What-if: Modeling Simulation Systems

The application of statistical techniques in the modeling of what-if systems is very useful to identify hypotheses and represent possible business outcomes considering several variables.

So, if you want to learn the basics and techniques of what-if now, continue reading this post.

What-if: Simulation systems

In a simplified way, what-if systems are solutions that allow the simulation of scenarios in a dynamic way. This means that those who use this technique to obtain different results from the visualization of hypothetical scenarios can change the main variables of these systems, such as:

• Sales volume
• Product Price
• Net Margin

Even today, a spreadsheet is the most used tool to simulate results, mainly because it is simple to use and widely spread in the market.

It is in this way, therefore, that many companies continue to make predictions: through electronic spreadsheets, such as MS Excel and Google Sheets, mistakenly leaving the specific modeling tools for this, such as what-if systems, relegated to projects of greater complexity.

And this makes less and less sense, as new alternatives and technologies for modeling and simulating scenarios are always emerging, such as Arena and FlexSim, for example.

In other words, with the proliferation of data warehouse (DW) projects and business intelligence (BI) tools, we expect the use of spreadsheets for scenario design to become increasingly obsolete.

And this will happen because it is not possible to use spreadsheets consciously, choosing to lose the advantages of manipulating large volumes of data with the speed of access allowed by BI tools.

What-if Systems: Difficulties

Most business intelligence solutions do not allow automatic scenario simulations or require major changes to the data model, which makes it difficult for the standard user to use them.

And it is important to remember that these systems are usually proprietary code and not modular, which unfortunately implies high cost and difficulty in using them in different environments.

In short, the importance of using data is gaining more strength every day and, as a result, companies and professionals are looking for data-driven solutions for decision-making and what-if systems are advanced resources created to meet the demands of this new market.

However, there are still many challenges. For example, the use of simple software such as Excel does not allow integration and/or direct connections with big data and databases. In addition, the most robust solutions (BI and DW) are expensive and complex to implement.

But don't be discouraged! With that in mind, our main goal with this article is to simplify the simulation process and demonstrate how Indicium works to overcome these challenges.

Concepts involved in a what-if system

Data is generated in operational applications such as ERPs, CRMs, POS systems, etc. Such applications often store this information in relational databases (SQL).

For analytical purposes, it is recommended that this data be stored in a data warehouse, or other equivalent structure, and transformed into information relevant to a business.

From there, BI tools, such as MS PowerBI, Tableau, Looker or Qlik, can be integrated into the already organized database, allowing the generation of intuitive reports and visualizations based on this information stored in DW.

Thus, graphs, tables, maps and other illustrations are generated in this process that we call descriptive analysis.

But the same information can be used in other processes, such as data mining or data science, for example, to identify patterns between the variables of a business and predict new values and situations through predictive analysis. And finally, simulation and optimization models can be created to support assertive decision-making.

Note that descriptive analytics is the starting point in the analytics process. Despite this, this is the stage that receives the most support from tools and technologies.

And, although much of the effort to achieve assertive decision-making occurs after descriptive analysis, the deficiency of techniques and methodologies lies precisely in the last two stages of analytics.

This is because it is there that the most advanced techniques and the most qualified teams are needed for the execution and use of the systems.

Past, present and future

BI applications allow you to answer questions about the past and present of business processes. For example, you can answer questions such as:

• what was the turnover of product X in the year 2020?
• Which customers have accounted for at least 10% of the company's revenue in the last five years?
• What is the average productivity of the company year over year?

Unlike BI, a what-if analysis wants to model the future of a business under a set of hypotheses, also called scenarios.

What-if analysis is a data-intensive simulation whose main objective is to inspect the behavior of a complex system (an organization or a department) under some probable scenarios.

For this reason, what-if simulations analyze the impact of changes in a set of independent variables on a set of dependent variables. In doing so, they offer a simplified representation of a business and illustrate its fundamental characteristics based on a company's historical data (Kellern et al., 1999).

Structure of a what-if system

To model a DW, there is a data transformation step called ETL. This process has similarities to the logical structure required for what-if modeling.

For example, in both models, an input data series is processed by a model or function that generates a new output data set.

But there are also differences, and the main one is that, when it comes to ETL, only historical data is used, while for the implementation of the what-if analysis, new datasets and external models are needed.

In order for the scenario to work in a what-if simulation model, complex relationships are established between corresponding variables and entities in a domain (e.g., sales, products, customers, costs, etc.).

