Valuable
Names the consumer and the decision it supports. If nobody can name both, bin it before it becomes another unused dashboard. Example: "Sales forecast used by the weekly pipeline call".
Data mesh without the theatre
Most data mesh efforts fail
If your board is asking what is data mesh, the useful answer is not a four-pillar diagram. It is a working data product: one owner, one contract, one catalogue entry, shipped in the stack you already paid for.
Built by Data Dune
BI consultants for Pfizer, McKinsey, Siemens.
- 7
- traits of a real data product
- 3
- practices that make it real
- 6
- Monday steps before buying tools
Data mesh, governance, and delivery speed
Consistently brought strategic thinking on architecture, data mesh, governance, and cost optimization, reducing our dashboard delivery from weeks to days.
Proactively solved issues, bridged tech and business communication, and strengthened our internal capability to scale independently. I’d confidently recommend Igor to any pharma leader looking for strategically aligned, sustainable data solutions.
Simon Skurikhin
Data Engineer, Pfizer
I have had the chance to work with Igor on a few data projects for Pfizer. Its been a wonderful experience so far.
He is very insightful and an expert at what he does particularly Tableau and data pipelines. Working with him has been fun and productive at the same time. I sincerely look forward to work and learn with him more in the future as much as I can.
Helina Freesgi
Data Engineer, Pfizer
There are countless tools available today, but none of them truly add value without a passionate individual behind them like Igor. At Pfizer Tableau has proven to be an exceptional asset, bringing clarity, insight, and control across multiple areas.
It’s not just the tool itself, but how it's used with purpose and dedication that makes the difference. Thanks Igor!
Jorge Castro
Application Manager, Pfizer
The problem
Sound familiar?
The idea is not wrong. The starting point usually is. Most teams optimise for architectural purity while the people opening Tableau are still asking which number they can trust.
Six months in, page one of the contract template is still blank.
Process-heavy mesh turns into thirty-page templates, quarterly councils, and nobody brave enough to ship the first domain. Governance becomes a waiting room.
Decentralised. Federated. Polyglot. Lovely. What number do I use?
Theoretical mesh sounds clever in steering committees and useless on Monday morning. Three words your Sales VP will never say do not fix three definitions of Active Opportunity.
We'll do mesh after the warehouse migration.
Replatforming-first mesh is how good ideas go to die. This works on your warehouse, your BI tool, Salesforce, Data Cloud, Tableau, Agentforce, and whatever else finance already signed for.
Architecture
What is data mesh, actually?
Treat data like a product. Owned by the domain that produces it. Discoverable by everyone else. Then keep the implementation brutally small: catalogue, contracts, ownership, one domain at a time.
Pick one domain
Clean cut. Recognisable. At least two stakeholder groups already consuming the data.
Name one owner
A person, not a team alias. Someone who can explain, approve, and answer the phone.
Write one contract
Schema, refresh, SLA, consumers, owner. One screen. YAML in Git if you want it tidy.
Add one catalogue entry
Confluence is fine. Data Cloud catalogue is fine. The point is one link, not another portal safari.
Use the changelog and versioning
Every schema change bumps a version and lands in the changelog. Trust is not never change. It is never surprised.
Stop, feedback loop, repeat
Live with it for three weeks. Listen to consumers. Fix the rough edges. Then package it for the next domain.
Traits
The traits of a real data product
Discoverable
One catalogue entry, one link, one source of truth. People should find the product before they rebuild it badly. Example: Catalogue + contract, not Slack archaeology.
Addressable
Has a permanent, unambiguous address: a URI, a table reference, a catalogue link. People and pipelines point at it without "which one was the right one again?". Example: stable://sales_pipeline/v1.2. Same address next quarter.
Trustworthy
Breakage is announced, not discovered. Trust is not never change. It is never surprised. Example: Versioned contract + changelog.
Self-describing
Carries its own schema, field meanings, and calculation notes. Consumers understand what "Active Opportunity" means without ringing the owner at 9pm. Example: Schema + field descriptions + calculation notes shipped with the data.
Interoperable
Plays nicely with other products. Shared dimensions like customer_id line up across domains so cross-product questions don't need a migration project. Example: Same customer_id format in CRM, Service Cloud, and Finance.
Secure
Access rules live in the contract from day one, not bolted on after a board scare. PII, regulated fields, and row-level security are stated, not assumed. Example: Row-level security in the contract; PII fields tagged and masked.
What this costs your team right now.
The Sales Pipeline war story is the proof. This was not a slow query problem. It was a leadership team that quietly stopped using the platform.
