The o9 Digital Brain is a cloud based solution designed to enhance the speed and scale of your supply chain planning. o9 empowers data scientists to flexibly and transparently process and refine historical data with best in class predictive models. Let's explore two questions, key to the success of any data science team. How can we develop the best predictive models?

How can these models leverage business wins? The answers are through a combination of platform and process, enabling business insight. Let's look at platform first. The o9 platform allows data scientists to 1.

Integrate data 2. Store and curate data. 3. Analyze and predict data. And 4. Visualize and publish data, Let's focus in on the data science machine learning elements of the platform.

It contains o9 predict AI for a fast no code entry into machine learning based forecast. For custom built models, the data science platform offers native support for R, Python and PySpark, allowing data scientists to utilize the largest library of open source models for statistics, simulations, optimization and machine learning. Data scientists can literally 'plug and play' in o9 by using copy and paste of R and Python models from open source libraries and then invoking the scripts. o9 is fully integrated with Jupiter Hub, so it is easy for data scientists to develop code on the platform using the tools that they are used to.

For example, o9 supports git the version control to publish models and then see the results within the o9 user interface. And o9 is designed to scale as needed using parallelization through a Big Data Execution Model on the Hadoop ecosystem, including HDFS Hive, and Spark. The supporting infrastructure allows elastic scaling vertically and horizontally and ensures that results are ready whenever needed. This means that data scientists can develop ML models from small test datasets to millions of items in locations at high frequency like week or day, and provide the insights to the business on the o9 user interface.

With the platform explained, let's turn to how data scientists can create the best possible models by following these process steps. First is collect key data, whether internal or external, structured or unstructured. Second is data preparation, which involves data cleansing and ensuring compliance and integrity, assessing gaps and defining the rules for outliers. Segmenting for forecast ability, including ABC/XYZ classification seasonality, trends, intermittence.

and engineering features such as customer and item attributes, time series based features, and data drivers. Third is highlighting impactful features. Use correlation analysis to find the most relevant variables and to remove variables with Multicollinearity. Essentially, this means using the concept that "less is more".

And running model architecture analysis to determine the best forecasting levels and time buckets for segments. And fourthly, completing model training and tournament evaluation to determine the best forecast recipes and select the right models or selection of models to use for the production forecast. So how can better models be translated into business wins? The answer is from insight gained from having experts in your own data.

Let's look at some examples demand drivers, product features and blended forecast. Demand drivers first. In this view, we can see how data scientists can distinguish between demand drivers that can be influenced by the business, such as promotions or pricing against the drivers that cannot be influenced, such as the baseline, trend, seasonality or holidays. Here, the contribution factor of each driver is shown and visualization of the forecast composition.

The data scientist can define and display these demand driver insights to the business, providing rich insight for event and pricing management, gap closure, and exploiting market opportunities. Now for product features. Data scientists can use neural embedding to represent product features in vectors. A new product and its attributes can be mapped to all the available product features from training data, to their respective vectors.

The importance or weight of each attribute is calculated through an ML algorithm that the data scientists can create and refine. The weights are used in a distance function to find appropriate similar items with a good amount of history. Product feature mapping can significantly reduce the time spent in NPI forecasts creation and these features insights are invaluable for product, portfolio, and market growth analysis. And as a final example, let's look at blended forecasts.

The data scientists can evaluate blending different forecasts together, sometimes called hybrid or merged forecasts. The idea is to compare the traditional single best fit forecast approach with a blend of multiple forecasts. Different blends can be used for different horizons, and these multi blended forecasts can provide significant improvements in accuracy, reduction of bias and increased stability, making supply chains more efficient and agile. So as you have seen, the o9 Digital Platform empowers data scientists to develop the best predictive models for their data and use these models to leverage business wins.