Supply chain software helps decide what to produce, buy, store and deliver, when and in what quantity. The term covers several families of tools, from demand forecasting to warehouse execution. This guide covers the planning side: forecasting demand, building the plan, adjusting it. Warehouse (WMS) and transport (TMS) software are covered in other guides.
What supply chain planning covers
Planning starts from demand and goes down to production and purchase orders. Three horizons are usually distinguished:
- Sales and operations planning (S&OP): over 12 to 18 months, it aligns sales, production, purchasing and finance on a single plan in volumes, reviewed every month.
- Master production schedule and material requirements: over a few weeks to a few months, it turns the plan into quantities per product, then into component and material requirements (MRP).
- Execution (S&OE): over the coming days and weeks, it adjusts the plan to disruptions (a late supplier, an urgent order, a machine breakdown).
One distinction matters when choosing methods: independent demand comes from the market and is forecast (sales of a finished product); dependent demand is calculated from it (that product's components, through the bill of materials).
Families of supply chain software
- ERP: holds orders, stock, bills of materials and material requirements planning (MRP). It is the reference database, but usually plans at infinite capacity and forecasts little.
- Demand forecasting software: produces sales forecasts by product, customer or site, from history and external information.
- S&OP or IBP (integrated business planning): supports the monthly sales and operations planning process, with scenarios and a financial view.
- APS (advanced planning and scheduling): calculates a plan that respects real constraints (machine capacity, labour, materials, supplier lead times), from the master schedule down to scheduling.
- DRP (distribution requirements planning): derived from MRP, it extends requirements planning to the distribution network, warehouse by warehouse.
- Control tower: gives end-to-end visibility of flows and stock, and flags gaps before they become stock-outs.
Large vendors offer suites covering several families. Since 18/03/2026, Gartner has published two separate rankings of planning tools, one for discrete industries and one for process industries, according to vendor press releases.
Demand forecasting methods
- Qualitative methods: sales team input, expert panels, customer information. They help when history is missing (a product launch) or to correct a calculated figure.
- Moving average: the forecast is the average of the last periods. Simple, but slow to react to change.
- Exponential smoothing: the next forecast combines the latest actual and the latest forecast, with a coefficient between 0 and 1 that sets responsiveness (Fn+1 = a × An + (1 − a) × Fn).
- Triple exponential smoothing (Holt-Winters): adds trend and seasonality, in additive or multiplicative form.
- Statistical models and machine learning: regressions, ARIMA models, then models that include external variables (promotions, weather, prices, market signals).
No method wins everywhere. Forecasting software tests several models on each product's history and keeps the one with the lowest error.
Measuring forecast error
A forecast is judged by its error, measured every month:
- Mean absolute error (MAE): the average gap between forecast and actual, in the product's unit.
- Root mean squared error (RMSE): penalises large gaps more.
- Mean absolute percentage error (MAPE): easy to read, but meaningless when actuals are zero or very low, which is common for spare parts.
- Mean absolute scaled error (MASE): compares the model with a naive forecast and works even with zero sales.
Tracking bias as well, that is whether forecasts are repeatedly too high or too low, prevents stock from building up unnoticed.
The bullwhip effect
A small change in end demand grows at each tier of the chain: the distributor orders a little more as a precaution, the manufacturer more still, the component supplier even more. This is the bullwhip effect. The known remedies: share real demand with suppliers, cut lead times and batch sizes, plan together rather than separately. Software alone does not prevent it: partners have to agree to exchange their data.
What manufacturers said at Tech for Industry Show
At Tech for Industry Show 2026, at Paris Expo Porte de Versailles:
- L'Oréal presented an AI sales forecasting solution deployed worldwide, which picks up weak signals to secure ingredients and produce faster when a product goes viral. Read the article
- Nestlé described its first predictive models, launched in 2019 on coffee, cocoa and sugar prices: after a year of comparison, the machine forecast better than the experts, and the organisation was redesigned. Renault described its control towers, covering more than 40,000 suppliers at every tier. Read the article
- Accenture showed agents that recalculate planning parameters (order frequencies, safety stock) as conditions change, with humans keeping control of major decisions. Read the article
- No Chain Technologies and Argon & Co highlighted the "last missing part" in aerospace: a single missing part blocks assembly and ties up the stock of all the others. Read the article
The round tables are available as replays: Supply chain: data, AI and resilience, Resilient and autonomous supply chains in the AI era.
How to choose supply chain software
- Start from the problem: stock-outs, overstock, an unstable plan, poor visibility of suppliers. Each problem calls for a different family of tools.
- Check data quality: sales history, bills of materials, lead times, capacities. A planning tool is only as good as its input data.
- Check the links with the ERP and other tools (MES, WMS): direction of exchanges, frequency, which system is the reference.
- Test on your own data: ask the vendor to replay a past period and compare its forecast error with today's.
- Plan the process as much as the tool: an S&OP without a monthly meeting and decision-makers at the table remains a spreadsheet.
- Compare prices over time: subscription, number of users, integration, training. Few vendors publish their prices. Example noted on 04/11/2026: Forecast Pro, a forecasting tool, lists 199 dollars per user per month for its Professional edition, billed annually with a 12-month commitment; collaborative editions are priced on quotation.
2027 exhibitors equipping supply chain planning
To meet at Tech for Industry Show 2027 (in alphabetical order):
- Celonis: process analysis from company system data, including the supply chain;
- Elisa Industriq: demand and inventory management, S&OP, planning, scheduling;
- KBRW Adventure: flow orchestration and real-time supply chain execution.
Frequently asked questions
What is supply chain software?
A tool that helps decide what to produce, buy, store and deliver. The term covers different families: forecasting, S&OP, APS, DRP, control tower, and for execution WMS and TMS.
What is the difference between an ERP and an APS?
The ERP records orders and stock and calculates requirements without considering capacity; the APS calculates a feasible plan under capacity, material and lead-time constraints.
Which demand forecasting method should you choose?
The one with the lowest error on the product's history. Exponential smoothing suits many products; Holt-Winters adds trend and seasonality; machine learning helps when external variables matter.
How do you measure a forecast?
By its error (MAE, MAPE or MASE) and its bias, tracked every month product by product.
Do you need AI to forecast well?
No. Statistical methods are enough for many stable products. AI helps when demand depends on many factors or when weak signals need to be captured.
