It is always puzzling for founders to solve the unexpected demand, especially in B2C/D2C business models. While we focus on acquiring customers, it is hard work to retain them. Unmet demand from the customers can immediately redirect them to our competitors.
Forecasting demand is a key to both revenue and cost management. If we forecast more than the actual demand, we would end up with increased inventory/holding costs. And a lower forecast results in revenue loss and customer attrition.
The ‘past’ is assumed to be an acceptable indicator of the future. It is important to look at the historical data and understand the demand behaviour for any trends/seasonality/cyclicality. The analyst should identify if the demand for a particular product is dependant on any other product/segment, (say, the demand for automobile components is dependant on the demand for automobiles/market growth).
The first month’s forecast would usually be a guess estimate based on the industry research and available information. If we have one month’s forecast (Ft) and the actual demand (Dt), we could try a simple exponential smoothing method to arrive at the next month’s forecast (Ft+1)
Ft+1 = αDt + (1-α) Ft
where α is the smoothing constant. Research suggests that a value of 0.1 to 0.3 would be good to be considered for the constant. If the difference between the Ft and Dt is less, then an α of ~0.1 to 0.2 could be considered.
Once we get the values of forecast and actual demand for three to four months, a simple three month moving average of the demand can be considered for the next forecast. We can even assign weights for the most recent period to arrive at a weighted moving average.
Forecasts should always be compared with the actual demand to understand the variance. We could technically look at the mean deviation, mean absolute deviation or mean absolute percentage variation. But we need to get a sense on the error margin, look for seasonality and keep improving our forecast process.
