The Bank Vault Approach: Data Vault Modelling

Opening Scene

An old bank vault holds rows of deposit boxes, each with its own number, never reassigned even decades after a customer closes their account. A separate ledger records which boxes have ever been linked to joint access agreements. And a third set of folders, filed by date, records exactly what was added or removed from each box and when. Nothing gets erased or overwritten — if an auditor ever needs to know what was in box 142 on a specific date in the past, the vault’s record-keeping makes that answerable.

In Plain English

Data vault modelling is a way of structuring data so that the history of what changed, and when, is preserved by design rather than as an afterthought. It splits information into three kinds of tables: hubs, which hold core, stable identities; links, which record connections between those identities; and satellites, which hold the descriptive details that change over time, each one timestamped. The result is a model built primarily for auditability and resilience to change, even if it’s more complex to query directly than other approaches.

The Old Way

Traditionally, a data vault is deliberately structured around this separation of concerns:

Hubs — the deposit boxes themselves. A hub holds the core, stable identity of something the business cares about (a Customer, a Product) and almost nothing else — just enough to say “this box exists and here’s its number.”
Links — the joint access ledger. A link records that two or more hubs are connected (e.g., a Customer is linked to an Account), without storing any descriptive detail about the connection itself, only that it exists.
Satellites — the dated folders. A satellite holds the actual descriptive, changeable detail about a hub or link (e.g., a Customer’s address, which might change over the years), with every change recorded as a new dated entry rather than overwriting the old one.

This structure means nothing is ever destructively updated — a change in a customer’s address doesn’t erase the old address, it adds a new dated satellite record alongside it, the same way the vault’s ledger never erases old entries, only adds new ones.

What’s Changing (and Why AI Is the Reason)

AI-driven anomaly detection benefits directly from a model that already preserves full history. Detecting unusual patterns over time — a sudden change in customer behavior, a suspicious sequence of account changes — is far easier when every historical state is already preserved by design, rather than needing to be painstakingly reconstructed from backups or logs.
AI lineage and explainability tools depend on exactly this kind of traceable structure. When an AI model produces a surprising prediction, being able to answer “what did the input data actually look like at the time this was generated” is far easier on top of a data vault’s timestamped satellites than on top of a model that simply overwrites old values.
AI is helping manage the complexity that data vault modelling is known for. Because hubs, links, and satellites multiply the number of tables involved, querying a data vault directly has traditionally required real expertise. AI-assisted query generation and semantic layers are starting to abstract that complexity away, making the vault’s auditability benefits accessible without requiring every analyst to master the underlying structure by hand.

The Metaphor, Fully Extended

Bank Vault Element	Data Modelling Concept
A deposit box and its permanent number	A hub
The box number, never reassigned even after closure	A hub’s stable, durable identity
The joint access ledger linking boxes to shared agreements	A link table
A dated folder recording what changed inside a box, and when	A satellite table
Never erasing old ledger entries, only adding new ones	The append-only, fully historical nature of satellites
An auditor asking “what was in this box on this date”	A historical query against the data vault
The vault’s layered, multi-step record-keeping process	The structural complexity of querying a data vault directly
A trained vault assistant who retrieves the right historical record instantly	AI-assisted query generation simplifying access to vault history
A security review tracing exactly which records were affected and when	AI-assisted lineage tracing through timestamped satellite history

For Beginners: What to Actually Do

Practice separating “what something is” (a hub) from “what we know about it right now” (a satellite) — this distinction is the heart of data vault thinking.
Get comfortable with the idea that nothing gets overwritten — instead, picture every change as a new dated entry added alongside the old one, never replacing it.
When you encounter a data vault model for the first time, trace one single hub’s full history through its satellites by hand — it’s the clearest way to understand why the structure exists.
Don’t be intimidated by the larger number of tables; recognize it as the direct cost of getting full historical traceability in return.

For Practitioners and Leaders: The Deeper Layer

Data vault modelling is chosen specifically for environments where auditability, regulatory traceability, or resilience to frequent source-system change matters more than simplicity of querying — it’s a deliberate trade-off, not a universal best practice.
Because data vault preserves raw historical detail so faithfully, it pairs well as a durable “source of truth” layer feeding into a separate, simpler dimensional model (Article 6’s star schema) built specifically for end-user analysis.
When implementing AI-driven lineage or explainability tooling, evaluate whether the underlying data architecture already preserves history the way a data vault does — retrofitting that traceability after the fact is far more expensive than designing for it upfront.
As AI-assisted query tools begin abstracting away data vault’s complexity, maintain rigor around hub/link/satellite definitions regardless — the abstraction only works correctly if the underlying structure remains disciplined.

Quick Recap

Data vault modelling splits data into hubs (stable identities), links (connections between them), and satellites (changeable, timestamped details).
Nothing is overwritten — every change is preserved as a new historical record.
This structure trades query simplicity for strong auditability and resilience to change.
AI-driven anomaly detection and lineage tools benefit directly from this kind of built-in historical traceability.
AI-assisted query tools are starting to make data vault’s complexity easier to work with, without removing the need for disciplined underlying design.

Where This Fits in the Series

Article 6 covered how data gets reorganized for fast analytical browsing through star and snowflake schemas. This article covered a different design priority entirely — auditability and historical traceability — through data vault modelling. Next, Article 8 contrasts modelling for fast-moving daily transactions with modelling for broad analysis, through the lens of a restaurant’s working kitchen versus its published cookbook.

Image Instructions

Image 1 — Header banner (~1600×600px, wide format): A wide illustration of a bank vault interior, split left-to-right. On the left (“old way”), dim lighting over rows of old brass deposit boxes, a clerk manually filing a dated paper folder into a drawer — muted gray/blue tones throughout. On the right (“new way”), the same vault interior transitions into a softly glowing digital archive, with timestamped records visibly stacking behind each deposit box in layered teal light, and a faint trace-line connecting several boxes through a glowing audit trail. Bitt the Beaver stands near the vault door holding a small dated folder. Flat vector illustration style, clean lines, minimal in-image text.

Image 2 — Supporting diagram (~1200×800px): Placed directly after “The Metaphor, Fully Extended” table. A simplified, abstract infographic showing a central box labeled “Hub,” a connecting box labeled “Link” joining it to a second hub, and a stack of small dated cards labeled “Satellite” layered behind each hub, each card showing a faint timestamp. The hub and link boxes are muted gray/blue; the most recent satellite card in the stack glows electric teal, with a small magnifying-glass icon beside it representing AI-assisted historical lookup. Bitt the Beaver appears small in the corner pointing at the glowing satellite card. Flat vector illustration, clean lines, minimal text, soft glow reserved only for the teal AI element.

Cross-series visual identity note: In every image across this series, muted gray/blue always represents “the old/traditional way” and electric teal/blue glow always represents “AI/the new layer.” Bitt the Beaver, the series’ recurring mascot, appears in every header banner and most supporting diagrams, with its prop or pose adapted to each article’s metaphor while its core design stays recognizable. Style is flat vector illustration with clean lines, minimal in-image text, and soft glow/gradient reserved exclusively for AI/new elements.