H&M Group’s Data Mesh Journey: From Ideation to Implementation

Data mesh was coined by tech leader Zhamak Dehghani in 2019 to refer to an agile, domain-based approach to setting up data architecture. As organizations become more data-driven, it’s time for IT and digital leaders to explore data mesh and its advantages. Erik Helou, the former Lead Architect at H&M Group, shares pertinent insights on how the organization utilizes data mesh, its benefits, and the challenges encountered during implementation.  

*This article is a recap of Erik Helou’s presentation at the session, Decentralized Approach to Becoming an Agile Business. 

 

Why H&M Group Adopted Data Mesh

H&M Group found itself in the same spot as many organizations, experiencing many iterations of data platforms, ways of working with data, and centralized teams. “In those days, we spent four or five years on new AI efforts, working with data in different ways,” Helou explains. H&M Group’s data systems eventually had to scale with the organization’s growth, and data mesh enabled that.  

Data mesh addressed the following needs: 

• An accelerated growth of data: “The organization experienced accelerated growth of data, resulting in the need to use data in newer and faster ways.” 

• A rapidly changing industry that pushed demand to scale AI and data: “The industry was changing as our company was evolving. We needed to onboard, facilitate, and make use of data capabilities at a scale and speed we hadn’t seen before.” 

• Business knowledge ownership inhibited growth: “A big thing that inhibited growth and expansion in the data space was how to scale and manage business knowledge ownership of the data; and the ways of interpreting the data.” 

• Business tech organizations move towards product centricity: “We saw a shift towards product centricity in the business development and technology departments, rather than the usual IT delivery way of operating.” 

 

H&M Group’s Interpretation of Data Mesh

H&M Group was drawn to data mesh, and more specifically, the distributed domain model because it solved many of the organization’s pain points. “The idea of this model is to define a map of your business in domains, subdomains, and data products,” Helou explains.  

“In a typical retail case, you would have the business data domain of sales, which is a difficult source-aligned product domain. All source-aligned data domain products should be the most correct and easy-to-use window into the operational business. That way, anyone in the company who needs the data view on insights, discovery, or operational software development knows where to go.”  

Helou explains that data mesh creates an official domain of data products that work well together. There is also a team behind them that the staff can contact to guarantee the operational qualities of that data. “It’s easy to find your way in a data mesh that represents the entire company’s activities,” he adds.  

Next are the aggregate products or the consumer-aligned products, which are refined or engineered products that transform operational business data for different purposes. Consumer-aligned products focus on specific needs in the business. In addition, the aggregate is a typical customer 360 product that takes information from a number of data products with a wealth of business knowledge into something more refined that can be used by people or systems.  

 

The Four Pillars of Data Mesh

1. Domain-oriented ownership: This enables the distribution of ownership, development, and maintenance of hundreds of data products in an enterprise like H&M Group. Employees need to understand their domain ownership and data product ownership. “That’s the operational model and culture that needs to be established in the organization,” Helou says.  
2. Data as a product: “Teams supply their data products to present what happens in the business domain they are responsible for. They serve that to themselves and to the rest of the company as an important asset.” 

To make that happen in a structured and sustainable way, there are technological tools that need to be in place: 

3. Self-service data platform: “This is so we don’t have 200 product teams purchasing software and designing things in completely different ways. There’s a lot to gain in terms of costs and interoperability between different products if they can be on the same platform and use the same tools.” 
4. Federated computational governance: This is important for the sustainability, compliance, and quality of H&M Group’s data products. “It’s things from the legal parts to logging to data quality to discoverability. You have one catalog where you can browse, discover, and understand the data assets you have rather than browsing through a big database.” 

   

Features of the Data Mesh Approach

• Enable all teams to autonomously ingest, share, refine, and consume data: “The autonomous way of working was the most important thing for us. The self-service data platform is key to distributing the work. We wanted all new and existing data product teams to be enabled on their own without too much dependency on bottlenecks, like how data engineering teams used to be in a data warehouse space. Everyone should be able to find and consume that data autonomously rather than asking for permission or knowledge. It needs to be self-explanatory and as automated as possible.” 
• Provide a self-service UI based on standardized infrastructure, modules, and platforms: Data products are shared through friendly UI which provides standardized infrastructure components. “You should be able to create the infrastructure and data pipelines on top of what’s already there for an entire company. Any additional platforms or modules that you need go through the Self-Service UI.” 
• Provide monitoring, logging, and alerting: As a background capability for the self-service UI, data mesh provides standard monitoring, logging, and alerting systems to maintain consistency among product teams, as well as ensure data quality and operational quality. 

 

Data Centralization is Key

“We want to stay clear of centralizing too much to reduce bottlenecking. Central engineering in data warehousing teams needed to do everything for all the source systems to be represented in the data warehouse. Business teams had to go through the heavily loaded, bottleneck engineering team. Data mesh allows us to get away from that,” Helou says.  

What H&M Group centralizes:  

• Ensure a holistic one-stop-shop data use experience: “This is the data catalog we need to offer centrally because it has to be federated. Everything needs to be collected in one place.” 
• Apply governance to all aspects: “We need to centralize the governance, legal aspects, security, compliance, and data quality.” 
• Establish reusable accelerators and toolkits: “These are the building blocks for all the product teams to establish their data pipelines. That way, the teams don’t have to build specific tools because they cost a lot of time and risk. This makes it easy to fix bugs across a lot large number of data products at the same time.” 
• Create schema, contract, and landscape fundamentals: “We need also to centralize hygiene factors like schemas, data contracts, and landscape fundamentals to enable stable and trustworthy operations at this scale and manage changes in the integration points.” 
• Massive communication effort: “The central team needs to continuously talk about how we use data mesh and why we use it.” 
• Documentation is key: “We need to centralize documentation describing the many data products we offer.” 

 

Data Mesh Lessons and Challenges

• Mindset shift and upskilling of employees: “It’s a cultural thing to distribute the ownership of data creation, data knowledge, and data use in this way, which is something that is very attractive for a lot of people. But it’s still a big shift and we need to educate each other on how the organization uses data mesh. There will be upskilling, and the addition of new technical tools.” 

• Decentralize technology (no central DevOps): “The decentralization of technological systems to find the right balance on what to centralize and decentralize.” 

• Architectural and technological change: “It has an impact on the entire data landscape on how things work in the operational backbone and how data flows.” 

• Manage legacy platform deprecation: “We need to take legacy into account. There are many well-functioning data solutions for reporting analytics that have to keep running. But we also have to steer our investments towards data mesh products and find our transition from legacy systems to this new platform.” 

• Onboard key data procedures early: “It’s important to onboard key data early and find a handful of data sets early that enable teams to migrate from legacy data solutions to the data mesh reality.”  

• Focus on business priorities: “What’s your strongest business case, either that you need to keep operating or something completely new? What data needs to be there? Then, you can get business stakeholders to buy in and find experts that can guide you to create those data product teams.” 

 

The Future of Data Mesh at H&M Group

“In the long run, we hope to gain agility and autonomy. As an online retail business, we need to keep scaling fast and adapt quickly to changes in the business world. We also need to get rid of any bottlenecks in this data expansion. Data mesh provides a strong foundation to enable innovation around data. It allows business analysts to uncover new potential, insights, and value propositions. It also lays the foundation for us to continue growing in the digital era and in the data-centric way of working.”