Sales forecasting is an art form, if, like me you appreciate statistical analysis as an art. To truly understand that art, this post will review some key points in general sales forecasting, before moving on to the key discussion point, sales forecasting through uncertainty.
In this article, we use the COVID-19 outbreak as a coefficient for our multiple regression model, if you are here because you unnderstand this concept already, please skip ahead to my thoughts on planning and preparedness for uncertainty and industry shocks.
If you are interested in understanding forecasting and preparedness strategies, then please, read on as we cover, the basics of sales forecasting, multiple regression models, the value of the data and lastly, applying this to system shocks, such as COVID-19.
When talking about the annual planning process, I regailed readers with a story on a sales director that set targets based simply on naive forecasting. Put simply, this was what we did this time last year, with a little bit more added on. It may be possible, although I have never been on the receiving end of it, with a little take off. Naive forecasting is the very basis of sales target settting and planning, but it is also the simplest and most prone to error. There is simply no way that this type of forecasting can deliver any value in the context of industry, or global shocks. Therefor it is simply in this post to highlight just that fact. If the shock, or change wasn’t in existence last year, naive forecasting simply cannot work.
Naive forecasting may well have been the way of the world at the begining of business. Luckily though, it didn’t take long for people to begin adding their own information into the mix to begin to start to move the goalposts on sales planning. I have worked in very large global companies where the the sales planning strategy is naive forecasting, with a lot of information layered on top of it. In all honesty, this is how most businesses work. However, you can’t apply to these feelings with any industry shock, be it an outbreak of a virus, or something more positive effectively with just gut instinct. Sure it was obvious that England making it through to the semi finals of a world cup. Along with a hot summer people would buy more beer. However, through statistical analysis, the amount could have been much more widely known and discussed.
In many businesses, that is all that is used. For me to comment on what is needed, depends on a business situation and it’s appetite for risk. Why risk? I hear people say. I say risk for a simple reason. With a tighter outlook on sales forecasting, through uncertainty and industry shocks, be it positive or negative, we have a greater propensity to derisk areas of business.
Derisking is fundamental within any business through a shock, or a time of uncertainty. The issues faced by big business and SME’s may, at some points, be different. Whilst at others the same. However, the takeaway here is that there are not many business should be actively risk seeking in all of their endeavours. Having a greater level of knowledge can, generally, reduce the risk that is faced by any business. I have outlined some examples of risks that can be mitigated with solid shock planning.
Having an understanding of the risk and potential impact on stock levels allows a business to order the correct quantities in advance. Don’t forget, these can be peaks, as well as troughs. For example, a pub may have to evaluate how much beer to order based on temperatures expected in the summer period, whilst this can be done with naive forecasting and a little knowledge that a heat wave is imminent. A legitimate data driven approach will give a much tighter start point, than simple naive forecasting. Allowing, if neccessary further input from the team to tighten the forecast further.
Knowing a spike or decline is set to occur gives businesses an opportunity to adapt to these changes from a labour perspective effectively. Going back to the beer analogy, knowing there will be a 46% increase in sales through a summer period, begining in 9 weeks, and that it takes a 3 week onboarding cycle. The hiring period would be around 8 weeks prior to the spike starting. Ensuring a good level of staffing where required.
Conversely, should there be a decline, it may be best served to begin certain process improvement projects, or reallocate staff to other areas of the business, not over-burdoning the business with extra labour in the wrong areas. Drastically impacting productivity and conversion costs. Adversely affecting the bottom line.
By understanding what the potential peak or trough of any potential impact of such a shock will allow a business to incetivise and retarget staff accordingly. This will reset expectations and instill confidence that management are aware of what the situation is, and are planning accordingly.
For situations such as COVID-19 – this could involve release of information on potential sales decline, and what is expected of employees. Allowing management to communicate effectively why decisions are being taken. What the plan is, and how it will progress with ongoing analysis.
If through conducting correct data analysis, it may become apparent that a business fortunes is directly linked to another element. This could be outside of it’s control. It may sound basic, but for ease of understanding, let’s go back to the pub. If a pub only makes profit due to soaring temperature and beer garden sales. What happens if there is an extremely cold winter. Noting just how much the reliance is on external factors allows businesses to diversify. Focussing on revenues not directly linked to that external factor.
I don’t want this to get too technical, ultimately this is a post about forecasting shocks into a model for business mapping. That business mapping tool, that will be discussed is a Holts-Winter model. It is not a step by step guide of how to do it for your business. Moreover, a why youd should do it.
If you are looking for a step-by-step guide on Holts-Winter, this is not the post for you. I reccomend looking through this website to get a better understanding of the methodology.
This post has been updated as the UK has moved through the peak of COVID-19. It has become clear, when looking at certain industries. Such as Hospitality. That even forecasting such as this will be inherently thrown simply due to the size of the shock.
Holts-Winter takes an external data source, and internalises that datas effects on sales. Holts-Winter itself takes into account seasonality, allowing the user to internalise the external dataset to the model. Through trial and error the user can determine if the external dataset is firstly relevant. Secondly, how it can relevant it is. Again without getting too techical, this is through minimising the standard error and mean squared error.
As a business you can apply the Holts-Winter approach to your dataset as of today. Especially if you are noticing sales movement in one direction or the other and can see no discernable reasoning. There are data-sets out there that you can use, just remember that this is a learning algorithm and a moving set of information with infections.
You can download infection numbers and introduce this into your dataset to review correlation and fit.
With any statistical analysis, there are some gotchas.
Truly the biggest gotcha with an industry shock, is having to believe that it is completely relevant to your business. Undoubtedly, COVID-19 will have an impact on business to some degree. Whether or not it is material or not is the question. This type of modelling can help identify this. It is far more of a science than naive forecasting. However the science isn’t exact. Especially with the limited data. To base decisions purely off of this data, would be a big gotcha. Basing it off of no data, in my opinion, would be an even bigger one.
It is hugely important to remember that forecasting is the basis of supporting a much wider strategy. Through this type of forecasting, it is possible for businesses to make contingency plans. Contingency planning and forecasting can allow management to communicate best case, expected case and worst case scenarios effectively. Simply putting the forecasting in place alone, is not enough.