Forecasting and Scenario Planning
Learn to build forecasts you can defend, from beating a simple benchmark to measuring error and bias honestly. No coding: you work on the data of Raket Parts, a fictional maker of rocket parts.
You will be able to
- test a forecast against a simple benchmark
- spot a break in a series and challenge a claimed early signal
- rebuild a forecast from its drivers
- measure forecast error and bias separately
Beat the simple forecast first
Compare honest planning forecasts with a simple benchmark on held-back months and state a cautious range.
Breaks, blips and early signals
Recognise a possible break, compare levels after it and challenge whether a claimed early signal actually leads.
Forecast from the drivers
Rebuild ordered value from its drivers, assign an owner to assumptions and compare the effects of changing them.
How wrong, and which way?
Measure the size and the direction of forecast misses, compare methods on the same months and score a hand adjustment.
Stories that change the decision
Design scenarios that could change a decision, compare every option in every scenario with a regret table and name signposts to watch.
Reviewing a colleague's changed forecast
Check a revised forecast before it is released.
Ranges by horizon and the forecast cycle
Set honest ranges and run the forecast round each month.
Integration clinic: the annual planning round
Bring every method together on one planning round.
Final assessment
A forecast and scenarios on an unseen case.
Forecasting and Scenarios / Module 1
Beat the simple forecast first
Teun is offered a method that matches old sales almost perfectly. Check whether it predicts months it never saw better than the simplest forecast, and show the uncertainty.
Sections
Forecasting and Scenarios / Module 1 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Forecasting and Scenarios / Module 2
Breaks, blips and early signals
Sales rise sharply out of the thin 2024 records. Bram wants to project the steep line; Teun checks whether it is growth, a one-off, or a change in the data.
This module builds on module 1. If you are new to this, start with module 1.
Sections
Forecasting and Scenarios / Module 2 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Forecasting and Scenarios / Module 3
Forecast from the drivers
Bram wants a sales plan he can challenge. Teun separates the assumptions into orders, lines per order and average value per line, then checks the result against history.
This module builds on module 2. If you are new to this, start with module 2.
Sections
Forecasting and Scenarios / Module 3 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Forecasting and Scenarios / Module 4
How wrong, and which way?
Teun has two forecasts for each segment's monthly sales. They can miss by about the same amount while one of them is too high every month. Learn to tell the size of a miss from its direction, and to compare methods fairly.
This module builds on module 3. If you are new to this, start with module 3.
Sections
Forecasting and Scenarios / Module 4 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Forecasting and Scenarios / Module 5
Stories that change the decision
Mira wants to decide how Raket Parts will handle machining demand. Teun compares options across stories that could change the answer.
This module builds on module 4. If you are new to this, start with module 4.
Sections
Forecasting and Scenarios / Module 5 / Free practice
Try it on your own question
Run it in the practice editor
The practice editors have the Raket Parts tables loaded. Type a query or a formula, run it and change it as often as you like. Nothing is marked.
Forecasting and Scenarios / Module 1 / 1.1
1.1 What history can you use?
Section 1.1 complete
Next: 1.2 The naive benchmark
Forecasting and Scenarios / Module 1 / 1.2
1.2 The naive benchmark
Section 1.2 complete
Next: 1.3 Hold some months back
Forecasting and Scenarios / Module 1 / 1.3
1.3 Hold some months back
Section 1.3 complete
Next: 1.4 A range, not one number
Forecasting and Scenarios / Module 1 / 1.4
1.4 A range, not one number
Section 1.4 complete
Next: Module 1 check
Continue to the module check Recap this section Back to module 1
Forecasting and Scenarios / Module 1 / Check
Module 1 check
Module 1 complete
Next: Module 2, Breaks, blips and early signals
Continue to module 2 Try it on your own question Recap module 1 Back to module 1
Forecasting and Scenarios / Module 2 / 2.1
2.1 A jump in the history
Section 2.1 complete
Next: 2.2 Level after the break
Forecasting and Scenarios / Module 2 / 2.2
2.2 Level after the break
Section 2.2 complete
Next: 2.3 One-off or new level?
Forecasting and Scenarios / Module 2 / 2.3
2.3 One-off or new level?
Section 2.3 complete
Next: 2.4 Does the signal come first?
Forecasting and Scenarios / Module 2 / 2.4
2.4 Does the signal come first?
Section 2.4 complete
Next: Module 2 check
Continue to the module check Recap this section Back to module 2
Forecasting and Scenarios / Module 2 / Check
Module 2 check
Module 2 complete
Next: Module 3, Forecast from the drivers
Continue to module 3 Try it on your own question Recap module 2 Back to module 2
Forecasting and Scenarios / Module 3 / 3.1
3.1 The driver chain
Section 3.1 complete
Next: 3.2 Does it reproduce history?
