Risk-Based Probabilistic Planning –

A Data Story

support@mcflosim.ch
 MC FLO – Risk-based Probabilistic Planning 1

Motivation
Flora Company has been crafting gift items for years, distributed through various channels such
as department stores, local pop-up shops, the internet, and intermediaries. The gift item market
is fiercely competitive and replete with innovations, as new competitors are constantly entering
the market and customers demand frequent surprises.

In the past, the company expended considerable effort in establishing budgets for the constantly
evolving products and steering the company through the year. While information on costs and
sales figures from past years is available and used for budget planning, the company remains
dissatisfied with the following issues:

The bottom-up budget values and rolling forecasts do not account for variability and uncertainty.

Setting an ambitious target value has proven difficult. The planning managers hold an
informational advantage over the board of directors, therefore a conservative and less ambitious
plan resulted.

Forecasting the next few months until the end of the year (the "latest estimate") through bottom-
up planning is very time-consuming. Many areas have undergone the same level of effort as
budget planning. Plan values do not match target values, or no countermeasures can be derived
from the latest estimate.

Alternatively, the latest estimate was elaborated based on past data, but trends or seasonality
could not be accurately derived. A simple trend calculation did not explain future causes of
possible deviations from the target value.

Given these identified inadequacies, the company wishes to take an innovative approach to
financial aspects, and apply this to the relatively new product A.

Solution
To address these challenges, Flora Company has decided to implement driver models to identify
the cause-and-effect relationship, which breaks down the entire value creation process into
individual components, such as prices and quantities.

To better account for variability in assumptions ("demand for quantity will vary between 4,700
and 11,500 units next year, with a most probable value of 9,000 units") and thus uncertainty, a
Monte Carlo simulation is now being added as part of the planning process. The computer-
generated simulation calculates possible outcomes based on the "bottom-up" derived
assumptions and creates thousands of "what-if" scenarios for the drivers.

MC FLOsim
support@mcflosim.ch
 MC FLO – Risk-based Probabilistic Planning 2

Figure 1: Distribution of the demand

In addition, dependencies between the drivers are considered based on expert opinions.

Figure 2: Dependency (correlation) between demand and costs per unit

By aggregating the driver variables, Flora Company can consistently recognize all possible
characteristics of its control variable, EBITDA. This enables the company to identify the main
drivers of its financial performance and derive realistic ambitions.

The number of possible realizations of EBITDA is available as a distribution, which reflects Flora
Company's prior knowledge. By incorporating this distribution into its financial planning and

MC FLOsim
support@mcflosim.ch
 MC FLO – Risk-based Probabilistic Planning 3

management, the company can account for variability and uncertainty in its forecasts and make
more informed decisions.

Figure 3: Distribution of the target variable EBITDA using driver model (prior knowledge)

According to the distribution of simulated prior knowledge, approximately 95% of possible EBITDA
outcomes fall within the range of CHF 31,470 and CHF 109,400. Flora Company has set a realistic
and ambitious EBITDA target of CHF 75,000 based on this distribution. So, only 28% of all possible
outcomes are above this target value.

In contrast, a target value of CHF 130,000 is above the maximum possible outcome and is
therefore not realistic. Having set the target value, all influence or explanatory variables, such as
quantities (9,189 units) and maximum costs per product can consistently be derived.

A tornado analysis reveals which variable has the greatest impact on EBITDA. This analysis helps
the company to identify potential risks and opportunities in its financial planning and
management.

MC FLOsim
support@mcflosim.ch
 MC FLO – Risk-based Probabilistic Planning 4

Figure 4:Tornado analysis of EBITDA

Flora Company's management is convinced that the new planning approach, incorporating driver
models, Monte Carlo simulations, and tornado analyses, has increased transparency, empowered
commitment, and highlighted the risks associated with the target value. However, the company
also recognizes that the distribution of EBITDA, based on prior knowledge, needs to be refined
using observed data for effective steering.

To refine its forecast, Flora Company plans to measure data on quantities sold and costs per unit
and compare them with their respective target values. This will allow the company to derive
necessary actions, create a new forecast for the end of the year ("Latest Estimate"), and adjust
the target value if necessary. In contrast to the previous "bottom-up" approach and time series
calculations, the company will only use current data to refine its prior knowledge, reflecting the
belief that these figures better reflect the current market reality.

To create a robust new forecast, Flora Company has collected results from the first five months.

