01_Designing
# 01
Bulldogjob
Data science Workshops: From a Job Board to a data-driven Digital Career Consultant
Business Problem
​
Bulldogjob is a job board portal with more than 10K job offers in the IT industry on the Polish market and 160K active users. Due to increasing competition, the company realized that its current business model might not meet the users' needs and become unprofitable shortly. The company decided to change its business direction based on our data-driven approach and data science solutions based on collected but unused data.
Solution
​
We conducted a series of interactive workshops focusing on:
-
defining business goals and desired value;
-
assessment of the availability and quality of collected data through data inventory and translating business processes into data flows;
We have designed a data science projects pipeline based on Machine Learning, NLP solutions and econometric models.
Result
Bulldogjob is building a new competitive business advantage surpassing other IT job portals. Based on a solid database, we supported pivoting a simple job portal into a data-driven career advisor in the IT industry.
# 02
Resi4Rent
Data-driven decision path
Business Problem
​
Resi4Rent is a project co-created by the most experienced representatives of the Polish real estate market. R4R is an institutional platform of brand new apartments, dedicated exclusively to long-term rental. R4R has 2.3K apartments and will expand its portfolio to 10K apartments within three years. Due to the rapid development, R4R has faced the challenge of a massive flow of new unstructured data that must be properly stored and processed to ensure better customer service and the efficiency of internal processes.
Solution
​
-
We have built a data warehouse and BI reporting system based on the AWS platform.
-
Currently, the system has become the basis for building Machine Learning models for customer profiles analysis (data clustering and segmentation), reprocessing (data engineering), and real estate prices prediction (forecasting and classification models).
Result
​
Our solution is used on a daily basis by decision-makers. The reporting and analytics platform enables further digital development of R4R and ensures the highest standard of service.
02_Explaining
# 03
Kross SA
Digital Twin:
Optimization of Production and Allocation with Limited Availability of Components
Business problem
​
Kross SA is one of the largest bicycle manufacturers in Europe. The company faced global value chain challenges from the COVID-19 pandemic. The main obstacle was the lack of many bicycle components necessary to maintain the continuous production of the products. The company's goal was to develop a bicycle production and distribution plan that would maximize customers' satisfaction and match the company's resources.
Solution
​
-
We built the Digital Twin production plant model, which included an NP-hard Mixed Linear Integer Programming optimization problem with 4M decision variables and over 100M business constraints.
-
The model was implemented in Julia programming language using the JuMP package with Julia linear algebra features as well as heuristics and algebra transformations.
-
The processing environment has been run on EC2 instances at Amazon Web Services. The data processing platform has been implemented in the Python language. The data exchange process with Kross SA was constructed with the use of CSV and MS Excel files.
Effect
The computational model allowed to increase the production of bicycles by 20 percent and brought about 10 percent higher total profitability than the ERP solution. Kross SA uses the recommendations of the Digital Twin model for production and distribution planning daily. The model is constantly adapting to the changing business environment and changes in the bicycle production supply chain.
03_Predicting
# 04
Confidential
Reducing Capital Costs of Warehouse Inventory
Business problem
Inventory management in the industrial machinery industry is a process that generates significant capital costs due to high inventory levels. The company's demand for spare parts and finished goods was subject to strong fluctuation, resulting in the need to store several thousand types of components, including the production and manufacture of larger equipment. The problem was the growing working capital and the cost of financing inventories.
Solution
​
-
We constructed a series of forecasting models to estimate future demand for each SKU in the stock.
-
We built a simulation model that showed the effect of correlated different stock levels and demand uncertainty on the capital costs of the warehouse.
-
We have developed an optimized solution to find and define optimal storage strategies.
Effect
​​
The optimization models cut inventory costs by around 12 percent. This also impacted the Service Level Agreement (SLA) for parts delivery, which increased by 7 percent.
04_Optimizing
# 05
Confidential
Mathematical Modeling and Optimization of Logistic Processes
Business problem
In the realities of remote work, the problem of assigning tasks to technicians acting as one-person contractors in different cities, especially in large organizations, has arisen. Each technician worked from a unique location and had different skills and tools. The goal was to optimize logistic processes and allocate responsibilities for each technician, considering the continuous flow of new tasks.
Solution
​
​We developed a custom optimization heuristics model to reduce estimated computation time from years to minutes. The mathematical model was used to develop the schedule of the working day of the technical teams.
Effect
The optimization of logistics processes by the model allowed to reduce by 90 percent the effort related to planning technicians' work (initially performed by four people). As a result, the technician fleet was reduced by 6 percent, and the overall cost reduction was ca. 9 percent.