Navy abstract cover with the eyebrow IFS Cloud | Data mesh and the headline Four principles, eight steps

Independent IFS Cloud practice · Data mesh series

What is data mesh: domain-owned data served as a product, in four principles and eight steps for IFS Cloud

Data mesh is a way of organising analytical data in which each business domain owns the data it produces and publishes it as a product for other teams. Zhamak Dehghani defined it through four principles; this page explains them and links every step of this series that applies them to IFS Cloud.

Key takeaways

  • Dehghani names four principles: domain-oriented decentralized data ownership and architecture, data as a product, self-serve data infrastructure as a platform, and federated computational governance.
  • A data product belongs to the domain that creates the data, and its owner answers for whether consumers can find it, read it without help and trust it.
  • In IFS Cloud, a projection (an OData service) can serve as the addressable interface of a domain’s data product. The API Explorer lists the projections in a system and provides their OpenAPI and OData specifications.
  • Access to a projection is granted through permission sets, so the access part of governance is configuration that can be reviewed, not an approval kept in someone’s mailbox.
  • The eight steps below are the order this series follows. They are not part of Dehghani’s definition, and each one serves at least one of her four principles.

1.What is data mesh?

Data mesh is a decentralized approach to analytical data: the teams that run a business area own its data end to end and serve it to the rest of the company as a product. Zhamak Dehghani introduced the idea in How to Move Beyond a Monolithic Data Lake to a Distributed Data Mesh (May 2019) and set out the principles in Data Mesh Principles and Logical Architecture (December 2020).

The model it replaces is a central data team that pulls data out of every source system into one warehouse or lake. That team owns data it did not create and often does not understand, so every new report waits in its queue, and every quality problem is investigated by people who are furthest from the transaction that caused it. Data mesh moves the ownership back to the domain and keeps a central team only for the platform and the shared rules.

For an IFS Cloud customer the consequence is practical. The purchasing team that releases purchase orders in IFS Cloud also answers for the purchasing data that finance and planning consume, including its definitions and its refresh schedule. How data mesh differs from a central architecture in more depth is covered in data mesh: difference and disruption.

2.What are the four principles of data mesh?

The four principles of data mesh, as Dehghani names them in her 2020 article, work only together: ownership without a platform overloads the domains, and a platform without governance produces data that cannot be combined.

Principle What it means Where it shows in an IFS Cloud landscape
Domain-oriented decentralized data ownership and architecture The business domain that produces the data owns it, including its quality and its meaning. Domains are drawn along business areas such as procurement, inventory or finance, which IFS Cloud already separates into functional areas. See mapping IFS functional modules to business domains.
Data as a product The domain serves its data to consumers as a product that comes with documentation and reliability commitments. A projection can be the interface a consumer calls, while the domain owner answers for its definitions and service levels.
Self-serve data infrastructure as a platform A platform team provides tooling so that domains can build and serve data products without specialist help. The integration and analytics tooling around IFS Cloud, run by one team and used by all domains.
Federated computational governance Global rules are agreed by domain representatives and the platform team, then enforced automatically by the platform. Projection grants in permission sets, plus shared data rules recorded in the enterprise book of rules.

3.What makes data a data product?

A data product is a dataset that a domain publishes for other teams and supports over time, and Dehghani’s 2019 article lists the basic qualities it must have. I add what each quality means in practice when the source is IFS Cloud.

  • Discoverable: a consumer finds the product in a catalog with its owner and description, instead of asking around. Step 7 covers this.
  • Addressable: the product has a unique, stable address. A projection endpoint is such an address on the IFS Cloud side.
  • Trustworthy and truthful: the owner commits to service level objectives for how accurately and how recently the data reflects reality.
  • Self-describing semantics and syntax: the meaning and structure are documented. The OpenAPI and OData specifications of a projection describe the syntax, and the domain still has to write the business meaning.
  • Interoperable and governed by global standards: shared identifiers and formats let data from different domains be joined, for example one supplier key across purchasing and finance.
  • Secure and governed by global access control: access is controlled per product under central policy. In IFS Cloud this is a projection grant in a permission set.

A projection is not a data product by itself. It is a technical interface over IFS Cloud data, and it becomes part of a product only when a named domain owner stands behind its content and its service levels. How to specify such a product is described in creating data product specifications in IFS Cloud.

4.How do you implement data mesh step by step?

Data mesh is implemented in eight steps in this series, ordered so that ownership is settled before tooling is bought and governance is automated before the number of data products grows. Each step has its own article.

