INVO
Demand forecasting

Forecast demand before you turn it into production and purchasing.

Forecast requirements for every product, customer and site. Factory Brain reads historical data, orders, seasonality and business signals, then converts the forecast into production volume and into ingredient and packaging requirements.

What the module does

Turn sales history and market signalsinto operational decisions.

INVO combines historical data with current orders, the calendar, promotions, prices, inventory and the other factors that move demand. It produces a forecast for the days and periods ahead, and every new piece of information updates it. The forecast becomes the shared starting point for sales, purchasing, inventory and production.

Module capabilities

  • Demand forecast by product, customer and site
  • Trend, seasonality and outlier analysis
  • Best model variant selected automatically
  • Forecast accuracy monitoring
  • Demand converted into a production plan
  • Production plan converted into purchasing needs

Main flow

  1. History and signals
  2. Forecast models
  3. Demand forecast
  4. Production plan
  5. Material requirement
  6. Purchasing
Forecast dashboard

Every forecast, variance and risk in one view.

The dashboard shows forecast demand, confirmed orders, the level of uncertainty and the gap between the forecast and what actually happened. You see the products most at risk of overproduction, shortage or lost sales.

  • Demand forecast for the chosen horizon
  • Confirmed and expected orders
  • Forecast accuracy
  • Largest variances
  • Overproduction and shortage risk
  • Effect of the forecast on production and purchasing
Forecast dashboard
Live
Forecast demand
184,200 units
Confirmed orders
142,800 units
Forecast accuracy
91.6%
Products at risk of shortage
8
Products that need attentionOverproduction risk 3.2%
  • Chicken tikka masala

    Raise the plan

    Forecast 12,400 · sold 13,900

    Forecast error −10.8%

  • Teriyaki bowl

    No change

    Forecast 9,200 · sold 8,700

    Forecast error +5.7%

  • Premium salad

    Lower the plan

    Forecast 4,600 · sold 3,100

    Overproduction risk

  • Bolognese pasta

    No change

    Forecast 7,800 · sold 7,640

    Forecast error +2.1%

Model inputs

Build the forecast on the full picture of demand.

Factory Brain does not read sales history alone. The model can take in the operational and business data that explains why demand is rising, falling or changing shape. It also marks the periods when sales were capped by a stockout, so a temporary shortage is not mistaken for a drop in real demand.

  • Sales, order and production history
  • Day of week, month, season and holidays
  • Customers, sales channels and sites
  • Promotions, prices and range changes
  • Inventory, stockouts, returns and waste
  • Weather, events and other external variables
Data sources
  • Sales
  • Orders
  • Calendar
  • Weather
  • Promotions
  • Prices
  • Inventory
  • Returns
  • Waste

Event markers on the series

Promotion
12–18.08
Holiday
15.08
Price change
01.09
Stockout
22–24.08
The stockout window is marked so it cannot drag the calculated demand down.

Forecast model layer

Several models,one best forecast for the case at hand.

No single model works equally well for every product. INVO looks at the length of history, seasonality, trend, sales frequency and demand volatility. It then builds several forecast variants, tests them against historical data, and picks the approach with the best accuracy and stability.

  • Baseline models

    Use the last known value, the equivalent day or the equivalent period as a reference point. They are the benchmark the more advanced methods have to beat.

  • Trend and seasonality models

    Read repeating daily, weekly, monthly and yearly patterns along with long-term shifts in the level of demand.

  • Models with extra variables

    Take in promotions, prices, weather, holidays, events, customers, channels and the other factors that move sales.

  • Intermittent demand models

    Handle products ordered rarely, at irregular intervals or in variable volumes.

  • Hierarchical models

    Produce consistent forecasts at every level: SKU, category, customer, channel, site and the whole organization.

  • Ensemble models

    Combine the results of several approaches where a set of models forecasts more stably than any one of them.

Best model for SKU-2842
  • Trend and seasonality modelerror 11.8%
  • Model with extra variableserror 7.4%
  • Ensemble modelerror 6.9%
Variant selectedEnsemble model
Backtesting and model selection

Choose the model on results, not on assumptions.

INVO tests the forecast variants against historical data. It replays earlier forecasting moments and checks how each model would have done at predicting the demand that actually arrived. A model is judged not only on average error but on the stability and consistency of that error — under-forecasting causes shortages, over-forecasting causes overproduction and waste.

  • WAPE and MAPE
  • MAE and RMSE
  • Forecast bias
  • Frequency of under-forecasting
  • Frequency of over-forecasting
  • Stability across consecutive periods
Model comparison
WAPEBiasShortage risk
Baseline model18.4%−7.2%HighRejected
Trend and seasonality11.8%−1.9%MediumAlternative
Model with extra variables7.4%0.8%LowGood
Ensemble model6.9%0.3%LowSelected
Multi-level forecasting

Forecast demand at the level where you make the decision.

INVO can forecast for a single SKU, a whole category, a customer, a sales channel, a site or a plant. The forecasts stay consistent with each other, so the sum of the product forecasts matches the forecast for the category or the organization.

  • Product and product variant
  • Customer and customer group
  • Sales channel
  • Site and region
  • Product category
  • The whole organization
Forecast hierarchy
  • Organization184,200 units
  • Central region72,400 units
  • Customer · hospital group28,900 units
  • Category · ready meals16,200 units
  • Product · Chicken tikka masala12,400 units
  • SKU-2842 · 420 g tray12,400 units
The sum of the lower level always reconciles to the level above it.

