← Perspectives

Series · Two strategies, one estate · Part 7 of 7

data governance · data quality · progressive rigor · traceability · AI

A data governance that gets results (II)

Quality where it matters, progressive rigor, and traceability for analytics and for AI.

Estudio RGC · 5 min

This note is part of the series Two strategies over the same estate, and it closes it. In the previous note we separated the record from the stream and left each domain owning its data. Here is the second half, and with it the point of all the governance.

Quality at the source, and measurable where it matters. Data is best fixed where it is born. Validating at the source, with simple rules and immediate feedback to whoever enters it, costs far less than rebuilding quality downstream. And where the data feeds a decision that matters, quality has to be visible and measurable: completeness, timeliness, accuracy, with an explicit threshold and a clear remediation path when it falls below. The rule is a single one: quality proportional to the decision that data feeds. Neither less where the decision is grave, nor a quality ritual where the decision does not call for it.

Progressive rigor, instead of solving everything at once. You do not start with a total governance program over all data, because that program never finishes and shows no value. You start light: visibility and contracts over the few highest-consequence flows. Then you add rigor where the consequence justifies it, measured and in plain view. Governance grows the way the estate grew, in layers, but this time guided by a criterion instead of by the week's urgency.

Traceability, for analytics and for AI. There is a purpose that orders all these pieces. Governance has to serve visibility, traceability, accountability, agility, and the secure, and when needed confidential and controlled, delivery of multidimensional information to two consumers at once: the classic analysis of always and analysis based on artificial intelligence. Both need the same thing underneath: data with context, with clear permissions, and with an origin that can be followed. A piece of data that cannot be traced is not an asset, it is a risk with good packaging. An AI model built on data no one can audit inherits that risk and amplifies it.

The close of the series. In the end, what all this is for. It serves to be able to say no to both extremes with the same criterion. No to taking the continuity strategy all the way to freezing everything, because freezing accumulates risk until it blows up. No to taking the sweep strategy all the way to replacing everything, because that flattens the domain and makes you hostage to a vendor. No to the governance that imposes on the flow the cadence of the accounting ledger, because it chokes the operation it claims to protect. The two strategies each have half the truth, and the operation needs both halves. The guiding strategy does not pick one of the two: it combines what is true in each and rejects the extreme of both. That is guided evolution, and that is what it means to have reliable guidance. In the next series we come down from the criterion to the ground: we take a single critical flow and show, step by step, how a data contract and a quality dashboard are put on it without stopping the shift.

Glossary
System of record
The official, auditable source of a piece of data: accounting, assets, contracts. It needs rigor, immutability and traceability, and can afford to be slower because its value is that no one can dispute it afterward.
Operational data stream
The data flow that is born in the field and has to move at the speed of the operation: reports, sensor readings, progress, crew events. It is useful even when incomplete and is corrected downstream; demanding record-level rigor of it suffocates it.
Data contract
Explicit agreement at the interface between two systems on the data schema, the meaning of each field, and the quality and frequency guarantees. It decouples producer from consumer, so each can evolve internally without breaking the other.
Fit-for-purpose governance (by consequence)
A governance model that classifies data by how it is used and by the consequence of it being wrong, and applies rigor in proportion to that consequence. It avoids both the lack of control where the decision is grave and the pointless ritual where it is not.
Quality at the source
Validating the data where it is born, with simple rules and immediate feedback to whoever enters it. It costs far less than rebuilding quality downstream and is the cheapest way to keep an error from propagating.
Progressive rigor
A governance strategy that starts light, with visibility and contracts over the few highest-consequence flows, and adds control where the consequence justifies it. It is the opposite of a total program that tries to govern everything at once and never finishes.
Vendor lock-in
A situation in which the operation ends up tied to the roadmap, prices and timelines of a single vendor or platform, because it bet everything on it. Standardizing the connective tissue instead of the application is what preserves optionality and avoids being held hostage.
Connective tissue
The set of connections through which the pieces of the ecosystem talk to each other: interfaces, identity, data contracts, observability. Standardizing the connective tissue, and not the application, is what lets you change any piece without redoing the whole.
AS-IS (current state)
The honest map of how the application estate stands today: what exists, what is obsolete, what depends on what. The real starting point, unretouched.
TO-BE (target state)
The direction the ecosystem is meant to be taken in, drawable and debatable, corrected as you go. Not a fixed snapshot, but a firm heading.

Next in the series

In the next series we take a single critical flow and show, step by step, how a data contract and a quality dashboard are put on it without stopping the shift.

The change implementation that drives your operation.

Contact us