  1. Define the vision and align the strategy. Assess where data work stalls today, tie the mesh to business goals and secure a sponsor with a budget. Serves all four principles, because it decides the scope.
  2. Identify data domains. Split the company into business-aligned domains with one accountable owner each, and start with the domains where the benefit is largest. Serves domain ownership.
  3. Form cross-functional data product teams. Put data engineers, analysts, a product owner and business experts in one team responsible for the full life of its data. Serves domain ownership and data as a product.
  4. Define and deliver data products. Give each product service levels, metadata, lineage and an interface, plus a feedback loop with its consumers. Serves data as a product.
  5. Build a self-service data platform. Provide shared tooling for ingestion, transformation, quality checks and monitoring, so domains work without waiting for specialists. Serves the self-serve platform.
  6. Apply federated computational governance. Agree which rules are global and which belong to domains, then enforce them through the platform rather than by review. Serves federated computational governance.
  7. Enable data discoverability. Register every product with its owner and metadata in a searchable catalog, automatically. Serves data as a product.
  8. Promote the cultural shift and training. Train product owners and domain teams in product thinking and make ownership part of their role. Serves domain ownership.

The phased plan for an IFS Cloud project is in data mesh implementation planning for IFS Cloud projects.

5.Which IFS Cloud mechanisms support a data mesh?

IFS Cloud exposes its data through projections, which are OData services, and two standard tools make them usable as data product interfaces.

The API Explorer gives an overview of the projections in a system, including the Premium APIs designed for integration, and provides OpenAPI v2, OpenAPI v3 and OData specifications for each. To find which projection a page in IFS Cloud Web uses, enable the Debug Console on that page: as the API Explorer documentation describes, it shows page info including the projection names. A domain team uses this to say exactly which interface its product is served through, and a catalog can register that specification instead of a hand-written description.

Permission sets are created under Solution Manager > Access Control > Permission Sets, and they grant access to projections, lobby pages, workflows and database tasks. When a consumer domain needs read access to another domain’s product, the grant is a reviewable object with a name, not an exception agreed by email.

What IFS Cloud does not do for you is the organisational part. No setting assigns a data owner, writes a service level objective or decides which definition of “on-time delivery” is the company’s. That work is the content of steps 2, 3 and 6: if nobody decides the owner and the service levels there, the projections and grants serve data that nobody answers for. If you want help with it, see my services for IFS Cloud.

6.Where do data mesh initiatives go wrong at the start?

A data mesh initiative fails early when the organisation changes the vocabulary but keeps the old ownership. These are the patterns I check for first:

  • The central team is renamed: the data team becomes the “mesh team” and still owns every dataset, so the queue stays where it was.
  • Domains copy the module list: each IFS Cloud functional area becomes a domain on paper, but nobody in the business has accepted accountability for its data.
  • Raw tables are published as products: consumers get access to an interface with no owner, no definitions and no service levels, so nobody answers for what they read.
  • Governance is a committee only: rules are agreed in meetings and checked by hand, so they hold for the first few products and erode after that. The committee itself is described in creating a data governance committee for IFS Cloud data mesh.
  • Every domain starts at once: the platform and the rules are not ready, and each domain invents its own.

7.Frequently asked questions

What is data mesh in one sentence?

Data mesh is a decentralized approach to analytical data in which each business domain owns the data it produces and publishes it as a product, on a shared self-serve platform, under global rules that the platform enforces automatically. Zhamak Dehghani defined it on martinfowler.com in 2019 and 2020.

Is data mesh a product I can buy for IFS Cloud?

Data mesh is not a product and IFS Cloud does not ship it as a module. It is a way of assigning ownership of data and running a shared platform and shared rules. IFS Cloud contributes standard mechanisms, such as projections that serve as data interfaces and permission sets that control access to them, while the ownership and service levels are decided by the organisation.

Is an IFS Cloud projection a data product?

An IFS Cloud projection is an OData service, a technical interface over IFS Cloud data, and not a data product by itself. It becomes the addressable interface of a data product when a named domain owner answers for the data behind it, documents its business definitions and commits to service levels for accuracy and freshness.

How is data mesh different from a data warehouse or a data lake?

A data warehouse or data lake is usually run by one central team that collects data from all source systems and owns it. In a data mesh the business domains own and publish their own data products, and the central team runs only the platform and the shared rules. A warehouse or lake can still exist inside a data mesh as part of the platform.

Where should an IFS Cloud customer start with data mesh?

An IFS Cloud customer should start data mesh with a written vision tied to business goals and a sponsor, then identify one or two business domains with a willing owner and a clear group of consumers. Tooling and catalog choices come after ownership is settled, because a platform bought before the domains exist ends up owned by the central team again.

8.About the author

Dariusz Myśliwiec: 18 years hands-on with IFS (Apps 7.5–10 and IFS Cloud). IFS Certified Associate Consultant. PRINCE2® 7. Based in Kraków, delivering remotely across Europe and globally through an independent practice. You talk to the consultant who does the work, not a sales layer.

Selected clients from my IFS work: Fugro · LGC · BVI Medical · Betafence (PRÆSIDIAD) · Barlinek · NGK Ceramics · Newag · Oleofarm.

IFS is a registered trademark of IFS AB; this practice is not affiliated with IFS AB.

Who owns your IFS Cloud data today?

If you cannot name the owner of your purchasing, inventory and finance data, or the definitions behind them, start there. On a call I will go through your domains and tell you which step of this series your organisation is at.

Book a call about data ownership in IFS Cloud Start with step 1: vision and strategy