Too subjective, simplistic, complicated or just plain wrong – that describes many forecasts in pharma. Yet effective forecasting is vital to deliver optimal growth at an acceptable level of risk, and brands must have realistic targets which are compatible with market realities.
Any forecast will (eventually) be wrong, and the further forward the forecast projects, the greater the degree of uncertainty, due to the scope for unforeseen factors to influence the outcome.
But people still expect forecasts to be accurate. Given we cannot predict the future, there’s an inevitable disconnect between expectations and reality.
When companies base the growth of their company, and their investment in R&D projects, on the predicted success of sales, this disconnect can end up having a profound effect.
If forecasting is doomed to failure, why bother? It’s about helping us to prepare for something that is in the future, to make decisions about what we will do.
Even if forecasts can never be 100 per cent accurate, the better their quality the better our decision-making will be, and the chances of our businesses prospering.
In the short term, those decisions may simply be ensuring resources are fitting demand; in the medium term quality forecasts provide management with a forward view of performance, enabling them to judge where the business will be at key milestones, in order to provide confidence, drive any necessary adjustments, and support immediate investment decisions.
It is the longer term where accuracy becomes vital. Not only is the margin for error bigger, but it is here where the crucial, strategic decisions are taken.
Here decisions are being taken about different strategies which will require differing levels of investment. Get it right, and the shorter-term tactical decision-making becomes less important. A strategy based on effective forecasting stands less chance of being knocked off course by unpredictable events.
Forecasting can be distilled into three tasks: reconciling, understanding and predicting/exploring.
Reconciling is about finding a compromise between the need to deliver optimal growth at an acceptable level of risk, and the need for brands to have realistic targets compatible with market realities.
With a strong level of understanding, predicting the future becomes more realistic: the sales level that can be committed to, the impact that can be expected from different activities, the external factors which could take the brand off-target.
A common error is forecasts based on too-subjective assessments. Often driven by politics, they can be produced to justify investment, linked to some pre-defined ‘magic’ number.
Another is to be too simplistic, looking backwards and hoping the same will happen in the future. In a world of constant change, this ‘projecting the past forward’ model simply does not capture how the market works. A variation is the global model which does not account for regional market variations.
Conversely, some over-complicate the process. Impossible to understand, difficult to replicate or update, and too complex for the job in hand, this kind of forecast is likely simply to be ignored.
The most serious common error is the forecast that is just plain wrong: using wrong or out-of-date data, over-reliance on historical or selective data, or not validating it properly.
The forecasting approach will be partly driven by the quality of the data available. The statistical model requires good quality data. This is the most reliable in stable-to-moderate market conditions, appropriate for better defined markets.
With little, or poor quality, data, a more judgemental approach is needed – more likely in highly uncertain, poorly defined or nascent markets, or where a new competitor has created an all-new paradigm. Let’s look at three approaches along that statistical-judgemental axis.
Data-driven
This involves the extrapolation of historical data or using analogues – accepting that over time the margin of error will inevitably increase. Its advantages are simplicity, the fact that there are quick, readily available tools to help build this kind of forecast, and ease of updating. Balanced against this is the assumption that history can predict the future, as well as the fact that it can hide seasonality. Either way, it requires reliable, accurate historical data.
Casual models
This approach allows the evaluation of the impact of changes in the market environment, and is therefore good for brand planning, although it requires quality data and an established relationship between cause and effect.
It can incorporate future changes in the market environment, and can be linked to impact marketing initiatives – so is more reflective of real market conditions. Against this, it can become very complicated, is time-consuming, and requires significant expertise in forecasting. Again, quality data is required, and it relies on linking the past to the future, assuming that planned initiatives will have the same impact as in the past.
Patient-based
Logical and easy to understand, this approach is based on our understanding of the market. It is replicable in different countries, and provides a tool to model the impact of future events. It can also be aligned to market drivers, and leads to better understanding of relationships in amongst the data.
Conversely, it is based on patient volumes rather than sales, so a measure of conversion will be needed. It also requires a sound understanding of market dynamics. Because it can become complicated and time-consuming, good design of the forecasting tools is vital.
We don’t live in a perfect world, so it is not always possible to ensure our forecasts are completely robust – in a perfect world, the data driving our forecasts would be available, complete and on time. While assumptions would ideally be backed up by robust references and research, there is not always a perfect analogue, especially for new-to-world strategies.
That perfect world would also allow us to know future events, and to validate market share gains and uplift through quantitative research and conjoint analysis. But the future is uncertain, and extensive market research is not always possible.
Often these annoyances are used by forecasters as an excuse for inaccurate forecasts, or, even worse, for delaying coming up with a forecast for so long that it is of little use in practical terms.
So build the forecast as soon as possible, so that you can establish the data requirements and identify the data gaps early. Government statistics can be useful to confirm figures for things like prevalence and population figures. Industry reports and horizon scanning can help identify major events that are likely to impact over the planning period.
Once you know where the data gaps lie, you can start to close them. Extrapolating existing data can be a useful first step, and this can be made more robust by cross-learning from the data available for other appropriate brands, therapy areas and markets. Tap into the knowledge of industry experts by using qualitative research as a base for uplift estimates.
Above all, don’t let the lack of data hold up the forecast. Instead, use estimates as placeholders – as long as you keep track of all the assumptions you have used, and tag the level of confidence in each of these assumptions, aiming to validate those ‘low-confidence’ assumptions as soon as possible.
Forecasting will never be an exact science in a world where things happen that we cannot foresee, and when we will never have perfect data all of the time. But this is not a reason for not striving to create the most accurate forecasts possible, because in the end these are crucial to helping us make the kind of strategic decisions which will define the future of our businesses and our industry.
- An extended version of this article is available at www.msi.co.uk
