The function of planning and forecasting goes back a long way and has mentions as an ancient practice in religious and historical texts. Like all other phenomena, the function deals with development and advancement, whether theoretical or technological, in a cyclic manner – Realise, Reconstruct, Refresh and Repeat. When it comes to modelling times series sales data, the math has remained fairly stagnant, what has changed is how we make use of it.
The more sophisticated forecasting software, including Planamind now offer Expert Selection as an option. This article seeks to analyse the role this feature plays in transforming how forecasting is carried out and explain the interdependency required to make it perform at peak efficiency.
But, first a little background.
In the early days of technology supporting the demand planner, the forecasting software had to be fed in parameter values such as alpha, beta and gamma for exponential smoothing. This means earlier systems, were only equipped to handle forecasting using simpler models. As technology advanced, newer softwares built up the capability to handle more and more complex models. Nowadays, the system runs through all available models using around 150 rules of selection- and the number of possibilities are reduced at each stage. Eventually, through this process of elimination the system is left with what it considers is the most optimum. What this also does is tell the user how perfectly or imperfectly the data is matched. A fully automated feature in this, provides the user on the best model and the best parameters for the model, given it’s current data set.
So why is the expert selection model a game changer? Today, the demand forecaster function is slowly merging with the demand planner role. What that individual needs is the ability to reasonably estimate demand in order to properly anticipate it . What a fully automated expert selection option does is free up his time from being bogged down in the mechanical mathematics of it and instead contribute his expertise and experience, in addition to the insights he draws from reports, to have a more robust forecast.
The next question is how – how does the demand planner maximise utility of this feature ?
By presenting it with scrubbed data, that identifies impactful external events and teases out the extent of their influence. By presenting it with data that notes and corrects according to decisions by the supply chain or marketing team. By presenting it with as much relevant data as possible.
The other factor that needs mention is of the time saved by using this option. Expert Selection seamlessly carries out in a matter of minutes what would take the demand planner hours to do. Consequently, availing of this feature helps streamline the overall planning process. It means planners work more productively and that more time is spent increasing the accuracy of the forecast itself.
The last aspect to this is the interdependency factor. A machine cannot think for itself. A machine does not look at an outlier and decide for itself that this should not factor in the decision making of the model. It needs the planner to do that. It needs the planner’s instructions to properly gauge the data and the trends it senses. The closest analogy we can think of is trying to solve a 10,000 piece puzzle. What the system does is figure out the corner piece and then try out the other 9999 in all angles till it finds one that fits in seamlessly. What the person does is collect all the border pieces, then segregate based on the image and so on. The fastest results come when the two things work in tandem. And so it goes for expert selection.
Please leave us your comments on the points made and let us know if we missed any matter of significance.
There are some events that temporarily affect demand as it happened during demonetization in India and then there are events that affect supply like local political issues. However, very rarely an event happens that affects both demand and supply making