Month Quantity Costs per

MC FLOsim
support@mcflosim.ch
 MC FLO – Risk-based Probabilistic Planning 5

Based on the given data, the correlation between the quantity sold and the costs per item is -0.88,
which is close to the assumed correlation of -0.9 in the budget plan. However, the observed
quantity sold in February is only 56 units, compared to 1,200 units in January, underlying the
volatility in the data.

To incorporate the observed volatility in the preparation of the new forecast, the company
randomly selects a value from the observed data for the months of June to December and add
them to generate a possible realization. Using a Monte Carlo simulation, the company can create
thousands of "what-if" scenarios and generate a distribution of the possible realizations of the
sales volume from June to December. By combining this with the observed data from January to
May, the company can create a forecast for the whole year.

Figure 5: Forecast for demand based on observed data

95% of the results are between 6,942 and 9,917 units, the derived target value of 9,189 units is
therefore within the forecast based on the observed data.

The innovative approach combines the prior knowledge and the forecast based on the observed
data to generate a posterior knowledge (based on Bayes’ rule). Analogous to the prior knowledge
and the forecast based on the observed data, the posterior knowledge is uncertain and given as
a distribution.

This can also be applied analogously to the costs per piece.

The aggregated EBITDA is therefore also available as a distribution (posterior knowledge). 95% of
the results of the posterior knowledge in relation to the EBITDA are between approx. CHF 48'800
and CHF 92'500. The uncertainty regarding the EBITDA has decreased compared to the prior.

MC FLOsim
support@mcflosim.ch
 MC FLO – Risk-based Probabilistic Planning 6

Figure 6: Posterior distribution of EBITDA

The posterior shows that the expected value is around CHF 67,800 and is therefore CHF 7,000
below the target value. The management choose a value from the posterior distribution as the
new forecast. The target level with an ambition of 28% leads to a forecast value of approx. CHF
73,700. This is only marginally different from the previous target value.

The company can now restart this process month after month. The new monthly data can be
merged with the posterior knowledge to obtain a new posterior knowledge. With each passing
month, the uncertainty regarding the EBITDA thus decreases and the most probable value of the
posterior knowledge approximates the true value by the end of the year.

Summary
The planning logic implemented by Flora has proven to be highly advantageous, resulting in a
positive outcome at the end of the year. One of the main benefits of this approach is that
management has delegated the determination of the drivers' characteristics and their
relationship to the employees, resulting in better-informed decisions. Additionally, transparent
communication of the target value to the board of directors based on previous knowledge has
increased commitment, while driver models have increased employee understanding of the
process and how their decisions impact it.

Moreover, the forecast based on observed data has significantly reduced the effort required
compared to traditional bottom-up approaches, while still maintaining cause/effect relationships.
The Monte Carlo simulation has acted as a catalyst for determining prior knowledge, observed
data forecasts, and posterior knowledge, all with minimal manual correction required. By
aggregating all driver values, Flora has achieved consistent plans for all planning stages, thus
eliminating inconsistencies.

MC FLOsim
support@mcflosim.ch
 MC FLO – Risk-based Probabilistic Planning 7

The combination of prior knowledge with measured data has led to a high level of planning
accuracy, and adjustments to observed data forecasts still preserves consistent statements from
the posterior knowledge. The new planning logic has bridged the gap to risk management,
quantifying all relevant uncertainties and taking them into account in the planning process, thus
replacing risk matrices in most cases. Finally, this approach is scalable and can be extended to any
planning period, resulting in a uniform planning process for both the short and long-term horizon.

While Flora acknowledges some disadvantages compared to the previous planning logic, such as
the increased time required to derive prior knowledge based on driver models, and the black-box
nature of deriving posterior knowledge, the benefits of the probabilistic planning approach far
outweigh the drawbacks.

Overall, probabilistic planning represents a lighthouse for Flora, particularly in volatile times, and
has transformed communication with plan figures by disclosing not only the point value but also
the 95% credible interval and the level of ambition.

Remark: the calculations were carried out on Microsoft Excel for Office and with the Monte Carlo
Add-In MC FLO. MC FLO is available in German, English and Spanish.

MCPlan is an integrated planning system that includes the income statement, balance sheet and
cash flow statement, supporting a driver-based modelling.

Training videos of MCPlan: https://youtu.be/m2ML9EpMqKY, https://youtu.be/bBM6PERC9Ps

Training video of MC FLO and Bayesian analysis: https://youtu.be/dNz9YTqyK14

Please let us know if you have any questions.

www.mcflosim.ch
LinkedIn: https://www.linkedin.com/company/14018968/

MC FLOsim
support@mcflosim.ch