Exclusive portal designed for our Customers and Partners

Log in

Request for Access

logo
logo
countries flags

Blog post series: Plan your predictive analytics and machine learning journey with CCH Tagetik - Part 4

Sep. 23 2020 by Prof. Dr. Karsten Oehler, Solution Architect - CCH Tagetik DACH / Marco Van der Kooij, Managing Director - ForSight Consulting

Performance Management Business Intelligence & Analytics

In the last part of this Blog Post series we will go ahead to deep dive you discovering how you can plan your journey into predictive analytics and machine learning.

Today, we want to talk about driver-based simulation.

Forecasting is a passive way to view the future. Active decisions which influence future developments can benefit from simulation of key indicators such as EBIT or Cash, but the quality of the results depends on the accuracy of cause and effect relationships in the model.

But too often, managers trust oversimplified simulation models. For instance, the effect of sales price variation is only considered in terms of revenue, when in fact volume is also important because if neglected you risk:

  • Wrong decisions due to over/under estimation
  • Taking only a partial view of the decision under consideration. For example, tax effects on cash might be omitted or only approximated
  • Failing to understand the effects of risk on overall value creation. Average calculations usually hide potential risks or ignore them completely

Management decisions in complex situations need to be quantified, and as management is interested in financial results, it is helpful when a core financial model is an integral part of the simulation. But non-financial cause and effect rules are also necessary to create a realistic view and contribute to the building of powerful scenarios.

Management can benefit from these simulations in many ways:

  • Working with different scenarios shows where future developments may lead
  • Scenarios can be assessed with an optimistic or pessimistic view to allow for extreme but realistic developments
  • Complex dependencies can be considered: for instance, constraints due to machine capacity, which only become clear after running production planning (including MRP and routing). Also the effect of currency variations is hard to assess manually, because it can affect both purchase costs and revenue and thus needs an integrated simulation model

Machine Learning and Simulation

Simulation is not machine learning. But you can use machine learning to improve your understanding of cause and effect and improve the quality of results. Simulation depends on a reliable relationship between input and output, and realistic, statistically-based input parameters. Machine learning can provide both.

Enterprise simulation is not a one-time task. A financial simulation usually needs a lot of input from different business areas. The following diagram shows the relationship between forecasting and simulation. The same machine learning methods for detecting forecast figures can be used to identify cause and effects for simulation models. The detected drivers derived from sales and cost forecasting should be used to enrich simulation models.

blog post series imm1

Figure 1: prework for simulation

For this the simulation model must be open to communicate interactively with machine learning tools. The output of machine learning is not necessarily an easily-transferred polynomial function and specific simulation tools are often closed shops. Imagine you want to run a revenue development simulation and estimate the effects of your discount policy. Every change in the discount requires a machine learning inference to be run, and the most obvious solution is to use inference directly in simulations. This makes simulation more accurate but requires close integration between machinery and simulation.

CCH Tagetik provides all that is needed to set up a simulation:

  • A strong starting point is the built-in financial intelligence including an integrated profit and cash model
  • Various predefined operation planning models like human resource planning or production costing
  • The capability to incorporate machine learning modules and link them directly with CCH Tagetik’s financial modelling

This solution enables the creation of a realistic base scenario – including discovered causes/effects. The impact of possible initiatives can be calculated accurately by including machine learning funded dependencies.

Handling uncertainty – Monte Carlo Simulation with CCH Tagetik

Using probabilities in forecasts delivers more insights into risk and opportunities. But they are also important for simulations. To what degree can the resulting simulation measures (e.g. EBIT or Cash) vary? Sensitivity analysis might be one way to find out, or a more realistic option is to work with scenarios based on random numbers, generated according to a certain distribution function. But where do you get the distribution functions? Traditionally they are based on individual assessment, but today machine learning could also provide distribution functions for use as input for a risk-based simulation.

It is one small step towards integrated risk modeling. And to aggregate KPI results, risks must be aggregated at an organizational level. But simply aggregating a distribution function does not produce accurate probabilities. Simulation with the Monte Carlo method is quite often used in this situation to aggregate risks (and opportunities) and is an easy and appropriate way to aggregate all kinds of distribution functions.

With Monte Carlo simulation, instead of single variable inputs you usually work with distribution functions (typical functions are normal, triangle, PERT, binomial distribution). The result of the simulation should be a distribution function of output KPIs. The resulting measure is usually an empirical distribution with a left tail (see right box in figure 2) due to certain risks. Random numbers are generated up to 100,000 times according to the distribution functions. Each unique set of parameters is a scenario. The results are calculated for each scenario using a financial model and the cause and effect rules provided by machine learning. A histogram shows the result distribution, giving the probability for each possible output. It is also possible to get the so-called “value at risk” for every probability. Correlation between the input variables can be included in the simulation to increase the model’s accuracy.

The following graphic shows the necessary steps towards an integrated simulation framework:

blog post series imm2

 

Figure 2: An integrated simulation view

The openness of the data model means simulation engines can be combined with the CCH Tagetik financial model. The simulation model combines deterministic formulas likes revenue = volume * price with stochastic results from the deployed machine learning model by handing over a vector of input variables for each simulation scenario. An empirical distribution function is generated from the resulting scenarios.

The results can improve management’s capability for accurate decision making. Results from predictive analytics can be used to improve the quality of the scenarios. The benefits include:

  • Meaningful business cases can be created through better assessment of activities/projects
  • Potential risks are integrated
  • Bandwidths provide a more realistic view on planning results

In four blog posts (combined in a whitepaper) we have explained how we are already in a position to apply new developments in Artificial Intelligence and specifically Machine Learning for collecting more granular operational and financial data, and reporting and analyzing this data. The Office of Finance can benefit from these new developments to provide more accurate, forward-looking insight with improved quality of outcomes to steer the business towards sustainable value creation. CCH Tagetik is very well placed to accomplish this.

Get the complete picture, download the whitepaper "Machine Learning for Controllers. Use cases for Forecasting, Planning, and Simulation" click here.

Innovation with tagetik

 

Share this post!