Do You Forecast Bookings or Shipments?
We see the need to have a broader discussion regarding the forecasting practice in Semiconductor industry of baking supply constraints into the demand forecast. While this serves the revenue forecasting needs, it obscures the true demand signal and can lead to self-perpetuating chip shortages in future.
Semiconductor companies have been extremely supply constrained for the last 2 years. This has put further strain on the IBP best practice of forecasting unconstrained demand and applying supply constraints to it later during supply review step to produce the constrained supply response. As semiconductor companies look to forecast revenue, it seems more intuitive to just forecast supply instead of demand, and then allocate constrained chip supply to competing Finished Goods products, channels, and large customers. Why forecast true unconstrained demand, if you cannot meet it in the foreseeable horizon? This article covers GitaCloud perspective on forecasting bookings (true demand supposedly unconstrained by supply) vs. forecasting shipments (which make the process of allocation planning and revenue forecasting easier).
Let's define bookings vs. shipments with an example. Raw Bookings are the sales order schedule lines representing customer requested delivery dates and quantities. Confirmed Bookings are sales order schedule lines representing confirmed delivery dates and quantities. For example, a customer order is received on 15th May 2022 (order creation date): for 500 units to be received (requested quantity) on customer docks by 15th June 2022 (requested delivery date). However, we were only able to confirm 300 units (confirmed quantity), that too a month late: 15th July 2022 (confirmed quantity). Now, let's say that we have only been able to ship 200 units (shipment quantity) on 15th August (shipment date), and the remaining 100 confirmed units are on backorder.
The question now is: what constitutes our sales history for statistical forecasting purposes: is it raw bookings of 500 units in June; or confirmed bookings of 300 units in July; or shipments of 200 units in August? If the planning process is largely looking to forecast revenue, one might argue that the raw bookings are quite meaningless to forecast, specially given the chronic chip shortages. It would be so much easier to simply see the trends of supply constraints and bake future supply constraints into the demand forecast (forecast for 300 units, not 500 units in this example).
In case of a gap between confirmed dates / quantities vs. shipment dates / quantities, we recommend to capture this with a separate metric: plan attainment (planned vs. actual shipments), not forecast accuracy metric (which should remain: forecasted vs. actual raw bookings). To the extent, we are able to provide alternative products to meet demand, the views of sales history from confirmed bookings or shipments can further distort the true demand signal.
Bookings Forecast should be a "supply unconstrained" demand forecast in order to capture the true demand. We do not want to pollute our demand forecast with anticipated supply constraints. It is important to understand that Bookings Forecast is not truly unconstrained in all aspects, as it is indeed constrained by demand constraints: such as Pricing, Product Transitions, Promotions, Market Trends, Competitive Activity, etc. For example, if we wanted to aggressively capture market share for a certain product family in a certain region, we could decide to cut prices by 50% and this will likely generate a very different statement of 'unconstrained demand' than demand we have based on the demand constraints, such as current pricing. The unconstrained demand is the likely demand our customers will place given our current and planned sales and marketing motions: including pricing, promotions, product transitions, etc.
To the extent, supply constraints impact demand shaping strategies, the 'unconstrained demand' is constrained by both demand constraints as well as impacted by supply constraints. For example, a 50% off promotion in Consumer Products industry that is likely to inflate demand by three times, may get scaled down, if we need to engage third parties to produce the excess demand beyond our manufacturing capacities, and if such outsourcing returns negative margin at the 50% price point. In this case, one could argue that the supply constraint within our in-house manufacturing network is driving a margin concern due to the mandatory outsourcing to meet promotion volumes. This supply constraint is changing the demand constraint (planned promotion), hence impacting the 'unconstrained forecast', to the extent we get a different unit forecast based on a weaker 25% off promotion.
Semiconductor companies have another challenge to deal with: customers inflate orders to get better allocation share. For example, if a product family is on allocation due to chip shortages, customers may double the ordered quantity if they expect to only receive 50% of what they order. Requested Quantity in SAP Sales Orders in this case will not be an accurate metric for forecasting future 'unconstrained bookings'.
Nevertheless, we recommend to load separate views for requested quantity (bookings: unconstrained demand) vs. confirmed quantity (shipments: constrained supply) within your IBP platform. Statistical Forecasting / Machine Learning should apply to requested quantity (bookings history), not confirmed quantity or delivered quantity (shipment history). It's critical to retain visibility into supply unconstrained demand forecast, even if there is a significant gap between bookings and shipment lines due to persistent supply shortages. Forecast error is a measurement of forecasted raw bookings vs. actual raw bookings, not forecast vs. shipments, which conflates the forecast accuracy metric with the plan attainment metric.
If Sales or Demand Planner teams start to bake in Supply Constraints into the Unconstrained Demand Forecasting process, they risk perpetuating the supply constraints for the mid-long term: as Supply Planners are not even challenged to find a way to resolve the Bookings-Shipment gap. Customers will simply learn to place orders in alignment with supply availability or in a worst case, find other channels to fulfill their unmet demands closer to when they want the product.
We may choose to cleanse raw bookings history, where we see demand inflation, to provide a statement of true unconstrained demand (bookings forecast). Customers looking to get a higher share of constrained supply may send in inflated orders: this pollutes raw bookings and should be corrected by demand planners on top of the outlier correction capability within IBP.
We recommend to stay with the demand review - supply review - reconciliation review - executive review cadence of the best practice IBP processes and manage revenue forecasts in the reconciliation review step. This enables both the capture of unconstrained demand during demand review step, constrained supply from the supply review step, and customer allocations during reconciliation review to generate the revenue forecast.
