Skip to main content

AI-driven personalization improves advisor outreach


Of branch advisors clicked on their personalized tips


Of advisor calls that were enhanced by personalized tips resulted in a personal meeting


New personal client meetings from personalized tips within the first three months

PARTNERS: DataSentics
CLOUD: Azure

To streamline customer service and boost customer engagement, Česká Spořitelna, part of Erste Group, recognizes the importance of empowering branch advisors with personalized conversation starters that cater to individual client needs and interests. As a leading financial institution with roots going back to 1819, the bank has continually emphasized the centrality of the client-advisor relationship. With the increasing complexity of financial products and a more demanding client base in the modern era, the organization was facing a challenge. Access to data was scattered among multiple teams using different tools (Oracle DB, Cloudera, Kafka), with no efficient way to leverage all data sources simultaneously. Manually analyzing each client’s transaction history to decipher their current requirements, life circumstances and preferences became untenable. This inefficiency subsequently led to increased workloads for branch advisors and lower conversion rates from outgoing calls due to the use of generic conversation openers.

Combining transactional, call center and digital data to enable advisors to talk to their clients at the right time about relevant topics

To enable personalization, the company embarked on a project that began by asking proactive advisors to create a wish list of notifications for more effective client interactions. Building on this foundation, a joint team of data scientists from Česká Spořitelna and DataSentics, an Eviden business, utilized the custom product Persona 360, built on the Databricks Data Intelligence Platform and deployed on Azure cloud. This integration capitalized on a suite of Databricks products such as Delta, Unity Catalog, Repos, Workflows, Feature Store, Dashboards and Alerts.

The Databricks Data Intelligence Platform centralized data from various domains and tools into a single coherent platform. With the governance provided by Unity Catalog, it ensured all team members had tailored access to data based on their roles. Data engineers could now interact with both batch data from Oracle and Cloudera, and streaming data from Kafka. At the same time, data scientists could draw from this wealth of information for their models, and data analysts could connect to it via the Tableau connector. Each team member continued to use their familiar tools: data engineers with SQL and workflows, data scientists with Python and notebooks, and data analysts via SQL editor and Tableau. Further enhancing the development process, Databricks Repos integrated with Github allowed the team to adopt best software engineering practices for data solutions.

The refined process produced a range of notifications, both deterministic insights — like a customer’s recent website visit or a significant incoming payment — and probabilistic insights, such as a penchant for travel, recent changes in app usage, income shifts or unexpected website behavior. These notifications are then channeled to the existing CRM system. Here, branch advisors are presented with a comprehensive list of notifications for their portfolio, while also having the option to view specific notifications for individual clients directly on their CRM page.

Using the Unity Catalog by Databricks

Data governance in the company’s architecture was essential. Without Unity Catalog, the platform built around Databricks wouldn’t align with their security policies. It not only allowed them to federate the data but also provided a means to govern it in a straightforward and effective manner. Before Unity Catalog, custom tooling for each organizational unit was required, but it wasn’t as efficient. The main use case for this project was utilizing Unity Catalog for documentation purposes, tracing data origins and troubleshooting issues in our pipelines.

Why Databricks?

“Thanks to Databricks, data scientists from Česká Spořitelna (part of Erste Group) who were used to working with SQL were able to onboard the project in a short amount of time and deliver results with ease,” said Lukáš L., Senior Data Scientist at DataSentics. “Databricks lakehouse architecture helped us to process and transform great amounts of transaction tables, call center data and Google Analytics exports efficiently and make the outputs reusable for training machine learning models in the future.“

The significant impact of personalized communication

Automated alerts for bank advisors are proving to be very effective. Half of the calls triggered by an alert lead to a client meeting, and another 27% result in clients setting up a follow-up call. Within just the first three months after launching, these alerts have driven a noticeable increase in meeting bookings, adding 1,000 more client meetings to the roster. Databricks has played a pivotal role in this process, aiding Česká Spořitelna’s mission to support their customers’ financial health by tapping into both historical and real-time data to provide valuable advice.

Using batch and real-time calculations, several marketing personalization strategies were developed on the Databricks Feature Store. This means that not only do advisors save time when preparing for meetings — instantly accessing the most relevant topics for each client — but the bank as a whole also benefits. The efficiencies gained lead to faster product launches and fewer hours spent on manual tasks.

“The introduction of Workflows by Databricks has been a game-changer,” said Tomáš Bouma, Data Engineer at DataSentics, “transforming how teams work together and boosting overall productivity. This combination of improvements directly contributes to heightened customer satisfaction and opens up more opportunities for upselling.”

Looking to the future, Česká Spořitelna is excited to make even greater use of streamed data, something that their previous on-premises solution couldn't accommodate.