Forecast horizons

Match the forecast to the decisionyou have to make.

The system can produce forecasts for several horizons. A short-term forecast supports production happening now; a longer horizon helps plan purchasing, capacity and supplier relationships.

  1. Next day

    Supports decisions about production, picking and packing happening now.

  2. Next 7 days

    Planning of shifts, lines, fresh ingredients and the nearest deliveries.

  3. Next 30 days

    Planning of purchasing, inventory, labor and available capacity.

  4. Next 60 and 90 days

    Long-term planning of supply, ingredient contracting and plant utilization.

From forecast to production

Convert forecast demand into real production volume.

INVO combines the forecast with confirmed orders, finished goods on hand, work in progress, safety stock and shelf life. That way a sales forecast does not automatically become a production plan — the system first checks how much product is already available. The overlap between the forecast and confirmed orders is removed so the same demand is not counted twice.

  • Forecast demand adjusted for orders
  • Target safety stock
  • Finished goods on hand
  • Production already in progress
  • Shelf life and freshness
  • Double counting of demand removed
Production requirement
Chicken tikka masala
  • Forecast demand12,400 units
  • Orders already in the forecast8,900 units
  • Target buffer800 units
  • Inventory on hand−2,100 units
  • Work in progress−1,000 units
To produce10,100 units
From production to purchasing

Convert the production plan into ingredients and purchase orders.

The purchasing requirement is not a separate, independent forecast. INVO runs the planned production volumes through the recipes, sub-assemblies, yield factors and process losses. It then compares the material need against inventory on hand, reservations, confirmed deliveries, buffers and supplier lead time.

  • Recipe and per-unit requirement
  • Planned losses and safety buffer
  • Available and unreserved inventory
  • Confirmed deliveries
  • Minimum order quantities and pack sizes
  • Shelf life and supplier calendars
From plan to purchasing
10,100 packs of product

Material requirement

  • Chicken breast1,696 kg
  • Basmati rice929 kg
  • Tomato sauce747 kg
  • Coconut milk384 kg

Recommended purchases

  • Chicken breastorder 620 kg
  • Coconut milkorder 160 kg
  • Other ingredientsno extra orders
Forecast intervals

Plan for a range of risk, not just a single number.

A forecast is not one certain figure. INVO can present a base scenario together with the range demand could fall in. That makes it possible to size the production and purchasing buffer to the level of risk and to the product's shelf life. For fresh products the system can recommend a tighter buffer to limit waste; for products with a high service level, more safety stock.

  • Low, base and high scenario
  • Recommended production volume
  • Shortage risk for each scenario
  • Overproduction and waste risk
  • Buffer sized to shelf life
  • Customer service level as a parameter
Demand scenarios
SKU-2842
DemandShortage riskOverproduction risk
Low scenario10,800 unitsHighLow
Base forecast12,400 unitsMediumMedium
High scenario14,100 unitsLowHigh
Recommended production12,900 units
Accuracy monitoring

See where the forecast works and where it needs improving.

INVO compares the forecast against actual orders, sales and production. Accuracy can be analyzed by product, customer, channel, site and horizon. The system also shows the direction of the error, so you know whether a model consistently over-forecasts and drives overproduction, or under-forecasts and causes shortages.

  • Accuracy by SKU and category
  • Comparison across horizons
  • Over- and under-forecasting analysis
  • Products with the largest error
  • Effect of error on waste and lost sales
  • History of model changes
Accuracy matrix
Forecast error by horizon
1 day7 days30 days60 days
Ready meals4.2%6.8%11.4%16.1%
Salads5.1%9.3%14.8%21.6%
Soups3.8%5.9%9.2%13.4%
Desserts6.4%11.2%18.7%24.9%
Selecting a product shows the history of the forecast against actual demand.
Factory Brain for demand

Spot a change in demand and update the operational decisions.

Factory Brain watches new orders, actual sales, the rate inventory is being drawn down, and the events that move demand. When the situation diverges from the earlier forecast, it recalculates and shows the effect on production, purchasing and ingredient availability.

  • Unusual changes in demand detected
  • Forecast updated as new orders arrive
  • Shortage and overproduction risk predicted
  • Production adjustment recommended
  • Purchasing adjustment recommended
  • The factors behind the change explained
Factory Brain · Demand shift

Change detected

Demand for SKU-2842 in the central region is growing faster than the forecast assumed.

New forecast

Previous forecast: 12,400 units · Current forecast: 14,100 units · Change: +13.7%

Impact

  • Production: raise the plan by 1,700 units
  • Chicken breast: 286 kg more
  • Packaging: 1,700 units more
  • Shortage risk if nothing changes: high
Update the planShow what drove the change

Integrations

Connect the forecast to sales,production, purchasing and inventory.

INVO works from data in sales systems, the ERP, CRM, inventory and production. The current forecast feeds production planning, MRP, purchasing and inventory management.

  • Sales, CRM and orders
  • ERP and historical data
  • Recipes and food cost
  • Production planning
  • Purchasing and suppliers
  • Inventory, MES and finance
Forecast data flow
  • Sales
  • CRM
  • ERP
  • Recipes
  • Planning
  • Purchasing
  • WMS
  • MES
Historical and current data go into the forecast; production and purchasing decisions come out of it.

One source of truth for what will be needed.

Sales sees expected demand, production knows how much to prepare, purchasing knows what ingredients to order, and inventory can react to shortage and surplus risk before it arrives.

Forecast demand first. Then let production and purchasing recalculate themselves.