curves almost never conform neatly to ideal curves, and perhaps the most common curve type might be the random line, resembling the profile of some distant mountain range. Hindsight comes into play when interpreting the pattern-each movement of the line up or down can be ascribed to some event, perhaps a micro-economic factor, such as the launch of a competitor product; a macro-economic factor, such as a currency instability in a particular market, or even something quite prosaic, such as a period of ill health of a sales rep. Seeing trends continue, calmly and orderly, from the past and into the future, with any blips obviously explained, can lend a business forecast a neat sense of infallibility.
The entertainment industry has been around long enough for real life to have caught up with the science fiction epics of the past. Some visions of “things to come” have now already “been and gone.” Contrary to 20th century movie writers’ predictions of the future, the society of today is largely bereft of flying cars, the population usually avoids wearing aluminium foil, and we tend not to let psychotic, self-aware computers run our space programs. The trouble with hindsight is when one tries to turn it into foresight. While lessons can be learnt from the past, the fundamental precept remains that the future is unpredictable, so that any prediction is almost certainly wrong. What most predictions need is some indication of how wrong they could be; that is, some idea of the risk that the prediction might be so wrong as to be pointless. Such considerations are difficult to make just from a series of graphs or data tables that go into making a marketing forecast.
Making predictions by spreadsheet is difficult for another reason. A prediction of future financial performance of a company is very rarely based on a single graph. With the wealth of data available comes the problem of data-flood, and with skills in collecting and presenting data in a meaningful format comes the need for further skills in being able to identify the key aspects of the data, and the integration of those key aspects into a coherent whole. One might very well come up with a model based on the careful scrutiny of hundreds of pieces of data, hundreds of graphs, facts, and figures spread over many spreadsheets and workbooks, but what happens when the robustness of the model is challenged? Is the prediction based on principles and assumptions so clear that it can be articulated coherently and withstand critique? Also, what happens when the forecasting is the job of a single individual who then leaves? Are the base information and mechanisms of prediction formation logical and rational enough to allow a new person to take up the reins and produce forecasts in a consist manner?
Attempts have been made to improve the robustness of historical data analysis, including changing the assumptions used to extrapolate trends into the future. For instance, some systems allow more recent data to have a greater weight in the final prediction than data further in the past,the idea being that the recent past will have more of an impact on the near future than the more distant past. While such tactics might improve short-term accuracy, the underlying issue of reliance on historical data remains.
Part of the problem is that reference to past trends largely ignores the causal relationships between events. For instance, if a previously fast-growing product starts to show a slow down in sales growth, it might be assumed that the graph of sales reflects an S-shaped curve, and the market is reaching its saturation point. However, closer analysis might reveal that manufacturing and supply issues may be the problem, or that the sales message is no longer reaching the target audience, or any number of factors that, once addressed, would lead to a period of renewed sales growth. Without the ability to examine and critique the data to pick out underlying causes, and a method to test which of the causes might be to blame, the value of forward predictions based on past data is only going to allow reactive responses rather than proactive ones. Methods based on multiple linear regressions are available that seek to identify which of a series of factors are related to the variable under examination. Although complex and not without their own issues regarding underlying assumptions, regression analysis methods do have the advantage of being based on well-recognised statistical principles. If relationships between factors are observed in the data, the basis for that relationship can be clearly and logically defined.
Understanding at least some of the reasons that might be responsible for the data trends means that more informed assumptions can be made regarding the predictions of future business performance. For instance, multiple regression analysis methods may identify the presence of a relationship between sales and client contact time in one particular market, but not in another; this information could help determine territory-specific strategic plans. In addition, the application of regression analysis tools provides measures of error or variability, which means some judgement can be made as to the quality of analysis results.
To get the full value of any predictive tool requires the recognition of its limitations as well as its strengths. The analysis of past activity as a means to predict future performance does have its place in a stable environment, but reliance on this analysis as a single means to generate a forecast can be a pointless undertaking. Historical trends usually carry a significant element of risk when extrapolated into the future, and such analysis is often more suitable for data that either show little inherent variability, or for long term strategy considerations where a high level of uncertainty is accepted within the context of the forecast. Looking backward into the data may give a company a very good understanding of where it has been, and might even give clues as to where it is going. However, using historical data extrapolations and interpretations in order to make decisions in a changing environment such as pharma will be fundamentally flawed and will miss much valuable information that can inform a more forward-looking approach. Understanding the mechanisms that drive trends observed on the spreadsheets will not remove the inherent uncertainty of the prediction, but it will provide a basis from which to improve and judge the soundness of business predictions. There are systems that examine history but do not use it to make judgements on current decisions. These systems more intelligently rely on an understanding of current market factors (including PCT, Managed Care, regulatory, pricing, physician and patient perceptions and other factors) to make far more accurate decisions about what needs to be done now to create a more accurate forecast and what to do to improve the forecast by a specific amount.
For more information on this topic and how the next generation of analytics can help you, contact the author, Dr Andree K Bates at www.eularis.com to discuss your needs in more detail.
Author: Dr. Andree K Bates, Eularis, www.eularis.com
