Monte Carlo in planning: ranges instead of three scenarios

What three scenarios cannot do

In most planning processes, scenarios are built by hand. Someone sets all assumptions to their favourable value, then all to their unfavourable value, and the plan sits in the middle. This has three weaknesses.

• No probability. Nobody can say whether the bad case occurs in one year out of three or in one out of a hundred.
• Everything moves together. It is rare for volume, price and material costs to hit their worst values at the same time. As a result, the bad case is too bleak and the good case too rosy.
• The effort grows with every question. Each additional scenario is another copy of the plan that has to be maintained.

How a Monte Carlo simulation calculates

The method is old and simple to describe [1]. For each uncertain assumption, a range is specified within which its value can lie. The calculation draws one random value per assumption from that range, runs the plan with it and records the result. It repeats this many thousands of times.

The outcome is a distribution. It shows the range in which the result fell in most runs and how often it fell outside.

An example with invented figures: a plan shows an operating result of €8 million for 2027. After the simulation, the statement is that 90 out of 100 runs fell between €7.2 million and €8.7 million, and 5 out of 100 below that. A minimum result of €7.5 million is missed in 16 out of 100 runs. The plan is the same. Management now knows how robust it is.

What this looks like in SAP Analytics Cloud

In SAP Analytics Cloud, the feature is called Compass. According to SAP's lesson on the topic [2], Compass calculates using the Monte Carlo method and relies on the relationship between drivers and the target measure that is stored in the model as a formula. The method has nothing to do with artificial intelligence.

For each driver, you set a minimum and a maximum value and choose whether the values in between are drawn from a normal or a uniform distribution. Drivers can also be constrained as Restricted Drivers [2].

Compass calculates with a choice of 1,000, 10,000 or 100,000 runs [2]. More runs take longer and give a more stable picture.

Compass requires an SAP Analytics Cloud for Planning licence, at the Professional or Standard level [2].

What the simulation requires

A model with drivers

The simulation needs a formula that links drivers and result. A plan in which the result is entered as a figure cannot be simulated. Driver-based planning is therefore the first step.

Ranges that someone can justify

Where does the assumption come from that volume will lie between two values? From previous years, from the order backlog, from the sales team's assessment. A range without justification makes the result more impressive and not one bit more accurate.

A look at dependencies

If the calculation draws each assumption independently of the others, although two of them are linked in practice, the spread is misjudged. Such relationships belong in the model as a formula before simulating.

Where the simulation does not help

A simulation calculates with the assumptions it is given. An event that does not appear in any range does not appear in the result either. It answers the question of how much a plan fluctuates under known uncertainties. For the loss of a supplier or a new regulation, the classic scenario remains the right tool.

How to read the result

Three questions can then be answered. How likely is it that the target will be reached? How far can the result fall in an unfavourable case? And which assumption moves the result the most? The third question is the most important for steering the business, because it shows where more precise planning makes the biggest difference.

The decision remains with management. The simulation ends the argument about whether the bad case has been painted too bleak.

What the simulation achieves in a finished planning model and what the result looks like is shown in the Monte Carlo simulation section of our SAP Analytics Cloud page.