Data science and advanced analytics

Billions of sensor data from the new electric car are processed by the Data Analytics Platform (DAP) in less than 5 minutes

Every second of driving generates thousands of records from sensors. The goal of the project at a major car manufacturer was to efficiently process these records and present the results in managerial dashboards.

Thanks to DAP, the data is processed into a comprehensible format within minutes.

Managers adjust the analyses and reports themselves.

The rate of customer attrition after the warranty period decreased by 33%

The automotive company wanted to retain its customers and offer them additional service and sales services.

Predictive analytics, with the help of machine learning, helped to reveal the probability with which a customer would leave the authorized service.

As a result, the precisely targeted campaign was three times more successful than the campaign in the previous year.

A third of the approached customers took advantage of the offered promotions for additional services (a year earlier, only 10% of the broadly approached customers took advantage of the offer).

Customer 360 gained a richer perspective on the bank clients’ behavior

A major bank in the Czech Republic wanted to enrich its view of customer data with location attributes, and at the same time, identify branches suitable for closure.

We employed machine learning models – clustering and spatial analysis.

Thanks to big data analysis, we assisted the company in obtaining a more comprehensive Customer 360 and conducting more analyses that allow them to make better strategic decisions based on facts.

Data visualization also provided clear map outputs for business departments and managers.

Predicting account balances increased loyalty and reduced indebtedness

A major Czech bank wanted to predict the future balance of a client’s current account.

Based on historical data, we identified transaction behavior patterns and compiled an overview of future expenses. Now, the client always knows how much money remains in their account until the next income (e.g., until payday) after paying all regular/planned payments.

Standardization of projects and processes unified the approach of data science teams

At a major telco operator in the Czech Republic, we streamlined the way external and internal data scientists work on joint projects through standardization.

We started from the proven CRISP model, expanded the range of GIT functionalities used, and introduced 4 new technologies (GIT, MLflow, Papermill, PySpark). An exemplary project was created, encompassing comprehensive data science processes, including patterns/templates. We eliminated the specific procedures of individual data scientists.

Everyone adheres to standardized practices in both production and development environments. The Data Engineering department can deploy data science models more quickly. Data scientists can more easily share and collaborate on work thanks to a clearly defined structure of scripts and Jupyter Notebooks.

Estimation of client income resulted in an increased number of loan applications

A major Czech bank wanted to effectively implement digital/online processes during the COVID era, accelerate the loan process, and offer a relevant indicative loan to clients.

Using a two-phase model that draws from external and publicly available data sources, and a combination of statistical elements and machine learning, we identified significant differences between low and middle-income clients.

First, we estimated the income group affiliation, then calculated the probable income.

The indicative loan now aligns with clients’ expectations – resulting in an increased number of loan applications.

The project was completed in 6 months and was conducted entirely online, including all meetings and project handovers to the client.

Overdraft offers for corporate clients are 6 times more precise

We have delivered a propensity-to-buy model to the marketing department of a major Czech bank’s retail banking division. This model detects greater interest in a selected product or service.

• Using machine learning, we estimate the chance of setting up a corporate overdraft for clients from the segment of small and medium-sized businesses.
• In the set of clients chosen by the model, the chance of selling an overdraft in a targeted campaign is almost 6 times higher than in a blanket campaign.
• We evaluate this using Lift, a popular marketing metric. This gauges how many times higher the probability is of selling a product or service to a select group of people than to an indiscriminate group.

Assessing the client’s real creditworthiness

We helped a large Czech bank to better assess its clients’ true creditworthiness. Today, many people use the services of multiple banks, and this may not be the client’s primary bank. In order to communicate with the client appropriately and offer them relevant products, the bank needed to know its clients’ true creditworthiness.

• This knowledge makes it possible for the bank to work better with clients both in the branch and in marketing campaigns. It allows the bank to pre-select and, consequently, offer products and services that correspond to the client’s actual needs and possibilities.
• We have reduced the weight given to primary indicators such as current account balances and inbound transaction volumes. We calculate creditworthiness of the basis of a copious variety of indicators (financial, behavioral, and so on).
• We take into account client characteristics emerging from historical transaction data, use of the current account and other banking products and services, the client’s payment categories, etc.
• We have divided an apparently disparate group of clients into several homogenous sub-groups with similar characteristics, with whom personal bankers can now more easily find common ground.

Just-in-time loan offers

Together with the bank, we used several years’ worth of transaction descriptions and a number of transactions of a specific type to identify a family with children.

• Analyses of their behavior patterns have shown that customers in this segment spend the most in September. As a result, the bank offered this family a personalized loan product at the right time.
• The approach was 10 times more successful than using standard procedures without analytics on the top of Big Data. Usually, simpler rules are used for credit offers (for example, account balance and credit limit).