Minimum Viable Data Architecture Rests on Eight Domains
This is Article 2 of 4 in the Q-Lana Data and Knowledge Management series. A minimum viable data architecture rests on eight data domains and six components that connect systems a bank already owns. Integrate first, automate second and scale last.
A minimum viable data architecture rests on eight well-defined data domains, not on a large technology programme. Ask a lender what it knows about a borrower, and you get a list of documents. Ask what it can prove on a Tuesday afternoon, without phoning anyone, and the list gets much shorter. That gap is a problem of categorization, not collection. This article sets out the eight domains, the six components that hold them and the order in which to build.
Categorization Decides What a Lender Can Prove
Every client relationship, credit proposal and monitoring report already creates the information. Whether it becomes an asset or a liability depends on an early decision. What kind of data is this, who owns it, and which system holds the correct version?
Many institutions postpone that decision until the portfolio seems large enough to justify it. By then, definitions have drifted, and records have scattered. The asset they meant to build has already lost its integrity. The banks that do well define the data model early. They standardize numbers and narrative together and assign an owner to each domain from the start.
The Eight Data Domains Behind a Minimum Viable Data Architecture
Eight domains give a full view of a client and a credit relationship. They are not equally difficult, and they are not equally valuable, which is why naming them in order is more useful than listing them.
The anchor is customer and counterparty master data
Legal entities, ownership structures, ultimate beneficial owners, group linkages, sector classification, geographic footprint. Everything else connects here, which is why the unique identifier matters more than anything else in this article. Get the identifier wrong, and every downstream domain inherits the error.
Two domains carry what the client tells you
Financials are the quantitative backbone: audited and management statements, projections, bank statements, tax filings, ERP extracts. Relationship and interaction data is the qualitative side: relationship manager notes, meeting minutes, site visits, call summaries, correspondence. The first supports ratio analysis and trend detection. The second carries the opportunities and the early warnings that numbers alone never reveal.
Two more carry what the institution did about it
Credit process data is the formal lending workflow, from application through analysis, committee memo, covenant and collateral to the decision itself. It is the institutional memory of judgment. Collateral and valuation data records the pledged assets: type, location, enforceability, inspection history, revaluation records.
One carries what actually happened
Behavioral and performance data is repayment patterns, arrears, restructurings, waivers, facility utilization.
And two set the context and the boundary
Operational and third-party data brings the outside world in: credit bureau reports, business registries, ESG self-assessments, geospatial and sector data, trade and invoice datasets. Policies and risk appetite data set the limits by segment, sector and obligor, the pricing grids, the concentration thresholds and the risk appetite metrics. Linking every exposure back to those parameters is what makes consistent portfolio steering possible at all.
Behavioural Data is the Most Underused Domain

Behavioural and performance data predicts credit outcomes better than any other domain. It is also the one lenders use least.
Why it goes unused
The reason is structural. It accumulates quietly, in systems built for servicing rather than for analysis. Nobody has to compile it, so nobody does. It sits in the loan management system as a by-product of collecting instalments, and it is the closest thing a lender has to ground truth about its own borrowers.
It is the domain competitors cannot buy
Bureau data is available to everyone who pays for it. Sector indices are public. The record of how a specific borrower has behaved through a specific cycle, under facilities the institution itself structured, exists in exactly one place. An early warning model trained on that record will outperform a bought model, not because the mathematics is better, but because the borrower universe is the institution's own.
That is the argument for capturing it deliberately rather than incidentally.
Key insight
Early discipline creates compounding advantage. Each domain adds proprietary knowledge competitors cannot replicate. Systems come and go. The data logic must endure.
The Trap of Starting Too Big
When banks decide to improve their data, the instinct is scale. They plan an enterprise-wide programme with dozens of systems, perfect integration and full coverage before launch.
That ambition usually produces paralysis, an inflated budget and a solution that is out of date on arrival. Requirements are written against today's systems, the build takes two years, and the bank has moved on.
The better path is smaller and less satisfying to present to a board. A minimum viable data architecture is a lean, purposeful setup that delivers value now and lays the foundation for later. It rests on three words.
What minimum viable data architecture means instead
A minimum viable data architecture is smaller and less impressive in a board presentation. It is a lean setup that delivers value now and lays the foundation for later. It rests on three principles.
|
Principle |
What it means |
|
Clarity |
Where each piece of data comes from and what it represents |
|
Connection |
Linked systems and teams, so manual file transfers and
version confusion disappear |
|
Control |
How information moves, who can change it and how each
decision traces to its source |
Using parts you already own
None of this requires replacing the core banking system or choosing a single vendor. Most banks already own a core banking or loan management platform, a CRM and some kind of document store. What they lack is those parts connected into one view.
Six Components of a Minimum Viable Data Architecture