Forecasting and Scenarios / Module 3 / 3.3
3.3 Owners and sources
Section 3.3 complete
Next: 3.4 Which driver matters most?
Forecasting and Scenarios / Module 3 / 3.4
3.4 Which driver matters most?
Section 3.4 complete
Next: Module 3 check
Continue to the module check Recap this section Back to module 3
Forecasting and Scenarios / Module 3 / Check
Module 3 check
Module 3 complete
Next: Module 4, How wrong, and which way?
Continue to module 4 Try it on your own question Recap module 3 Back to module 3
Forecasting and Scenarios / Module 4 / 4.1
4.1 Same average miss, different story
Forecasting and Scenarios / Module 4 / 4.2
4.2 Size and direction
Section 4.2 complete
Next: 4.3 Compare on the same months
Forecasting and Scenarios / Module 4 / 4.3
4.3 Compare on the same months
Section 4.3 complete
Next: 4.4 Did the hand adjustments help?
Forecasting and Scenarios / Module 4 / 4.4
4.4 Did the hand adjustments help?
Section 4.4 complete
Next: Module 4 check
Continue to the module check Recap this section Back to module 4
Forecasting and Scenarios / Module 4 / Check
Module 4 check
Module 4 complete
Next: Module 5, Stories that change the decision
Continue to module 5 Try it on your own question Recap module 4 Back to module 4
Forecasting and Scenarios / Module 5 / 5.1
5.1 High, middle, low changes nothing
Section 5.1 complete
Next: 5.2 Which uncertainties matter
Forecasting and Scenarios / Module 5 / 5.3
5.3 Consistent stories
Section 5.3 complete
Next: 5.4 Outcome and regret tables
Forecasting and Scenarios / Module 5 / 5.4
5.4 Outcome and regret tables
Section 5.4 complete
Next: Module 5 check
Continue to the module check Recap this section Back to module 5
Forecasting and Scenarios / Module 5 / Check
Module 5 check
Module 5 complete
Next: Module 6, coming later
Back to the Path Try it on your own question Recap module 5 Back to module 5
Forecasting and Scenarios / Module 1 / Recap
Recap card
1.1 What history can you use?
What history can you use?
- Check coverage: reconstructed months may be thin
- Match dates: invoices use the invoice month
- No annual pattern: the usable run is too short
Common mistake
Common trap. Treating rebuilt old records as full observations can invent growth.
Forecasting and Scenarios / Module 1 / Recap
Recap card
1.2 The naive benchmark
The naive benchmark
- Naive: next month equals this month
- Three-month average: only earlier completed months
- Error: distance from what happened
Common mistake
Common trap. Including the target month in its own prediction is hindsight.
Forecasting and Scenarios / Module 1 / Recap
Recap card
1.3 Hold some months back
Hold some months back
- Commit first: fit on earlier months
- Holdout: later months remain unseen while choosing the rule
- Compare errors: evaluate the same test months
Common mistake
Common trap. Picking the winner using the test months makes the test unfair.
Forecasting and Scenarios / Module 1 / Recap
Recap card
1.4 A range, not one number
A range, not one number
- Centre: a baseline based on completed history
- Width: grounded in observed misses
- Longer horizon: usually needs more room
Common mistake
Common trap. A narrow band is not evidence of accuracy, especially with little history.
Forecasting and Scenarios / Module 1 / Recap
Recap card
Module 1: Beat the simple forecast first
What history can you use?
- Check coverage: reconstructed months may be thin
- Match dates: invoices use the invoice month
- No annual pattern: the usable run is too short
Common mistake
Common trap. Treating rebuilt old records as full observations can invent growth.
The naive benchmark
- Naive: next month equals this month
- Three-month average: only earlier completed months
- Error: distance from what happened
Common mistake
Common trap. Including the target month in its own prediction is hindsight.
Hold some months back
- Commit first: fit on earlier months
- Holdout: later months remain unseen while choosing the rule
- Compare errors: evaluate the same test months
Common mistake
Common trap. Picking the winner using the test months makes the test unfair.
A range, not one number
- Centre: a baseline based on completed history
- Width: grounded in observed misses
- Longer horizon: usually needs more room
Common mistake
Common trap. A narrow band is not evidence of accuracy, especially with little history.
Forecasting and Scenarios / Module 2 / Recap
Recap card
2.1 A jump in the history
A jump in the history
- Check the source: rebuilt records can look thin
- Separate evidence: a count jump is not automatically demand growth
- Choose comparable months before projecting
Common mistake
Common trap. A steep line across inconsistent source periods is not a safe forecast.
