Ráčková, Laifr, T-Mobile: We didn’t expect it—teams started creating their own data products independently

April 7, 2025

T-Mobile has decided to democratize data and its usage. What strategy have they chosen? How successful have they been in breaking down data silos, managing data governance, and aligning expectations with reality? Our guests are Adéla Ráčková, Tribe Leader of the data team, and Vojtěch Laifr, Data Mesh Product Owner, from T-Mobile.

Read the podcast as an interview:

(The interview has been shortened and edited using ChatGPT.)

Ivana Karhanová: Adéla, what sparked the decision to build a data mesh?

Adéla Ráčková: Several things. On the one hand, evolving data architectures required a rethink. On the other hand, we adopted an agile operating model, which pushed us to change how we handle data and analytics.

Ivana Karhanová: But data democratization means new technologies and architecture, and, obviously, investment.

Adéla Ráčková: Exactly.

Ivana Karhanová: So, what convinced management to go ahead with the investment?

Adéla Ráčková: Our existing infrastructure couldn’t keep up with developments like generative AI. We needed modern tools to stay competitive and respond effectively to innovation.

Ivana Karhanová: I assume the agile shift was also driven by tech and market changes?

Adéla Ráčková: Precisely.

Ivana Karhanová: From a business point of view, how do you track progress? Do you use KPIs?

Adéla Ráčková: We’re early in the journey. Defining KPIs is something we’re working on this year. So far, we’re tracking how much data we’ve migrated, how many business domains are active, how many teams operate in the new architecture, and how many use cases are running there. We’re also assessing public cloud costs.

Ivana Karhanová: Are you using both private and public cloud?

Vojtěch Laifr: Yes. We started with the private cloud to test the technology. For some data-heavy use cases, the private cloud is more cost-effective. Public cloud, though, is our long-term priority due to its flexibility and access to modern tools.

Ivana Karhanová: So, will you stay hybrid?

Vojtěch Laifr: Yes. A hybrid model helps avoid vendor lock-in. But we aim to have a public cloud as the main processing and storage environment.

Ivana Karhanová: How far along are you with data mesh adoption?

Adéla Ráčková: We’ve implemented eight business domains so far, out of a planned 47. So, we’re still at the beginning.

Ivana Karhanová: Eight doesn’t sound bad.

Adéla Ráčková: It’s a good start, but it could be faster.

Ivana Karhanová: What’s slowing things down?

Adéla Ráčková: We’re a corporation, so this isn’t our only focus. We’re supporting day-to-day operations and other major transformations that compete for our capacity.

Ivana Karhanová: Did you consider simply upgrading your existing data warehouse?

Vojtěch Laifr: Our warehouse has served us for decades. But our transformation showed it’s time to move on—to centralize the analytics infrastructure and shift toward data microservices.

Ivana Karhanová: What’s your plan for expanding data mesh to the rest of the business units?

Vojtěch Laifr: We have a roadmap and plan to onboard additional domains based on new use cases. If a domain launches a new product or initiative, we aim to build it directly in the data mesh environment. Business needs drive the order of rollout.

Adéla Ráčková: And it’s not just about architecture. We’re changing the operating model, building user skills, and implementing data governance. Technology is only one piece.

Ivana Karhanová: How is the architecture structured?

Vojtěch Laifr: We use both private and public clouds. The private one runs on hardware we manage ourselves in our data centers. That gives us flexibility, but setup is slower compared to the public cloud, where provisioning is almost immediate.

Ivana Karhanová: But you offer cloud services to clients as well, right?

Vojtěch Laifr: Yes, but the data mesh infrastructure we’re discussing now is strictly for internal use.

Ivana Karhanová: Which use cases are better suited for private vs. public cloud?

Vojtěch Laifr: It depends on data volume and processing complexity. Use cases with constant high computational demands are more cost-effective in the private cloud. For everything else, we prefer the public cloud.

Ivana Karhanová: Could you give a real-world example?

Vojtěch Laifr: Our pilot was the TV Tribe, which manages the television service. It generates large volumes of data monitoring video quality. These are time-sensitive but short-lived, ideal for the private cloud due to high processing needs and limited retention.

Ivana Karhanová: And the public cloud?

Vojtěch Laifr: For virtually everything else—ideally, the public cloud is our default choice unless there’s a compelling reason to use private infrastructure.

Ivana Karhanová: Adéla, how are you onboarding users into this system?

Adéla Ráčková: We launched a Data Academy alongside the data mesh rollout. Initially, it focused on expanding self-service analytics. Now it covers training for both business and analytics teams on using the architecture, tools, and principles of data mesh.

Ivana Karhanová: If each unit is to create its own data products, don’t they need strong data skills?

Adéla Ráčková: Yes, and not all are there yet. Some teams have mature analytics capabilities; others are just starting. We’re adjusting the operating model to accommodate this variance.

Ivana Karhanová: What roles does a team need to build high-quality data products?

Adéla Ráčková: We started with reporting—basic analytics roles to deliver reports and insights. Now, we’re expanding into data science. Some teams already have their own data scientists and are self-sufficient even in AI development.

Ivana Karhanová: How do you ensure data governance guidelines are followed?

Adéla Ráčková: We’ve introduced new governance roles like data owners and stewards into each domain. We’re in the adoption phase—clarifying responsibilities, access rights, and aligning on standards.

Ivana Karhanová: But with decentralization, how do you enforce data quality?

Adéla Ráčková: We combine governance with a clear methodology. Teams are expected to follow our framework to maintain consistency and quality.

