Demand planning: a reliable, traceable sales forecast shared by the whole supply chain
The demand planner produces the sales forecast that feeds replenishment, production and the budget. They start from history, clean it of stockouts and exceptional events, then enrich it with what sales and marketing know. Most of their time still goes into collecting and cleaning data, at the expense of analysing variances.
Between sales, marketing, the supply chain and finance
The demand planner prepares a baseline statistical forecast by item, by customer or by site, then leads the demand review with sales and marketing, who bring promotions, launches, listings and delistings. They pass a consolidated forecast to replenishment and production, and a value view to management control. Their value lies in the forecast’s accuracy over time and in their ability to explain why it was wrong.
A typical day.
- MorningCheck yesterday’s or last week’s sales and spot marked variances from the forecast, by family and by customer.
- MorningAlert replenishment to items whose demand is dropping off or surging, before stock shows it.
- MiddayBring in information from sales and marketing: confirmed promotions, new listings, launches, discontinued products.
- AfternoonCorrect sales history for stockouts and exceptional events to recover true demand.
- Every weekRerun the statistical forecast, review items outside tolerance and circulate the updated forecast.
- Month endPrepare the S&OP demand review: forecast accuracy, bias, assumptions, variances against budget.
What the role is measured on.
- Forecast accuracy by family and by horizon
- Forecast bias (persistent over- or under-forecasting)
- Value added by manual adjustments
- Share of tracked items outside tolerance
- Variance between volume forecast and value budget
- Adherence to the demand review calendar
Indicators are named, never quantified: levels depend on each company, and the diagnostic measures them.
The problems that come up most.
Polluted history
Stockouts, promotions, store openings and exceptional events stay in the history and distort the statistical forecast if they are not corrected.
Commercial information that arrives late
Promotions, listings and launches are known to sales people well before they reach the demand planner, by e-mail or in a file.
Adjustments that are never measured
The forecast is reworked at every review with no record of who changed what, and no measure of the effect of those changes on accuracy.
New products with no history
Launches are forecast on intuition, for lack of a shared method to build on comparable products.
Forecast levels that do not talk to each other
Sales forecast by customer, supply forecast by item and site, value budget by family: the three are reconciled by hand.
Too many items to review
The catalogue is too large for a detailed review of every item; time gets scattered instead of going to the items that matter.
One role, several realities.
- Fashion and luxury
In fashion and luxury, the demand planner separates continuity lines, forecast on a flow basis from history, from collection newness, forecast by analogy with comparable styles and then readjusted from the first sales. The forecast is broken down by boutique, by channel and by size, with strong seasonality.
- Retail
In retail, forecasting is done by store and by item, often weekly, with strong effects from promotions, weather, the calendar and store openings or closures. The difficulty is volume: a very large number of store-item pairs to follow.
- Cosmetics and beauty
In cosmetics and beauty, demand is driven by launches, seasonal gift sets and operations with retailers. Retailers’ sell-out data, when shared, shows what really sells on the shelf beyond the orders received.
- B2B distribution and wholesale
In B2B distribution and wholesale, forecasting relies on customer order history, contracts and sales people’s information on key accounts. Often irregular demand calls for methods suited to slow-moving items.
- Industry
In manufacturing, the sales forecast feeds the MPS and, through it, the MRP. It combines firm orders, customer forecasts sent by EDI or portal and a statistical forecast on catalogue items, over a horizon consistent with production lead times.
- Pharma and healthcare
In pharma and healthcare, forecasting takes account of hospital tenders, reimbursement changes, generics and critical products whose availability is monitored. It is often built country by country, each with its own registration constraints.
Automate the task, keep the judgement.
What gets automated
- Collecting sales, orders and sell-out shared by retailers, consolidated daily from the ERP, the tills and the portals.
- Cleaning history: detecting stockouts, promotions and outliers, and proposing corrected demand.
- The baseline statistical forecast by item, customer or site, with the method chosen according to the demand profile, in a forecasting tool connected to the ERP.
- Continuous measurement of accuracy and bias, and of the value added by each manual adjustment.
- Preparing the demand review: variances, assumptions and items outside tolerance assembled automatically, then given a first commentary by an AI agent.
What stays human
- Bringing in market information the data does not contain: a negotiation under way, a competitor, a regulatory change.
- Choosing launch and promotion assumptions, and arbitrating between the sales forecast and the statistical forecast.
