Demand sensing is a collection of next-generation forecasting methods that leverages new mathematical techniques and real-time (or near real-time) data, including social data, leading indicators, and more to create an accurate forecast of demand, usually based on constraints within the supply chain. It gives you the ability to incorporate real-time demand, or “new” data (sometimes from POS locations), into short-term forecasts. The results can be quite impactful to downstream operations when implemented correctly as part of an overall forecasting framework. The goals are to identify trends earlier, detect potential under-performance issues, and to tighten plan-versus-actual performance.
By leveraging data daily, new demand signals may be uncovered quickly. These demand signals are generally compared to the statistical demand forecast to determine the significance of the signal. All too often companies believe implementing demand sensing alone will solve their problems, however, it can create more problems if done in a silo.
As with many systems (BI, Analytics, ERP, etc.), it comes down to the data. Garbage in, garbage out. Ensuring you have clean data is critical. Ensuring you understand your data is also important. When leveraging demand sensing technology, following are some examples of data sources:
- Shipment/fulfillment data
- POS (point of sale) data
- EDI transaction data
- Social media data
- Leading indicators
Remember, if you have stockouts, typical demand sensing solutions will want to include this data and not exclude it as some demand management solutions do.
Demand sensing vs. demand management
Demand sensing focuses on the short-term changes in demand and enables the forecaster to focus on these signals to make decisions around the tactical redistribution of inventory. Demand management looks at the supply chain holistically and typically does not focus on day-to-day fluctuations at a store level. Demand sensing generally uses data in an hourly time frame, where demand management may use it on a weekly or monthly level.
The typical performance of demand sensing systems can reduce near-term forecast error by 30% or more compared to traditional time-series forecasting techniques. This is true only if implemented as part of an overall framework for forecasting and performed by a company that has a deep library of models to perform extensive back-testing.
So, is demand sensing hype or hot? The answer is that it is still too early to tell. For retailers who sell high volume goods that go in and out of fashion quickly, this technique can provide value as part of a larger framework. For other companies, the value may not be there just yet.