Six components form the backbone. They connect existing systems and release data for reporting, analytics and decisions, without forcing a replacement programme.
A system of record
One truth, not many. Define which system is authoritative for which data. The core banking and loan management platform anchors customer and facility data. This is a decision, not a technology, and it is the one most often left unmade.
A master and metadata layer
A common language. A lightweight master data management layer unifies the key entities, customers, facilities and collateral, alongside a business glossary that defines each field and each metric.
An event log
Traceability by design. Who did what, when, on which record. Approvals, covenant changes, waivers, site visits, rating updates. This is the component that answers the reproducibility question from the first article in this series.
An analytical store
One place to think. A warehouse or lake house consolidating cleansed data for reporting, risk monitoring, early warnings and scoring, separate from the systems that run the business.
A document and knowledge hub
A unified store with strict version control and full-text search across relationship manager notes, committee memos and legal documents.
Pipelines and APIs
Automate what matters most. Scheduled flows from core to analytical store, CRM to interaction data, document store to references. Manual file movement disappears, and updates arrive close to real time.
Integrate First, Automate Second, Scale Last
The order of work decides whether the programme survives its first year.
Why is a Sequence Important?
- Integrate first. Connectivity is what unlocks visibility, and visibility is what buys the political capital for everything after it.
- Automate second. Automate what matters most rather than what is easiest.
- Scale last. But design for it, so that the third year does not require demolishing the first.
Institutions that follow that sequence achieve more in three months than they do in years of scattered technology projects. Institutions that invert it, scaling before they have integrated, spend more and show less.
Four early wins in the first quarter
Four early wins are available inside the first quarter, and each of them is visible to someone who was skeptical. A true Client 360 view. A portfolio risk appetite dashboard. A standardized underwriting pack. Consistent exposure and concentration views across the book.
None of these needs a finished architecture. All of them need one started correctly.
Building Lean on Purpose

A well-designed architecture does not attempt to anticipate every future need. It makes continuous improvement possible: each new product, process or insight plugs into a framework that is already clean, connected and controlled.
That is a different design goal from completeness, and a harder one to hold under pressure, because completeness is easier to describe in a steering committee.
Build for today, with a clear path to tomorrow. The next article deals with the discipline that keeps it standing: who owns the data, which rules are enforced without exception, and the point in every implementation where institutions quietly give up.
Next in This Series
The next article covers the discipline that keeps the architecture standing. It deals with data ownership, the rules enforced without exception and the point where most implementations quietly give up.
Check Your Data Foundation
The eight domains and the six components make a short checklist. It returns a coverage view across the domains, an architecture readiness score, and the output that matters most: which of the four early wins is actually unlockable next, given what is and is not in place.
The Q-Lana white paper Data & Knowledge Management for SME and Corporate Lenders covers this in full. It sets out the domain definitions, the architecture and the sequencing rules.
Download the white paper
All Articles in This Series
- Usable Data Is What Most Banks Are Missing
- Minimum Viable Data Architecture Rests on Eight Domains (this article)
- Data Governance: Small Rules, Ruthlessly Enforced
- AI in Lending Amplifies Whatever Data Is Already There
Follow Q-Lana
One short email a week on credit risk, lending and impact finance, free. Or browse more guides in the Q-Lana Knowledge Center.
Get Q-Lana Weekly
One short email a week on credit risk, lending and impact finance, free. Or browse more guides in the Q-Lana Knowledge Center.