Demand Forecasting Recalibration After a SKU Rationalization Project
SKU rationalization projects get a lot of attention while they're happening, the analysis, the stakeholder alignment, the actual discontinuation decisions, and then very little attention to what happens to demand forecasting immediately afterward. I've seen this gap cause real problems, forecast accuracy that quietly degrades for months because the forecasting model is still, in effect, calibrated for a SKU catalog that no longer exists.
Why rationalization breaks existing forecast models
Most demand forecasting approaches, whether a relatively simple moving average model or a more sophisticated statistical forecasting engine, rely on historical demand patterns to project forward. When you discontinue a meaningful number of SKUs, several things happen simultaneously that a model trained on the old catalog doesn't automatically account for.
Some portion of demand for discontinued SKUs transfers to remaining SKUs, as customers substitute toward the closest available alternative. Some portion doesn't transfer at all, representing genuine lost demand from customers who specifically wanted the discontinued item and won't substitute. And the remaining SKUs' own historical demand patterns may have been influenced by the presence of the now discontinued SKUs, cannibalization effects that existed before rationalization and have now changed.
A forecasting model that isn't explicitly recalibrated after rationalization will, at best, produce forecasts based on stale historical patterns that no longer reflect the actual catalog and demand dynamics, and at worst, will completely miss the demand transfer effect, understating forecasts for surviving SKUs that are now absorbing former demand from discontinued items.
Estimating demand transfer before you have new data
The hardest part of recalibration is the period immediately after rationalization, before you have enough new sales data on the surviving SKU set to statistically re-derive demand patterns. I'd approach this in two phases.
In the immediate aftermath, before meaningful new data exists, build an explicit demand transfer estimate rather than just guessing. Look at historical substitution behavior where possible, situations where a SKU was temporarily out of stock in the past and you can observe which alternative SKUs saw a corresponding demand increase during that stockout. That gives you an empirically grounded substitution pattern to apply to the rationalized SKUs, rather than assuming transfer rates based on intuition alone. Where that data doesn't exist, category management or sales input on likely substitution patterns, treated explicitly as an estimate to be revised, is a reasonable starting point.
Setting a defined recalibration checkpoint
I'd set an explicit checkpoint, typically ninety to one hundred twenty days after the rationalization takes effect, to formally recalibrate the forecast model using actual observed data rather than the initial transfer estimates. This isn't a one time correction, it's a planned step built into the rationalization project timeline from the start, not an afterthought triggered only if someone happens to notice forecast accuracy has degraded.
At that checkpoint, I'd specifically compare actual demand for surviving SKUs against both the pre-rationalization baseline forecast and the initial post-rationalization transfer estimate, which tells you not just whether the forecast needs adjustment but whether your original transfer assumptions were directionally correct, informing how much to trust similar estimates in future rationalization decisions.
Safety stock implications during the transition
Forecast uncertainty is elevated during this transition window, and I'd treat that explicitly in safety stock policy rather than leaving safety stock levels unchanged and hoping the point forecast is close enough. Temporarily elevated safety stock on the SKUs most likely to be absorbing transferred demand, until the recalibration checkpoint provides real data, is generally a better trade off than risking stockouts on items experiencing genuine but not yet well quantified demand growth.
Watch for false signals in the first few weeks
Immediately after a rationalization takes effect, there's often a short term demand spike on some surviving SKUs as customers who would have bought a discontinued item make a one-time substitution purchase, followed by a settling to a new, different steady state level. I'd caution against reacting too quickly to the first two or three weeks of post-rationalization data as if it represents the new normal demand level, since that initial spike can be misleading if treated as a stable trend rather than a transition artifact.
Building this into future rationalization projects from the start
The plants and organizations that handle this well treat forecast recalibration as a defined phase of the rationalization project itself, with an owner and a timeline, rather than something the demand planning team discovers they need to deal with reactively after the fact. If you're planning a future rationalization effort, I'd build the ninety to one hundred twenty day recalibration checkpoint into the project plan from day one, alongside the discontinuation decisions themselves, rather than treating forecasting as a downstream concern to be addressed once the main project is considered complete.
