What are the purchase patterns of your customers and are they changing? What are the trends in your manufacturing costs? Data mining enables to find patterns and relationships in historic data. Algorithms such as decision trees, k-means, or Naive Bayes are applied especially to big data to identify insight behind the numbers. The data is then usually clustered to identify for example high value customers or similar products. Clustering is also needed for forecasting and modeling.

How will the market develop over the next three years? What will my sales be next quarter? Forecasting algorithms can address these questions, but they need to be carefully chosen. With new products or new markets there is little or no historic data available. Customer demand is then projected using Bass diffusion models based on the degree of innovation and imitation. With growing history and more knowledge about customer behavior, causal models are developed based on econometrics & regression analysis to identify dependencies among economy, available substitutes, and complementary products. With long history, time-series analytics such as exponential smoothing and moving average models are applied to project future trends, cycles, and seasonalities. It’s critical here though to keep in mind that past performance is not necessarily indicative of the future.

What is the variability and risk in my sales forecast? Most forecasting techniques are point estimates, but since all projections are inherently wrong, it only makes sense to apply scenario analyses to get better understanding of what may happen. Monte Carlo simulation is an advanced modeling technique that allows to analyze thousands of scenarios almost instantly to identify a range of potential outcomes instead of focusing on only one. This modeling approach also allows to take into account market and business context instead of blindly trusting the results of algorithmic calculations.