Ivana Karhanová: What motivates teams to comply?

Vojtěch Laifr: The process itself does. Access to shared data goes through a catalog and requires approval from the data owner. That forces clear documentation and encourages collaboration between data providers and consumers.

Ivana Karhanová: What’s worked well for you so far—and what hasn’t?

Vojtěch Laifr: Besides the technology, which works great in cloud environments, I’ve been impressed by team autonomy. Even small teams, when empowered, can build use cases faster than if they relied on centralized IT.

Ivana Karhanová: Can you give an example?

Vojtěch Laifr: Our TV Tribe developed tools to handle customer complaints. When someone calls about bad video quality, the team can now access transmission metrics, correlate them with network data, and resolve issues more effectively—without IT’s involvement.

Ivana Karhanová: Would that have been possible with a centralized model?

Vojtěch Laifr: It would have taken much longer and involved more bottlenecks. This way, the team owns the product and improves it as needed.

Ivana Karhanová: Are there limits to what business units can share?

Vojtěch Laifr: Each data owner decides. Their responsibility is to evaluate the sensitivity and compliance of their data. Legal and security teams provide further guidelines.

Ivana Karhanová: How do you shift people’s mindset to embrace data sharing?

Adéla Ráčková: It’s not easy. We’ve built a data culture initiative—bringing together colleagues involved with data to discuss challenges, share updates, and refine our methodology. We also host an internal data conference and run educational content to promote awareness.

Vojtěch Laifr: A great example is our “Birell and Pizza” sessions—informal meetups to talk data over food. It created an open space for discussing mesh architecture and data practices.

Adéla Ráčková: And we’re currently preparing broader internal communication about data mesh to boost understanding across the company.

Ivana Karhanová: What lessons have you learned? What would you avoid next time?

Vojtěch Laifr: Involve business from day one. Data mesh isn’t just an IT project. Use cases must be co-created with business stakeholders.

Ivana Karhanová: Who should that be?

Vojtěch Laifr: Depends on the use case. In the TV Tribe, we had colleagues from both technical and business sides involved in development.

Ivana Karhanová: Adéla, what would you do differently?

Adéla Ráčková: We’d focus more on early communication with managers. Starting with small proof-of-concepts was fine, but we underestimated the need to explain the big picture sooner.

Ivana Karhanová: Any specific moments when that became clear?

Adéla Ráčková: Usually, when we get emails showing someone doesn’t understand the process, it’s a signal we didn’t communicate well enough upfront.

Vojtěch Laifr: Another point—data mesh stands on three pillars: technology, governance, and education. These need to progress together. If we train people too early, they expect tools that aren’t ready. If we build tech first, no one uses it. Synchronization is key.

Ivana Karhanová: Beyond the TV Tribe, what else is in production?

Vojtěch Laifr: Our network teams are building use cases using technical signal quality data to predict and prevent outages.

Ivana Karhanová: And the data helps guide field teams?

Vojtěch Laifr: Yes—operations teams analyze the data and give instructions to those in the field. Ultimately, this improves service stability and customer experience.

Ivana Karhanová: How are you supporting AI initiatives?

Adéla Ráčková: The challenge wasn’t data quality, but outdated tools. Many modern AI tools run in the public cloud and weren’t compatible with our old infrastructure. Now that we’ve shifted, our data scientists have access to cutting-edge tools.

Ivana Karhanová: Could you even do AI properly without a public cloud?

Vojtěch Laifr: You could, but it would be far less efficient.

Ivana Karhanová: Any final thoughts?

Vojtěch Laifr: Decentralizing data science has improved time-to-market. Having domain-level data scientists is more effective than relying solely on a central team.

Ivana Karhanová: So, the goal is for every unit to have a data scientist?

Vojtěch Laifr: Where it makes sense, yes.

Adéla Ráčková: Exactly.

Ivana Karhanová: You also mentioned the role of data owner—who typically fills that?

Adéla Ráčková: Usually the tribe lead or technical lead. They’re supported by stewards and engineers responsible for metadata and catalog updates.

Ivana Karhanová: What’s your vision for the future of data democratization at T-Mobile?

Adéla Ráčková: We want T-Mobile and Slovak Telekom to be truly data-driven—where even non-analysts can use self-service analytics in everyday decision-making.

Ivana Karhanová: What’s the main barrier?

Adéla Ráčková: Skills and experience. Many still lack the knowledge to work with data confidently.

Vojtěch Laifr: And legacy systems. Older monolithic platforms just aren’t designed for broad user access.

Ivana Karhanová: As Product Owner, where would you like to take data mesh?

Vojtěch Laifr: Our job is to manage the shared infrastructure and methodology. Where the data products go should be up to the data owners in each domain.

Ivana Karhanová: Technically, what’s the direction?

Vojtěch Laifr: Toward a microservices architecture tailored to each domain’s needs. It’s aligned with the broader shift across the company.

Ivana Karhanová: How long will it take to bring the remaining 40 units into the data mesh?

Adéla Ráčková: Years. As my boss once said, “never”—because it’s a continuous process. It’s not just migration; it’s enabling entirely new capabilities we didn’t have before.

Vojtěch Laifr: Exactly. New products emerge, old ones retire. Over time, the old world will shrink naturally.

Ivana Karhanová: Thank you both for sharing your experience with data democratization at T-Mobile. Adéla Ráčková, Vojtěch Laifr—great to have you here.

Adéla Ráčková: Thank you for having us.

Vojtěch Laifr: Thank you!