Forecasting and Scenarios / Module 2 / Recap
Recap card
2.2 Level after the break
Level after the break
- Compare levels: use an explicit group of recent months
- Do not assume: a peak does not repeat by definition
- Name the window: dates make the conclusion checkable
Common mistake
Common trap. A single peak is not evidence that the same level will continue.
Forecasting and Scenarios / Module 2 / Recap
Recap card
2.3 One-off or new level?
One-off or new level?
- One-off: a notable month that may not persist
- Practice rule: examine at least three consecutive comparable months
- Source matters: reconstructed periods can distort a pattern
Common mistake
Common trap. Calling a single high month the new normal can overstate the plan.
Forecasting and Scenarios / Module 2 / Recap
Recap card
2.4 Does the signal come first?
Does the signal come first?
- Lead: the signal must arrive before the outcome
- Test: compare one month with the following month
- Caution: association is not a reliable prediction by itself
Common mistake
Common trap. Same-month association alone is not advance warning.
Forecasting and Scenarios / Module 2 / Recap
Recap card
Module 2: Breaks, blips and early signals
A jump in the history
- Check the source: rebuilt records can look thin
- Separate evidence: a count jump is not automatically demand growth
- Choose comparable months before projecting
Common mistake
Common trap. A steep line across inconsistent source periods is not a safe forecast.
Level after the break
- Compare levels: use an explicit group of recent months
- Do not assume: a peak does not repeat by definition
- Name the window: dates make the conclusion checkable
Common mistake
Common trap. A single peak is not evidence that the same level will continue.
One-off or new level?
- One-off: a notable month that may not persist
- Practice rule: examine at least three consecutive comparable months
- Source matters: reconstructed periods can distort a pattern
Common mistake
Common trap. Calling a single high month the new normal can overstate the plan.
Does the signal come first?
- Lead: the signal must arrive before the outcome
- Test: compare one month with the following month
- Caution: association is not a reliable prediction by itself
Common mistake
Common trap. Same-month association alone is not advance warning.
Forecasting and Scenarios / Module 3 / Recap
Recap card
3.1 The driver chain
The driver chain
- Orders: count the orders in a dated period
- Lines per order: divide lines by orders
- Line value: use the observed average over matching lines
Common mistake
Common trap. Detail without a historical check does not establish greater accuracy.
Forecasting and Scenarios / Module 3 / Recap
Recap card
3.2 Does it reproduce history?
Does it reproduce history?
- Reconcile: recreate a recorded result using matched records
- Rounding: keep exact values for the check
- Timing: an order and an invoice are not the same event
Common mistake
Common trap. Calling an order-to-invoice timing difference a model error without investigating it.
Forecasting and Scenarios / Module 3 / Recap
Recap card
3.3 Owners and sources
Owners and sources
- Source: table and period behind each figure
- Owner: person responsible for the assumption
- One controlled entry: do not keep conflicting copies
Common mistake
Common trap. A forecast with anonymous assumptions is difficult to inspect or improve.
Forecasting and Scenarios / Module 3 / Recap
Recap card
3.4 Which driver matters most?
Which driver matters most?
- Sensitivity: move one driver and hold the others fixed
- Compare: give each proposed change and the euro effect
- Challenge: cross-check assumptions that may influence each other
Common mistake
Common trap. Increasing modelling detail without checking the links can create false confidence.
Forecasting and Scenarios / Module 3 / Recap
Recap card
Module 3: Forecast from the drivers
The driver chain
- Orders: count the orders in a dated period
- Lines per order: divide lines by orders
- Line value: use the observed average over matching lines
Common mistake
Common trap. Detail without a historical check does not establish greater accuracy.
Does it reproduce history?
- Reconcile: recreate a recorded result using matched records
- Rounding: keep exact values for the check
- Timing: an order and an invoice are not the same event
Common mistake
Common trap. Calling an order-to-invoice timing difference a model error without investigating it.
Owners and sources
- Source: table and period behind each figure
- Owner: person responsible for the assumption
- One controlled entry: do not keep conflicting copies
Common mistake
Common trap. A forecast with anonymous assumptions is difficult to inspect or improve.
Which driver matters most?
- Sensitivity: move one driver and hold the others fixed
- Compare: give each proposed change and the euro effect
- Challenge: cross-check assumptions that may influence each other
Common mistake
Common trap. Increasing modelling detail without checking the links can create false confidence.