- The final commitment on the forecast, made collectively in the demand review and at S&OP.
How to automate, case by case.
Each case starts from a question leadership teams ask. The starting point, what gets automated, the tools involved, and what stays in the teams’ hands.
How do you automate sales forecasting?
- Today
- The demand planner extracts sales history from the ERP, assembles it in a spreadsheet and extends trends item by item. The forecast is rebuilt at every cycle, depends on the person building it and only covers the main items in detail.
- What gets automated
- A forecasting tool connected to the ERP retrieves history every week, chooses for each item a method suited to its demand profile and calculates a baseline forecast over the useful horizon. Items outside tolerance are isolated for review; the others go through without intervention. The approved forecast is sent back to the ERP and to the replenishment tool.
- What stays human
- Approving forecasts for important items and special cases stays with the demand planner.
- Forecasting tool
- ERP
- Business intelligence (BI) tool
How do you automate the measurement of forecast accuracy and bias?
- Today
- Forecast accuracy is calculated now and then, on a selection of items, with a formula that varies from person to person. Nobody can say whether the forecast is improving, or where it is most wrong.
- What gets automated
- Every forecast version is kept, and the tool automatically calculates error and bias every period, by family, customer, site and horizon. It separates the statistical forecast from the adjusted forecast and measures the value added by each adjustment step. The results feed the demand review.
- What stays human
- Choosing which measures to track and the actions on the least accurate families are decided by the demand planner with supply chain management.
- Forecasting tool
- ERP
- Business intelligence (BI) tool
How do you automate the inclusion of promotions in the sales forecast?
- Today
- Promotions are decided by sales or marketing and passed on by e-mail or in a file, sometimes late. The demand planner adds an extra volume by hand, with no measure of the real effect of past operations.
- What gets automated
- Operations are entered in a shared promotional calendar connected to the forecasting tool. The tool estimates the sales uplift from comparable past operations, taking into account the mechanism, the discount and the knock-on effect on the following weeks. After each operation, the variance between forecast and actual enriches the promotion history.
- What stays human
- Volume assumptions for a new kind of operation, with no past equivalent, are approved by the demand planner with sales.
- Forecasting tool
- Promotional calendar
- ERP
- Business intelligence (BI) tool
How do you automate forecasting for new products with no history?
- Today
- A launch is forecast from the product manager’s opinion and a few items judged comparable, chosen by hand. The first sales are not always used quickly enough to correct the forecast.
- What gets automated
- The tool proposes analogue products according to their attributes (family, price, channel, season) and builds a launch curve from their sales. From the first sales, the forecast is recalibrated automatically, and significant variances are flagged to the demand planner and to replenishment.
- What stays human
- Choosing the analogues and the launch volume are approved by the demand planner with marketing and the product manager.
- Forecasting tool
- Product master data
- ERP
How do you automate the cleansing of sales history?
- Today
- Stockouts, promotions, store openings, a customer’s one-off order: these events stay in the history and distort the forecast. Correcting them takes line-by-line work that the team only does on the main items.
- What gets automated
- The tool cross-checks sales, stock and the operations calendar to detect stockout periods, promotions and outliers. It proposes corrected, so-called unconstrained, demand, with the reason for each correction. The demand planner approves in batches and only reviews ambiguous cases in detail.
- What stays human
- Classifying an event as exceptional or lasting remains the demand planner’s decision.
- Forecasting tool
- ERP
- WMS
- Promotional calendar
How do you automate the collection of sales team forecasts?
- Today
- Every month, sales people send back their forecasts by customer in files of varying formats and completeness. The demand planner chases them, consolidates them and compares them with the statistical forecast by hand.
- What gets automated
- Each sales person receives, for their customers, a forecast pre-filled from the tool, which they adjust in a simple form with a reason for each change. Responses are consolidated automatically, reminders go out on their own, and significant variances from the statistical forecast are flagged for the demand review.
- What stays human
- Arbitrating between the sales person’s forecast and the statistical forecast stays with the demand planner, in the demand review.
- Forecasting tool
- CRM
- Online form
- Messaging
How do you automate forecasting by store and by size?
- Today
- The forecast is made by item at national level, then the planner splits it between stores and sizes using fixed keys. Stores receive sizes that do not sell there, while others run short.
