Collaborative supply chain management makes a lot of sense. In today’s day and age of ubiquitous internet connection, why should your suppliers be kept in the dark concerning your upcoming purchase orders? After all, if your company is capable of producing accurate forecasts about your upcoming orders, sharing these forecasts with your suppliers would certainly be of great help to them, which, in turn, would yield better service and/or better prices.
Yes, but all of this relies on a flawed assumption: order forecasts ought to be accurate. Unfortunately, they won’t be. Period. So whatever follows is merely wishful thinking.
Companies frequently get back to us asking if Lokad could forecast the sequence of upcoming purchase orders. After all, we should have everything it takes:
- daily/weekly future sales levels (forecasted)
- current stock levels, both on hand and on order
- purchase constraints
By combining these different elements mentioned above, we could certainly roll-out a simulation, and consequently forecast the upcoming purchase orders for a given period specified by a client. However, although this is something which is possible to do, the results of such an operation would be disastrous. In this short post, we share our insights on this issue to help companies avoid wasting time on such forecasting attempts.
Statistics are terribly counter-intuitive. As mentioned in our previous posts, “intuitive” approaches are most certainly wrong; and the “correct” approaches are unsettling at best.
The central problem with supplier’s order’s forecasting is that the calculations involved are relying on an iterated sum of forecasts; which is very wrong on multiple levels. In particular, forecasting the next purchase order includes not one but two variables: the date of the order and the quantity ordered. Depending on the supply chain constraints, the quantity ordered might be something relatively straightforward to forecast: if you have a minimal order quantity (MOQ), the order is likely to equal the MOQ threshold itself. On the other hand, if the item is expensive and rarely sold, the next quantity to be ordered is likely to be a single unit.
The true challenge lies in forecasting the date of the next purchase order, and even more challenging, forecasting the date of the following purchase order. Indeed, not only does the date of the next purchase order likely to have 20% to 30% error (like pretty much any demand forecast), but the date of the order that follows this last purchase order will have (roughly) twice the error, and the one after that (roughly) three times the error, etc.
As illustrated in the schema above, the uncertainty regarding the date of the Nth upcoming purchase order grows so fast in practice, that it becomes a worthless piece of information for the supplier. The supplier will be much better off doing her own forecasts based on her own demand history, even if this forecast can’t leverage the most recent demand signal, as observed downstream.
However, while forecasting purchase orders and sharing them with the suppliers doesn’t work, moving towards more collaborative supply chain management remains a valid business goal; it just happens that this type of forecasts is not the right way to execute this objective.
Stay tuned, we will make sure to discuss here in due course how collaborative supply chain management can be correctly executed from a predictive perspective.