Forecasting and Scenarios / Module 4 / Recap
Recap card
4.1 Same average miss, different story
Same average miss, different story
- Error: actual minus forecast; minus means the forecast was too high
- Bias: misses that fall on the same side month after month
Common mistake
Reading size as lean. Rocket builder forecasts missed by similar amounts, yet only the three-month average was high all three months.
Forecasting and Scenarios / Module 4 / Recap
Recap card
4.2 Size and direction
Size and direction
- Mean absolute error: average size of the miss, signs dropped
- Mean error: average signed miss; minus means forecasts ran high
- Percentages break when actuals are zero or tiny
Common mistake
Reading mean absolute error as bias. University naive missed high in all three months, April to June 2025; only the mean error shows it.
Forecasting and Scenarios / Module 4 / Recap
Recap card
4.3 Compare on the same months
Compare on the same months
- Fair comparison: same held-back months, same horizon
- Data leak: a forecast that used the actual it predicts
Common mistake
Comparing different months. The colleague's three-month win flips to naive on matched April to October months.
Forecasting and Scenarios / Module 4 / Recap
Recap card
4.4 Did the hand adjustments help?
Did the hand adjustments help?
- Forecast value added: plain miss minus adjusted miss, on the same months
- Working days differ by month, so compare per business day too
Common mistake
Judging a change on one month. The practice +€50,000 helped April and made April to October worse.
Forecasting and Scenarios / Module 4 / Recap
Recap card
Module 4: How wrong, and which way?
Same average miss, different story
- Error: actual minus forecast; minus means the forecast was too high
- Bias: misses that fall on the same side month after month
Common mistake
Reading size as lean. Rocket builder forecasts missed by similar amounts, yet only the three-month average was high all three months.
Size and direction
- Mean absolute error: average size of the miss, signs dropped
- Mean error: average signed miss; minus means forecasts ran high
- Percentages break when actuals are zero or tiny
Common mistake
Reading mean absolute error as bias. University naive missed high in all three months, April to June 2025; only the mean error shows it.
Compare on the same months
- Fair comparison: same held-back months, same horizon
- Data leak: a forecast that used the actual it predicts
Common mistake
Comparing different months. The colleague's three-month win flips to naive on matched April to October months.
Did the hand adjustments help?
- Forecast value added: plain miss minus adjusted miss, on the same months
- Working days differ by month, so compare per business day too
Common mistake
Judging a change on one month. The practice +€50,000 helped April and made April to October worse.
Forecasting and Scenarios / Module 5 / Recap
Recap card
5.1 High, middle, low changes nothing
High, middle, low changes nothing
- Scenario: a story that could change which option wins
- Demand-only sets often keep the same winner
Common mistake
Three sizes of one forecast. At demand -5%, middle and +5% keeping overtime wins every time.
Forecasting and Scenarios / Module 5 / Recap
Recap card
5.2 Which uncertainties matter
Which uncertainties matter
- Rate an uncertainty by whether it could change the winner
- Read anchors as series: machining swung from 12,079 to 17,396 hours a month
Common mistake
Two months as a trend. September to October looks like growth; April was higher than October.
Forecasting and Scenarios / Module 5 / Recap
Recap card
5.3 Consistent stories
Consistent stories
- Consistent story: orders, hours and overtime move together
- Different stories pressure different options
Common mistake
A story that cannot happen. Orders up 20% with machining hours flat needs a stated reason.
Forecasting and Scenarios / Module 5 / Recap
Recap card
5.4 Outcome and regret tables
Outcome and regret tables
- Regret: cost above the cheapest option in the same story
- Lowest worst regret: the option never badly wrong; signposts say which story is starting
Common mistake
Picking the cheapest in one story. Buying wins the surge but carries €50k regret if demand stays steady.
Forecasting and Scenarios / Module 5 / Recap
Recap card
Module 5: Stories that change the decision
High, middle, low changes nothing
- Scenario: a story that could change which option wins
- Demand-only sets often keep the same winner
Common mistake
Three sizes of one forecast. At demand -5%, middle and +5% keeping overtime wins every time.
Which uncertainties matter
- Rate an uncertainty by whether it could change the winner
- Read anchors as series: machining swung from 12,079 to 17,396 hours a month
Common mistake
Two months as a trend. September to October looks like growth; April was higher than October.
Consistent stories
- Consistent story: orders, hours and overtime move together
- Different stories pressure different options
Common mistake
A story that cannot happen. Orders up 20% with machining hours flat needs a stated reason.
Outcome and regret tables
- Regret: cost above the cheapest option in the same story
- Lowest worst regret: the option never badly wrong; signposts say which story is starting
Common mistake
Picking the cheapest in one story. Buying wins the surge but carries €50k regret if demand stays steady.