DEMDemandStop 11 of 12
- Forecasting fundamentals
- Demand patterns & classification
- Exponential smoothing & ETS
- Intermittent demand: Croston, SBA, TSB
- ML & global forecasting models
- Forecast accuracy & bias
- Probabilistic forecasting
- Hierarchical forecasting
- Promotions, cannibalization & halo
- New products & phase-outs
- Demand sensing & shaping
- Consensus forecasting & FVA
Demand sensing & shaping
Reading the latest short-term signals to correct the near-term forecast, and using price and promotion to steer demand toward supply.
Read atChange any time
These two sound alike but do opposite jobs.
Demand sensing answers: what is changing right now? A monthly forecast built on two years of history might say 100,000 units next month. Demand sensing looks at the latest signals, such as daily store sales (POS), distributor orders, retail stock levels, online searches, weather and local events, and corrects the next few days or weeks. If a heatwave hits and retailer stock of a sports drink is falling fast, the near-term forecast might move to 110,000. That update flows straight into replenishment, production and trucks. Sensing improves the short horizon; it doesn’t replace the long-range forecast.
Demand shaping answers: how can we influence demand to fit what we can supply? Instead of only predicting, the company pulls levers:
- a discount on an overstocked item, or a promotion moved to a week with spare capacity,
- recommending an available substitute when a product is short,
- offering a cheaper slower delivery slot when the fast lane is full,
- adjusting prices when demand runs ahead of supply.
Shaping needs to know how customers respond, such as price elasticity and promotion lift, and it needs supply, sales and finance working together, because every lever has a cost.
Words to use
- POS data
- Point-of-sale data: what actually sold at the till, store by store, day by day.
- Price elasticity
- How much demand changes in percentage terms when price changes by 1%.
- Short-term horizon
- The next few days to weeks, where sensing adds the most accuracy.
How demand sensing works
- Start from the baseline forecast produced by the regular process.
- Collect fast signals for each item and location: daily point-of-sale data, open customer orders, customer stock levels, shipments, weather forecasts, local events, and online search or traffic.
- Learn the patterns that link those signals to the forecast error of the next few days or weeks, usually with a regression or machine-learning model that weights recent days most.
- Adjust the near-term forecast, often one to four weeks ahead, and pass the change to deployment, replenishment and transport.
- Measure the value: compare the sensed forecast with the baseline at short lags. If it isn’t more accurate, it isn’t helping.
The baseline forecast for next week is 25,000 units in one region. Then:
- Store sales over the last three days run 18% above forecast.
- Retailer warehouse stock has fallen to 4 days of cover against a normal 7.
- The weather service predicts temperatures 3°C above normal next week.
A sensing model that has learned how such signals relate to forecast errors might lift next week to about 27,500 units, and deployment sends the extra 2,500 to that region before shelves empty. The factory’s monthly plan doesn’t change unless the signal persists.
These numbers are illustrative: the size of a real adjustment comes from how well the signals predicted errors in the past.
How demand shaping is sized
Shaping uses measured customer responses, such as price elasticity, to move demand:
- An overstocked item sells 10,000 units a week. With a price elasticity of −2, a 10% discount lifts sales by about 20%, to 12,000 a week, clearing 2,000 extra units a week. Whether that pays depends on the margin given up against the holding and obsolescence cost avoided.
- When an item is short, recommending an available substitute online, or nudging customers to another delivery slot, moves demand to where supply is.
Every lever has a cost, so shaping decisions need sales, finance and supply in the same conversation.
Why sensing helps only in the short term
Recent signals carry information about the next few days that a monthly statistical forecast doesn’t have: a heatwave, a competitor’s stockout, a viral post. That information fades fast. A spike in this week’s store sales says a lot about next week and little about next quarter. So sensing improves accuracy most at the shortest lags, and it should feed decisions with short lead times, such as deployment, picking, transport and short production runs, not long-range capacity plans.
The data it needs
- Downstream data: retailer sales and stock data, shared through portals or collaborative arrangements such as CPFR (collaborative planning, forecasting and replenishment).
- Low latency: a signal that arrives a week late is history, not sensing.
- Clean mapping: the retailer’s item codes, stores and calendar must map to the supplier’s products and locations.
Research comparing retailer-order forecasts built from point-of-sale data with those built from order history (Williams and Waller, 2010) found that POS data often improves them, but not in every case, which is why sensing should be tested by product group and horizon rather than assumed to help everywhere.
Overreaction and double counting
- Chasing noise: a sensing model that reacts to every blip creates nervous plans and extra transport. Use thresholds and act only on changes that matter.
- Double counting: if planners also adjust for the heatwave they read about, the forecast moves twice. Agree who adjusts what, and when.
- Feedback loops: shaping changes demand, and sensing then sees the result as a signal. Record shaping actions so models don’t learn from your own interventions.
Shaping with care
- Training customers to wait: frequent discounts teach customers to buy only on deals, which lowers the baseline and raises volatility.
- Brand and fairness: prices that change too often, or that seem to exploit a shortage, damage trust and in some markets attract regulators.
- Bullwhip: promotions to clear stock create spikes and dips upstream. Shaping should smooth demand, not add swings.
Interview kit
Practise it out loud
Answer each question yourself before opening the model answer.
Demand sensing uses recent, high-frequency signals such as POS, orders, retailer stock, weather and events to correct the short-term forecast faster than history-based models can. Demand shaping is the active side: using price, promotions, substitution and delivery options to steer demand toward available supply. Sensing tells you what’s changing; shaping changes what customers do; and both feed supply and inventory decisions.
What is the difference between demand sensing and demand shaping?
Sensing is detection: it updates the near-term forecast using the latest signals. Shaping is influence: it uses commercial levers such as price, promotion and substitution to move demand toward what the supply chain can deliver profitably.
Give an example of demand shaping helping the supply chain.
A plant is short of capacity for the 1 litre pack but has plenty of 500 ml. Running a promotion on the 500 ml, or recommending it as a substitute online, shifts some demand to the available product and avoids stockouts and expensive overtime on the constrained line.
Connected stops
Where this idea feeds in, or gets its inputs from.
Sources and further reading
- Chopra, S. & Meindl, P. Supply Chain Management: Strategy, Planning, and Operation. Pearson. Chapters on demand forecasting and on pricing and revenue management.
- Hyndman, R. J. & Athanasopoulos, G. Forecasting: Principles and Practice (3rd ed.). OTexts.
- Williams, B. D., & Waller, M. A. (2010). Creating order forecasts: Point-of-sale or order history? Journal of Business Logistics, 31(2), 231–251.