What Happened to Data Products? | Jarkko Moilanen

Open original

Data mesh hype cooled, but the underlying need for findable, trusted, owned data hasn't gone away. The author argues data products are shifting from a full enterprise restructuring into a practical discipline built into existing platforms, catalogs, and governance work.

  • Data mesh asks for a big operating model change, and adoption stalled: a 2024 BARC study of 197 professionals found 20% had it underway and 24% planned it, down from 35% planning it in 2023.
  • A data product is defined by a known consumer, a stated purpose, an accountable owner, and measurable use, not by the asset format itself.
  • KPMG's 2025 survey of 250 executives found 92% saw data products as critical, yet only 35% reported extensive value, a gap between intent and execution.
  • AI agents raise the stakes: without clear ownership, meaning, quality, and permitted use, they make weak decisions faster, so governed, portable product definitions matter more.

Data mesh got a lot of attention, but the author's point is that the hype cooled because reorganising a company is much harder than redesigning a data architecture. Data mesh asked business domains to own the meaning, quality and lifecycle of their data, with a central team providing shared services and controls. That is expensive to fund and hard to operate. The 2024 BARC study of 197 data professionals found 20% had data mesh underway and 24% planned it, down from 35% planning it in 2023. The underlying need, though, hasn't gone anywhere: finance still needs trusted profitability data, commercial teams still need approved customer views, and operations still needs reliable performance figures. The more useful shift is that "data product" has become a practical delivery discipline rather than an architecture debate. A table, dashboard or API isn't a data product on its own. It becomes one when a known consumer uses a governed output for a stated purpose, an accountable owner stands behind it, and its value can be measured over time. Gartner's 2025 guidance says to start with business-critical use cases, build reusable minimum viable products and agree KPIs between producing and consuming teams. The author draws a useful line between catalogs, which describe the estate, and product management, which decides what to invest in, what the product promises and when to retire it. The gap is visible in KPMG's 2025 survey of 250 executives: 92% called well-built data products critical, yet only 35% reported extensive value from their initiatives. The usual failure is not the technology but unanswered questions about which decision the product supports, who owns the outcome and how anyone knows it's being used. AI is what makes all of this urgent. An analyst who finds an odd metric can investigate or ask a colleague, but an AI agent will act on it at scale, so missing context becomes weak decisions made faster. Agents need to know which source is authoritative, what a measure means, how current it is and what they're permitted to do. The author cites a 2026 Data Product MCP paper describing an agent that discovers, requests and queries data products under enforceable contracts, though its evaluation with 16 governance experts is early design work rather than proof of adoption. The author's practical recommendation is to avoid declaring a data-mesh transformation and instead pick one important decision, workflow or AI use case, then define its consumer, owner, output, quality expectations and usage measures. To keep such products portable across warehouses, catalogs and clouds, the author points to the Open Data Product Specification, a machine-readable definition of purpose, owner, access, quality and governance that catalogs and marketplaces can share rather than rebuild.

Ask about this article