Chocolate Factory in the Rhythm of the Season | Module 1 of 9

The Seasonal Volume:
How much is produced before the peak

A large share of annual revenue with pralines and seasonal figures falls into a few weeks. Whoever plans on last year plus gut feeling produces too much in January and delivers too little in December. A model on sell-in, sell-out and the promotion calendar forecasts demand per item, channel and week.

Model calculation - fictitious sample company

This module is part of a nine-part model calculation on a fictitious mid-sized chocolate manufacturer (€95 million net revenue, 140 items, 4 lines). All euro amounts are derived from the documented assumptions of the sample company. They are not results of real customers and not a performance promise. Which values are achievable on your data is shown only by a pilot.

01 The problem - seasonality that wrecks plans

Three peaks define the production year, and every peak is different

Chocolate is among the most seasonal foods: Christmas, Easter, Valentine's Day. The Easter date moves, March weather affects bar sales, and one retailer promotion shifts the whole mix. With 140 items this cannot be planned in a spreadsheet.

Today sales delivers a forecast from last year plus a mark-up, production corrects at its own discretion, and the warehouse receives the plan shortly before the peak. The result: overproduction of items nobody orders, and shelf gaps for those the retailer is waiting for.

Cost of the forecast error in the sample company: €1,960,000 per year

€570,000 write-offs after the best-before date, €1,140,000 revenue lost through shelf gaps, €250,000 extra shifts and express freight

Sell-in and sell-out
Season position and promotions
Gradient boosting
Forecast 12 weeks
Production plan

02 The model - one model per horizon, reconciled across the hierarchy

Gradient boosting with quantiles at item, channel and week level, reconciled to category and total

▸ Example values - illustration of the method, not an analysis carried out
Backtest 12 weeks, weighted forecast error: 9.4 %
Deviation of the last-year method in the same window: 21.8 %
Seasonal profile of three categories - pralines, seasonal figures, bars (index, annual mean = 100)
↳ Three different worlds

Pralines are gift goods with peaks at Christmas and Valentine's Day. Seasonal figures live from two week-windows in spring and Advent. Bars run almost flat with a summer dip. One model for all categories falls short; the model learns each category's profile and reconciles item forecasts to category and total. External signals such as weather or social media mentions are features among many, not a module of their own.

03 Business impact - fewer write-offs, fewer gaps, less rush

Three levers from one forecast

€331,500
Savings / year
−30 %
Fewer write-offs
€85,500
Recovered contribution margin
Three levers - base today and savings with model
Model calculation · How the amount arises - derived in the open

Every assumption comes from the sample company and is stored centrally. With your figures the input changes, not the method.

PositionValue
Write-offs after the best-before date / year€570,000
Lever: Less overproduction (30 %)€171,000
Revenue lost through shelf gaps / year€1,140,000
Lever: Recovery (25 %) at contribution margin (30 %)€85,500
Extra shifts and express freight / year€250,000
Lever: Fewer rush costs (30 %)€75,000
Result: savings / year€331,500

Assumptions of a sample company - in a real project your data replaces these values.

↳ The quiet lever: reliability

The biggest gain is not in the table: the trust of the retail chains. Whoever delivers on time and in full at the peak holds the better cards in the conversation about the secondary placement. Whoever fails leaves the space to the competition. The forecast is the foundation of the retail relationship, and module 02 builds on it.

04 Next steps in your factory

From the model to the weekly planning template

① Data export

Three years of sell-in from the ERP, sell-out of the main chains and the promotion calendar. Raw format is enough; cleaning is part of the pilot.

② Pilot per category

One model for the most seasonal category. In the backtest, the retrospective check on your own previous years, you see the forecast error against your current method, week by week.

③ Planning template

Automatic forecast every Monday with an uncertainty band, a note on deviation, hand-over to production planning.

All 9 modules: AI in the chocolate factory