Skip to content

Structure of the FAIR Business Value Framework

The FAIR Business Value Framework is structured as a three-layer hierarchical model, supported by a set of parameters and definitions.

Level 0 — Strategic Business Drivers

Four top-level drivers represent the primary dimensions of business value, namely

  • Trust — confidence in data quality, integrity, and compliance
  • Speed — acceleration of data access, analysis, and decision-making
  • Cost — efficiency gains and reduction of operational overhead
  • Effectiveness — improvement in outcomes and business impact

Level 1 — Business Value Areas

Each strategic driver is decomposed into 3 Business Value Areas, for a total of 12 business value areas at Level 1.

These represent key domains of value creation aligned with real enterprise challenges, which were derived from industry input and validation.

Level 2 — Business Value Drivers (Quantitative)

Each value area is further decomposed into quantitative drivers, for a total - currently- of 38 quantitative drivers at Level 2.

Each driver: - is defined through at least one explicit formula - includes input parameters - supports transparent calculation of value

Parameter Layer

The framework includes: - 195 parameters (as of August 2026): 183 in active use, plus 12 retained in a deprecated state so that earlier calculations remain traceable - enabling: - customization to organizational context - scenario modeling - alignment with real-world data

Reading metrics that depend on a FAIR assessment

Several drivers — most of them under Trust — take counts or rates of datasets that satisfy named FAIR sub-principles (F1, F2, I1, R1.2 and so on). The framework asks for those numbers; it deliberately does not prescribe how to produce them, and that choice has consequences worth stating plainly.

The FAIR principles are expressed as high-level guidelines rather than prescriptive technical specifications. Communities and tools interpret them differently, and the available FAIR evaluators and maturity indicators encode different interpretations. Their outputs are generally not interoperable and not directly comparable across tools. Automated evaluators at least apply identical criteria to every dataset, so comparisons within one tool are meaningful; manual and survey-based assessments are more subjective, and two assessors can score the same dataset differently.

Three rules follow, and every driver that consumes an assessment-derived input assumes all three:

  1. Hold the method constant. Measure the baseline and the FAIR state with the same assessment methodology, on the same dataset scope. A before/after pair produced by two different evaluators is not a measurement of change; it is a measurement of the difference between two tools.
  2. Read the result as relative, not absolute. These drivers quantify movement against your own baseline. They are not a FAIR score to be benchmarked against another organisation that used a different evaluator.
  3. Record which method you used. State the evaluator or survey instrument alongside the result. Without it the number cannot be reproduced or defended, and a reviewer is right to discount it.

Where a driver aggregates several tiers or sub-principles into a single figure, check that driver's own narrative for how the aggregation behaves — some are weighted indices rather than shares of a portfolio, and the two must not be quoted interchangeably.

This limitation is well documented in the literature, and reviewers of this framework have rightly raised it. Useful starting points:

Catalogues of the tools themselves: https://fairassist.org/tools and the EOSC FAIR metrics working group at https://fair-impact.eu/metrics-data.

Reading the AI-ready data drivers

The three drivers under Trust → AI-Ready Data are published in a deliberately high-level form, and readers should treat them differently from the rest of the framework.

The community does agree on the foundation. FAIR data is the basis for AI-ready data: without findability, accessibility, interoperability and reusability, a dataset cannot reliably be located, retrieved, combined or reused by a machine at all. The FAIR-for-Pharma Community of Experts equally agrees that FAIR is necessary and not sufficient. FAIR makes data usable by a machine. It does not establish that the data is fit to train or ground a model.

What remains open is how to qualify and characterise the attributes that close that gap. The Community discussed the question directly in August 2026 and the conversation is ongoing: several credible formulations are in circulation, they disagree on scope, and the group identified a prior question that shapes all of them, namely which kind of AI is in view. The criteria that make a dataset ready to train a predictive model are not the criteria that make it suitable to ground a generative one, and neither set is the same as the criteria for inference on live data.

The framework therefore does not encode a definition it cannot defend. The AI-ready training-data driver asks for the number of datasets meeting the user's own AI-readiness criteria, before and after, and prices the difference. The arithmetic is sound. The judgement it rests on belongs to the user, and the result is only as good as the criteria behind it.

Two rules follow:

  1. State your criteria alongside the result. A number produced from unstated readiness criteria cannot be reproduced or challenged, and a reviewer is right to discount it.
  2. Hold the criteria constant across the baseline and the FAIR state, for the same reason the assessment-derived drivers above require one method throughout.

In version 1.0 these drivers are most useful qualitatively: as a prompt to decide what your organisation means by AI-ready, which is a conversation most organisations have not yet had. The quantitative treatment will be revised once the community settles a definition, and this section will be replaced when it does.

Data Product Perspective

The FAIR Business Value Framework is not only a conceptual model; the framework includes of atomic definitions (in .md format) for the three levels and all parameters. There are currently simple hierarchical relationships between drivers (a tree structure). The FAIR Business Value Calculator represents one operational instantiation of this data.