We start with workshops and problem-solving sessions to thoroughly understand the challenge, define the scope and KPIs against which we will measure the success of the solution.

We identify various data sources, connect them, verify our understanding with the business unit and run exploratory data analysis to check the initial hypothesis and build up our business understanding with data-based reports.

We build up data pipelines that help us process and combine multiple data sources that are needed for the problem we are solving. We clean the data and prepare it in the format needed for modeling.

We develop machine learning algorithms for a specific task, optimise the parameters, verify multiple approaches or model types and train them on the data prepared in previous steps. We also take into consideration the complexity of the model.

Based on previously defined KPIs, we evaluate the model’s performance, run error analysis and identify areas for improvement and refinement of the model. This also helps us to better prioritise our work.

Once we select the final model, we deploy it in production and monitor the results. It is important to verify the results over time to make sure that the distribution of the input data and results do not shift.