- 14 → 1
- dashboards collapsed into one certified contract-backed product
- 9 → 1
- builders replaced by one named owner everyone could find
- 3 → 1
- conflicting definitions of Active Opportunity made boring
- 0 → 1
- owners to one catalogue entry with a contract and changelog
ROI
Before and after the pilot domain
| What | Before: dashboard sprawl | After: data product |
|---|---|---|
| Sales pipeline reporting | 14 dashboards | 1 contract |
| Accountability | 9 builders | 1 owner |
| Active Opportunity | 3 definitions | 1 definition |
| Findability | 0 people | 1 catalogue entry |
| Executive behaviour | Sales VP in Excel | Sales VP back |
- 6 weeks
- to first working domain
The Sales VP came back. Excel binned.
Use cases
Questions data leads face every week
Data mesh only matters if it changes decisions. Here is the practical version: fewer arguments about definitions, more confidence in the numbers, and less begging central data teams for every answer.
Director of Analytics
"Which dashboards are safe to use for the Monday revenue meeting?" Start with catalogue coverage and certified contracts. The answer becomes a short list of owned products, not a folklore tour through abandoned workbooks. Sources: catalogue, contracts.
VP of Data
"How do we scale ownership without turning governance into a ministry?" Domains own the data. Platform owns the rails. Central teams provide templates, contract standards, and guardrails; they do not approve every bloody field change. Sources: ownership, platform.
Head of BI
"Why are we still rebuilding the same metric in five places?" Because nobody made the trusted version discoverable and contract-backed. One product with listed consumers stops duplication before it becomes another Tableau junk drawer. Sources: contracts, catalogue.
Head of Data
"Are we AI-ready or just putting agents on top of dodgy data?" Agentforce, Copilot, and custom agents multiply the blast radius of bad data. Contract-backed products give agents the same trusted definitions your humans use. Sources: AI-ready, ownership.
Get started
Start with the cheatsheet. Go production with us.
The free playbook helps your team answer what is data mesh without buying another platform. The paid work turns one domain into a working proof people can copy.
Free cheatsheet
Price: Free. One-page data product checklist. Seven-trait scorecard to grade any dataset before you call it a product. Three practices that work without buying a new platform. Six concrete steps your team can run on Monday morning. Pilot-domain selection with your stakeholders and a contract template adapted to your stack are not included.
1-day discovery workshop
Pick the pilot. Map candidate domains and consumers. Find the first valuable data product. Agree owner, contract shape, and catalogue home. Works with your warehouse and BI tool. Production implementation and team coaching over six weeks are not included.
6-week pilot delivery
One domain, from discovery to handoff. One working contract-backed data product. Catalogue entry, changelog, and ownership handoff. Data Dune coaches your team while shipping. Package the playbook for the next domain. Executive readout with next-step roadmap.
FAQ
Questions buyers ask.
The practical bits: replatforming, ownership, regulated data, and how quickly you can get something real into the hands of one domain.
What is data mesh, in one paragraph?
Data mesh treats important datasets and dashboards as products: owned by the domain that understands them, described by a clear contract, and discoverable by everyone else. The grown-up version of what is data mesh is not decentralisation theatre; it is a working pattern that makes trusted data easier to find, change, and use.
We're not Salesforce-shaped. Does this still apply?
Yes. Salesforce, Data Cloud, Tableau, and Agentforce are one example stack, not the only stack. The same pattern works with Snowflake, BigQuery, Databricks, Power BI, Looker, dbt, APIs, and the BI tool your teams already open every morning.
Do we need to replatform first?
No. Please do not make this another two-year migration dependency. Start with one domain and one product on what you already own. If the pilot proves a platform gap, fix that gap with evidence instead of vibes.
How small can we start?
One dataset or dashboard. One owner. One contract. One catalogue entry. Use it for three weeks, listen to the consumers, then repeat. If nobody is already using the data, pick a different domain; empty demand makes fake products.
Who owns what: domain teams or central platform team?
Domains own the data because they understand the meaning, trade-offs, and changes. The platform team owns the rails: templates, infrastructure, shared governance, observability, and the standard that keeps domains from inventing chaos in five dialects.
How is this different from a data catalogue tool?
A catalogue tool is a place to list things. Data mesh is the operating model that makes the listed things worth trusting. You still need ownership, contracts, consumers, versioning, and a habit of announcing changes before people find broken dashboards.
What about regulated data, GDPR, or SOX?
Regulation makes ownership and contracts more important, not less. A contract should record classification, allowed consumers, retention expectations, lineage, and change controls. The pilot does not bypass governance; it makes governance concrete enough to inspect.
How long until we see results?
A focused workshop can pick the right pilot in a day. A six-week delivery can produce one working domain with a contract, owner, catalogue entry, and handoff. That is fast enough to prove the pattern and slow enough not to ship bollocks.
Next step
A pattern takes weeks. Trust takes a working contract.
Take the cheatsheet. Pick one domain. If you want the first data product shipped properly, Data Dune will help you build it with your team, in your stack.
No replatforming sermon. No governance theatre. One useful domain first.