This scenario is composed of the following elements:

1. the set of variables that are the object of a simulation;
2. variable sources;
3. the functional way in which the sources relate to each other;
4. the model;
5. and the set of additional parameters used to generate the result.

Each scenario describes ways and alternatives to construct a prediction of interest to the company from actual historical data, parameters, or additional hypotheses.

The hypotheses of a scenario can be divided into:

1. hypotheses about business parameters;
2. model parameters;
3. and the model itself.

This means that having the history of the data stored in a relational database is not enough to create hypotheses for a scenario. And that it is necessary to identify the functional way in which this data relates, and combine this information with other variables that are not directly linked to the business.

While BI tools answer questions about the past and present of a business, a what-if analysis seeks to answer what will happen in the future or in atypical hypotheses, if any key variable of the business is changed. For instance:

1. What will be the company's profit if it increases the price of product X by 20%?
2. What will be the impact on sales if the company reduces 15% of its advertising investment in area Y?

What-if analysis: simulations and results

To answer various questions and business hypotheses, such as those we have cited as examples in this article, a simulation model must be developed.

It needs to be able to express the complex relationships between the dependent variable you want to predict (such as annual profit) and the independent variables you want to "adjust" (such as the price of product X).

To do this, historical data is used to "calibrate" the model and make a prediction (or prediction) of future profits.

Translating to the what-if simulation model, from the scenarios we defined above, it needs to contain:

• A list of all the variables necessary to estimate a company's profit, such as sales revenue per product, costs in general, the evolution of the prices of each product, etc.
• the input data of the required variables, such as "the 20% increase in the price of product X".
• The forecasting model that relates profit to other business variables in a deterministic or non-deterministic way.
• the parameters of the model that define how many years to consider in the forecast, what is the confidence interval of the estimate, etc.

In addition, a what-if simulation model should contemplate three perspectives of a simulation:

1. functional: describes how the variables relate to each other;
2. Static: defines which variables will be used and aggregated in the creation of each object (dataframe);
3. Dynamic: Explicitly describes the order in which tasks are performed (sequential, concurrent, or alternative).

Conclusions on the modeling of what-if systems

What-if systems are key to taking decision-making to the next level. That is, truly including forecasting and predicting the future in business.

Therefore, modern companies that already care about the future and use what-if (or similar) resources will gain more and more space in their markets and gain an important analytical and competitive advantage in the coming years.

However, there are obstacles in the way.

We know that the efficiency of models depends on a well-structured data project that contemplates the past, present, and future.

In other words, both the configuration of past and present (through DW and BI tools) and the development of what-if models require a robust and complex data structure, and not all companies have the necessary resources to invest in this.

But although it seems like a complex subject, here at Indicium we have developed methodologies and data systems that enable the complete execution of these projects in an objective and simplified way.

We do everything from the implementation of ELTs, data warehouses and data lakes, to the configuration of custom data app systems.

Therefore, it is possible to say that we develop complete solutions for past, present and future so that companies gain an analytical advantage over their competitors and achieve objectives such as:

• Increase in sales
• Cost Reduction
• Increase in quality
• Speed Boost
• More productivity
• Risk Reduction

Did you like this content about what-if systems?

Subscribe to our newsletter and visit our blog to learn more about the most advanced concepts, tools, and techniques in the world of data.

‍

References

GOLFARELLI, M.; RIZZI, S. What-if simulation modeling in business intelligence. International Journal of Data Warehousing and Mining. Hershey, Pennsylvania: IGI Globalout, Oct. 2009.

GOLFARELLI, M.; RIZZI, S.; PROLI, A. Designing what-if analysis: towards a methodology. In: Proceedings International Workshop on Data Warehousing and OLAP. Arlington, USA: 2006. (pp. 51–58).

HAAS, P. J. et.al. Data is dead... Without what-if models. In: Proceedings of Very Large Data Bases Endowment - PVLDB, v. 4, n. 12, p. 1486-1489, Aug 2011.

HAAS, P. J. et.al. Splash: simulation optimization in complex systems of systems. In: Proceedings of 50th Annual Allerton Conference. Pennsylvania State University, CiteSeer x, 2012. CEFKIN, M. et.al. Splash: a progress report on building a platform for a 360 degree view of health. In: Proceedings of the 5th Informs Workshop on Data Mining and Health Informatics, DM-HI. Austin, TX, 2010.

‍