- What gets automated
- The tool forecasts at the aggregated level, where demand is more stable, then splits it by store and by size according to each store’s recent sales, its seasonality and the size profiles observed. The split keys are recalculated at every cycle instead of staying frozen.
- What stays human
- Rules specific to certain stores (opening, refurbishment, particular clientele) are set by the demand planner with retail operations.
- Forecasting tool
- Till and sales
- ERP
- Business intelligence (BI) tool
How do you automate the detection of shifts in demand trends?
- Today
- A drop or acceleration in demand only becomes visible after several weeks, when the variance is already established and stock bears its mark. Tracking tables are too broad to pick up weak signals.
- What gets automated
- Every week the tool monitors each item’s sales against its forecast and its usual behaviour, and flags variances that exceed normal variability. Alerts are ranked by their effect on stock and service, with a proposed new forecast.
- What stays human
- The decision to revise the forecast, or to wait for confirmation, stays with the demand planner.
- Forecasting tool
- ERP
- Business intelligence (BI) tool
- Messaging
How do you automate the use of retailers’ sell-out data in the forecast?
- Today
- Some retailers share their shelf sales and stock, but in files of different formats, sent at varying intervals. This data remains little used even though it signals upcoming orders.
- What gets automated
- Sell-out files are retrieved automatically from portals or e-mails, read, standardised and mapped to the company’s items. The tool compares sell-out, retailer stock and orders received (sell-in) to anticipate upcoming orders and spot destocking or overstocking at the customer.
- What stays human
- Interpreting variances with the retailer and reflecting them in the forecast stay with the demand planner and the key account manager.
- Retailer portal
- Forecasting tool
- ERP
- AI document-reading agent
How do you automate the reconciliation of volume forecasts with the budget?
- Today
- Supply forecasts in quantities by item; finance budgets in value by family and by month. The two are reconciled by hand before each S&OP, with prices and scopes that differ.
- What gets automated
- The tool values the volume forecast with current prices and discounts, then aggregates it according to the budget’s families and calendar. Variances are presented by family, by channel and by cause: volume, price or mix. Management control receives the same view at every cycle.
- What stays human
- The choice to revise the forecast or the budget, and the explanations given to management, stay with the demand planner and management control.
- Forecasting tool
- ERP
- Business intelligence (BI) tool
What teams look to improve.
- Which forecasting method should be chosen for each item?
- It depends on the demand profile: marked trend and seasonality, stable demand, irregular or intermittent demand. A forecasting tool tests several methods on history and keeps the one that would have performed best, item by item. The demand planner approves the choice on important items and can override it when a change is known to be coming.
- How do you measure forecast accuracy?
- By comparing the forecast frozen at a given horizon with actuals, by family, by customer or by site. Two measures complement each other: average error, which says by how much the forecast is wrong, and bias, which says whether it is always wrong in the same direction. A connected tool calculates these measures automatically every period.
- Should the statistical forecast be adjusted by hand?
- Yes when known information is not in the history: a promotion, a listing, a product being discontinued. No when the adjustment rests on a hunch. Measuring the value added by each adjustment shows which ones actually improve the forecast.
- Does AI forecast better than a classic statistical model?
- Not systematically. It helps most when many variables are at play (price, promotions, weather, calendar) and history is sufficient. The right choice is measured on the company’s own data, by comparing methods over the same period.
- How do you forecast demand when a product is out of stock?
- Sales during a stockout understate true demand. The tool identifies these periods from stock levels and replaces observed sales with an estimate of demand, based on comparable periods and similar items.
- How should the demand review be organised?
- With a fixed calendar, a baseline forecast prepared in advance and a short list of topics: items outside tolerance, new commercial assumptions, variances against budget. The decision and its reasons are recorded, so that its effect can be measured.
- At what level of detail should you forecast?
- At the level where the decision is made: item and site for replenishment, family and month for the budget. The tool forecasts at the most useful level, then aggregates or splits, so that all levels stay consistent with one another.
- Do you need a dedicated forecasting tool or is the ERP enough?
- ERP forecasting features often cover simple needs. A dedicated tool is justified when the catalogue is large, demand is irregular or the drivers are many. It then connects to the ERP, which remains the reference for orders and stock.
Describe what’s holding your teams back.
A first conversation to qualify your need. Then a scoped diagnostic: the scope, the solution, the estimate.
Discuss your projectReply within 48 business hours.