Forecasting (sales, operations & finance)

We are living in an era of high market vulnerability in challenging events. Those events are not anymore considered as “Black Swans” as they occur more and more frequently having a tremendous impact on the world’s economies. CCOs and COOs are heavily dependent on accurate forecasts to serve demand, produce with sufficient raw material quantities purchased at the best possible price. CFOs need all those estimates to be reflected in the cash flow projections.

As a result, companies are trying to answer the below questions:

• How accurate is our forecasting? Is there any way to improve it?
• Are we using the optimal algorithms?
• How could we minimize the effort and time spent in producing our forecasts each time?
• How predictable is our product portfolio? At which level?
• Are there any additional data we could use that would potentially improve our forecast accuracy?
• Is there any platform that could integrate the overall process, so that all employees involved could collaborate? Could this be done in an affordable manner?

WITSIDE offers end-to-end forecasting solutions combining historical (internal) data, with exogenous data that are improving the accuracy of the forecasting models. We have developed a unique framework to address our clients’ needs, consisting of 4 steps:

We deploy AI-driven software solutions to enable our customers perform effortless forecasting, embedding their personal experience or calendar events that affect its accuracy. We assist them to measure their accuracy and continuously improve it under a solid S&OP process.

Machine Learning (commercial and operations)

Machine learning (ML) is a widely (if not abusively) used term to describe an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed.

In WITSIDE we believe that manual tasks performed during traditional Data Science (DS) processes could be replaced by automated machine learning (auto-ML). Auto-ML takes advantage of the strengths of both humans and computers. Humans excel at communication, engagement, context and general knowledge, as well as creativity and empathy. Computers and software systems are ideal for repetitive tasks, mathematics, data manipulation, and parallel processing — providing the power and speed to master complex solutions.

The tasks that auto-ML automates are shown in the below graph:

In a nutshell, our approach is based on the context and needs of our clients during a project. We deploy auto-ML solutions when:

• Internal DS resources are scarce (in number or availability)
• Quick results is a vital success factor
• Business users want to use ML and transform into citizen DS
• We need control over the ML build process

We deploy classic ML when:

• Specific ML approaches that auto-ML does not cover (such as unsupervised & reinforcement learning) are required
• Complex data types exist such as text or network data
• Feature engineering and domain knowledge are demanding

ML or auto-ML is can be used standalone or combined with other techniques. For instance, we can “train” a model based on drivers’ behavior to predict delivery time of an order. Or we can “sniff” social media content to identify posts about specific products, services, companies, or individuals and get a sense of their meaning. In production, ML models can identify quality issues in output images. In insurance, to spot fraud patterns in claims’ processing or predict customer churn and in retail to segment customers and maximize their lifetime value.