BI product — Data

What the product guarantees about the data it handles.

≈ 4 min readReading

On this page

Data sources and flows#

Where does the data come from, and how does it arrive?

Level	Signature	Criteria to meet
0 · Handcrafted	Files built by hand and kept locally.	Starting point: no criteria.
1 · Locally tooled	Local data extracted from an ERP (or another management system).	The data comes from an extraction of the source system, not from manual entry. The extraction is described (query, scope, date).
2 · Shared	Scheduled extraction to a common source.	The extraction is scheduled. The data is deposited in a single shared location. The date of the last update is visible to users.
3 · Industrialized	Industrialized flows: recovery after failure, logging, volumes tracked.	Each flow is logged (start, end, volumes, errors). An interrupted flow can be restarted without duplicates or loss. Expected volumes are checked at every load.
4 · Integrated and governed	Multiple sources consolidated in a common model.	Several sources (internal and external) are consolidated. Keys and reference data are common across sources. Discrepancies between sources are detected and handled.
5 · Autonomous	Continuous flows and data contracts: the product adapts to changes in the sources.	Data arrives continuously (near real time) when the need justifies it. A data contract (schema, freshness, quality) is defined with each source. A schema change in a source is detected before it reaches users.

Effect on the user

The user no longer has to wonder whether the data is up to date, complete, or where it comes from.

Can the data be trusted, and how do we know?

Level	Signature	Criteria to meet
0 · Handcrafted	No checks: errors are noticed when they become visible.	Starting point: no criteria.
1 · Locally tooled	Manual visual checks by the author.	The author checks totals and outliers before sharing.
2 · Shared	Defined and documented check rules.	The expected quality rules (permitted values, uniqueness, consistency) are written down. Known anomalies are listed.
3 · Industrialized	Automated checks and anomaly reports.	Checks run automatically at every load. An anomaly report is produced and read. A blocking anomaly prevents publication.
4 · Integrated and governed	Quality measured by dimension, visible to users, with data owners.	Quality indicators by dimension (completeness, validity, consistency, timeliness, uniqueness) are tracked over time. The quality level of a dataset is visible to users. Each dataset has an owner responsible for its quality.
5 · Autonomous	Automatic corrections and prevention at the source.	Known corrections are applied automatically, in a traced and reversible way. Recurring anomalies are fixed at the source, not downstream. Quality thresholds drive publication (automatic or suspended publication).

Effect on the user

The user no longer has to check the figures themselves: trust is carried by the product.

Does each indicator have a single, shared and controlled definition?

Level	Signature	Criteria to meet
0 · Handcrafted	Each file recalculates its own indicators.	Starting point: no criteria.
1 · Locally tooled	Formulas copied from one file to another.	The indicator formulas are described somewhere.
2 · Shared	A shared model: common tables and relationships.	Reports are based on a common model. The main indicators are calculated there only once.
3 · Industrialized	Indicators defined once, glossary published.	Each indicator has a written definition (formula, scope, unit). A glossary is accessible to users. A given measure has only one name and one definition.
4 · Integrated and governed	Governed semantic layer: reusable, versioned, multilingual.	The model is reused by several reports and applications. Changes to the model are versioned and announced. Labels are available in the users' languages.
5 · Autonomous	Indicators “as code”: tested and deployed automatically.	Definitions are stored as code, with history. Each change triggers regression tests on the indicators. The impact of a change (affected reports, users) is calculated automatically.

Effect on the user

The user no longer has to wonder which definition applies: one figure, one meaning.

Can the product be understood, used and audited without its author?

Level	Signature	Criteria to meet
0 · Handcrafted	Tacit knowledge, held by the author.	Starting point: no criteria.
1 · Locally tooled	Personal notes.	The essentials of how it works are noted somewhere.
2 · Shared	Single, summary, accessible documentation.	Documentation exists in a known location. It distinguishes usage (user) from operation (technical).
3 · Industrialized	Documentation updated with every version, data dictionary.	The documentation is updated with every version. A data dictionary describes each field (meaning, source, rule). Changes are recorded in a change log.
4 · Integrated and governed	Catalog and end-to-end traceability, glossary.	A catalog lists the data and its owner. You can trace a displayed figure back to its source. You can measure the impact of a change of source or rule.
5 · Autonomous	Documentation generated automatically from the model, checked at every delivery.	The technical documentation is generated from the model and the metadata (reverse engineering). Outdated or incomplete documentation blocks delivery. The documentation is translated and published automatically.

Effect on the user

The user finds on their own the answer to “what does this figure mean and where does it come from?”.