Post 10 of 19
And what a working governance model actually looks like
Forrester analysts estimated in 2016 that between 60% and 73% of enterprise data goes unused for analytics. It is collected, stored, and paid for — but it never feeds a decision. That is not primarily a technology problem. The cost of storing data has collapsed by orders of magnitude in the last decade. The real bottleneck is governance: nobody owns it, nobody trusts it, nobody knows what it means, and as a result, nobody uses it. Data governance has a reputation for being bureaucratic, but the working version of it is anything but. It is the practical discipline that makes data findable, trustworthy, and safe to use — and in regulated sectors it is increasingly the difference between a clean audit and a regulatory enforcement action. This post lays out what governance actually entails, why so many organisations are still operating without it, and a five-step starter framework for closing the gap.
Data governance is the system of decision rights, accountability, and processes that determines how data is managed across an organisation. It is not a single document or a steering committee. It is the operating model that ensures every dataset of any consequence has a known owner, a documented definition, a defined quality standard, and a clear access policy. The discipline rests on four pillars that together cover the lifecycle of every data asset.
Ownership: every dataset has a named business owner — not an IT proxy, but the person whose decisions the data informs. They are accountable for definitions, quality thresholds, and access decisions.
Quality: agreed standards for completeness, accuracy, timeliness, and consistency, monitored continuously with automated tests and clear escalation when thresholds are breached.
Security: classification of data sensitivity (public, internal, confidential, restricted), role-based access controls, and audit trails of who accessed what and when.
Lineage: an end-to-end record of where every metric comes from, through every transformation, back to its source system — so that any number on any dashboard can be explained in minutes, not days.
These pillars are interdependent. Ownership without quality is theatre; quality without security is a leak waiting to happen; lineage without ownership is documentation nobody maintains.
The cost of poor governance is rarely a single line item. It surfaces as regulatory exposure, missed analytical value, and operational friction that compounds quietly across years. In regulated sectors — finance, energy, healthcare, anything covered by GDPR — the cost of ungoverned data has moved from theoretical to balance-sheet material. Maximum GDPR fines now stand at 4% of global annual turnover, and recent enforcement actions against Meta, Amazon, and others have demonstrated that regulators are willing to apply that ceiling.
Up to 73%
of enterprise data goes unused for analytics, by Forrester analysts' estimate (60–73%)
Source: Forrester Research, 2016€1.2B
largest GDPR fine to date, imposed on Meta in 2023 after an EDPB binding decision
Source: European Data Protection Board, 2023 ↗These figures are the headline costs. The everyday cost — analysts unable to find datasets, decisions deferred because nobody can vouch for the numbers, audit responses that take weeks instead of hours — is harder to quantify but typically larger in aggregate.
Master Data Management (MDM) is the operational backbone of governance. It is the discipline of maintaining a single, authoritative version of the core entities the business depends on — customers, suppliers, products, employees, locations, accounts. When the same customer exists under three slightly different names in three different systems, every downstream analytical question becomes ambiguous. MDM resolves these identity collisions through a combination of deduplication rules, golden-record creation, and stewardship workflows where humans adjudicate the ambiguous cases. It is unglamorous work, but it is the prerequisite for almost every other analytical capability that depends on accurate counts of customers, products, or events. In our experience, organisations that try to skip MDM and go directly to dashboards end up rebuilding their warehouse twice — once on the cracked foundation, once after the cracks become visible.
Identify the master entities that matter most to your business — typically five to ten, not fifty
Designate the system of record for each: which system holds the authoritative version, and which systems read from it?
Define matching rules: deterministic where possible (e.g. tax ID), probabilistic where necessary (e.g. name + address fuzzy match)
Build stewardship workflows: when a match is ambiguous, who decides, and how is the decision logged?
Publish the golden records back to downstream systems via APIs or scheduled feeds, so they actually get used
Governance maturity is a multi-year journey, but the first quarter of work delivers most of the practical value. The objective is not to govern everything; it is to govern the data assets that drive your most consequential decisions, and to extend the discipline outward from there. The framework below has been applied successfully in mid-market and large-enterprise engagements across the sectors we serve.
Inventory: list every dataset that informs a leadership-level decision. Stop at fifty if you can — comprehensive coverage is a year-two goal.
Assign owners: every dataset gets a named business owner with explicit accountability. No 'TBD' entries allowed in the inventory.
Define key metrics: the top twenty business metrics get a written definition, an owner, a calculation specification, and a tested SQL implementation in the warehouse.
Stand up a data catalogue: a searchable, browsable record of every governed asset — definition, owner, freshness, lineage, sample query. Tools like DataHub, Collibra, or Atlan are commodities now; the choice matters less than the consistent use.
Establish a governance forum: a monthly meeting of data owners and the analytics lead to review quality incidents, definition changes, and access requests. Light-touch, high-cadence, written-record.
This five-step model is deliberately minimal. It is what produces the 80% of the value with 20% of the effort. Once it is in place, the organisation has both the muscle memory and the credibility to extend governance to the next tier of assets.
Governance programmes that fail tend to fail for the same reasons, regardless of sector. They are launched with too much ambition, framed as a control function rather than an enablement function, and starved of executive air cover the moment they need to override a powerful business unit. The successful programmes invert all three of these defaults: narrow scope, enablement-first framing, and explicit executive sponsorship that survives the first political confrontation.
Scope creep: trying to govern everything at once produces a registry nobody updates and a steering committee nobody attends
Compliance framing: positioning governance as a control function rather than as a way to make analysts faster ensures grudging compliance instead of genuine adoption
No executive air cover: the first time governance overrules a powerful business unit, the programme needs unambiguous backing — without it, it collapses quietly
Tool-first thinking: buying a catalogue product before establishing ownership and definitions produces an expensive piece of shelfware
Forgetting the operating model: governance is an ongoing practice, not a project; without a sustained cadence and clear accountability, it decays within a year
The bottom line
The gap between data collected and data used is not a technology problem — it is a governance problem. Closing even a meaningful fraction of it requires ownership, definitions, lineage, and the institutional discipline to keep them current. The organisations that do this well treat governance as an enablement function: it makes analysts faster, regulators easier to satisfy, and leadership decisions more defensible. Delta BI builds governance frameworks alongside the warehouse and pipeline work, so the discipline is established at the moment the platform comes online rather than retrofitted years later. The cost of retrofitting is always higher than the cost of building it in.
Start with the decision
Tell us about your challenge. We will reply to arrange an assessment and identify a practical route to delivery.
Request a data assessment