The Auto-Labeling Invoices project used a multi-class classification approach on tabular data using lines from invoices, to automatically categorize your expenses and streamline the accounting process. This enabled the salespeople  to spend more time to focus on sales, boosting their productivity and driving more revenue to the pharmaceutical business.

The modern data infrastructure project aimed to develop our customer data infrastructure and ecosystem by dropping information silos and allowing cross-department data usage through a self-service data culture, using five steps: Extraction, Ingestion, Transformation, Delivery, and Quality.This project kickstarted the usage of cloud infrastructure in our Pharma customer, an important milestone for the company.

The Routing Matching was a data engineering initiative that aimed to help a logistic startup company. The project harvested logistics data from an API or web scraping and delivered recommendations of loads to purchase. The system was able to find good loads with trucks that were on the road in almost real-time and match them with cargo in the spot market. This enabled the company to secure loads that were 20% above the current market rate, giving them an advantage in terms of revenue.

The Cross Sell Recommendation Engine is a project that involved building a recommendation engine for a major telecommunications company in Iberia. The company was struggling to identify which customers to contact with new offers for premium TV channels, leading to outbound calls that resulted in complaints from clients. The project aimed to predict which customers were more likely to buy the new products by consuming large amounts of past customer data to build a gradient boosting model that incorporated client characteristics, current customer packs, and TV usage data.

We helped our customer migrate part of their data to Google Cloud Platform (GCP), kickstarting a cloud revolution at the company. One of the key challenges faced during the transition is maintaining the quality of data while ensuring compliance with regulatory requirements.To address this, the team wanted to build an architecture and solution that was able to address the topics of data quality, lineage, discovery and privacy.

Our data-driven sales project aimed at increasing a specific brand penetration. The team used a nearest neighbor approach and K-means clustering to rank points of sale that will likely have more demand for a premium brand, resulting in an extra 30% sales in the target group compared to the control group.

Shelf Space Optimization is data-driven sales project that aimed to understand which brands can be replaced on the shelf for new products using a simulation model. The team used a Random Forest Simulation Model to increase the average sales and profitability of each shelf in the supermarket, maximizing expected revenue by using data. The goal was to choose the most appropriate SKUs that minimized the risk of cannibalization and achieve higher ROI.

The goal of this project was to optimize promotions for specific SKUs by determining the ideal promotion percentage and timing to maximize sales while maintaining margins. The team used a boosting model that analyzed past prices and sales data to simulate a grid of prices for the future. These simulated prices were then used to predict a sellout and the chosen promotion was the one that achieved an optimal balance between market share and margin.

This project aimed to build an assortment optimization tool that provided guidance to the commercial team on which SKUs were appropriate for each client. The team used Maximum Likelihood Estimation per Cluster of Clients to determine which SKUs were most suitable for each customer group. The result was a guidance framework for assortment recommendations that improved sales performance. The team worked with 100GB of data and used DataBricks FileSystem to extract and pre-process data, perform K-Means clustering, and calculate Maximum Likelihood Estimation.

We helped our customer developing a pricing model using artificial intelligence to predict the best price for a Silicon Metal / FerroSilicon product.  The model uses past proposal data to determine the probability of winning new proposals based on price. The aim was to create a pricing governance system based on the balance between expected revenue and margin.

Our customer developed a model to predict which customers are likely to buy new financial products, using a Tree-Based Classification Model. This helped them rank customers and increase sales commissions on new products by 7.8%, particularly by mapping customers with higher likelihood of buying other products from their portfolio.