Skip to main content
AI-readiness assessment framework · operational companion

Operationalizing the AI-readiness assessment

The AI-readiness assessment framework scores a national statistical organization on twelve dimensions across two pillars. This page sets out, for each dimension, the program workstreams, open-source tools, reference implementations, and cookbook recipes that support progress to a higher maturity level. It also identifies the evidence that each resource produces for the assessment form.

Assessment frameworkCookbook self-assessment
Dimensions
12
Dimension questions
69
Addressed by a program component
49
Cookbooks, chapters
10, 82
Purpose and structure

Using the assessment results

  1. 1

    Assessment

    Complete the assessment framework to obtain a maturity level (A to D) for each dimension. The cookbook provides a short self-assessment for a first orientation.

    Assessment framework →
  2. 2

    Mapping

    For each dimension where the gap between the current and target levels is largest, consult its entry below. The entry lists the program resources that apply and the questions that remain the organization's own.

    Dimension entries →
  3. 3

    Implementation

    Apply the cookbook recipes and tools identified for the dimension. Each produces a document, dataset, or measurement that can be attached to the assessment form as evidence.

    Cookbook →

The program's contribution is concentrated in Pillar II, the readiness of data products. In Pillar I it addresses specific questions: governance instruments (a policy, a register, a map of model-assisted tasks, a statement of model use), licences and terms, skills through the cookbooks and their reading paths, computing and security for model servers, privacy-preserving techniques, evaluation, costs, and partnerships. Strategy, oversight bodies, legal frameworks, connectivity, and funding structure remain the organization's own decisions. The matrix below records this.

Coverage by dimension

Program support by assessment dimension

Each cell indicates the extent of one kind of support for a dimension. Selecting a dimension opens its entry below.

DimensionOpen-source toolsToolsMethods and researchMethodsCookbook recipesRecipesReference implementationsReferenceStandards and guidanceGuidanceCoverage
Pillar I · Institutional readiness for AI adoption
partial
partial
partial
partial
partial
partial
Pillar II · Readiness of data products and services for AI
direct
direct
direct
direct
direct
direct

substantial   some   none

Dimension entries

Program resources and evidence for each dimension

Pillar II · Readiness of data products and services for AI

2.1 Metadata standards and ontologies

Program components address most questions

The program publishes the metadata schemas used by NADA and the Metadata Editor, develops methods to complete and check records with AI under curator review, and works on classifications, concept links, and a question bank. The cookbook translates these into a three-layer check and templates that an organization can run on its own catalog.

Questions in this dimension 5 of 5 addressed by a program component

  • 2.1.1Adoption of metadata standards
  • 2.1.2Classifications, concepts, vocabularies
  • 2.1.3Linkage between data and metadata
  • 2.1.4Machine-readable publication and web discoverability
  • 2.1.5Provenance and lineage

Program resources

Open-source tools
  • Metadata EditorTemplates, validation, export to SDMX, schema.org, Croissant, DCAT; publishes to NADA
  • NADA catalogCatalog pages with schema.org markup, search, a catalog administration REST API, managed access; semantic search and MCP access through NADA AI, being integrated into NADA
  • Metadata AugmentationDDI-style variable groups from data dictionaries
Methods and research
  • Ontologies and knowledge graphsWorkstream: concept links across catalogs
Reference implementations
Standards and guidance
  • World Bank metadata schemasJSON Schema for indicators, microdata (DDI), geospatial (ISO 19115), documents, tables, images, scripts
  • Global Question BankWorkstream: reusable questions and concepts across surveys
  • AI-ready metadata and standardsWorkstream: standards adoption guidance for AI-ready data

What it takes roles and effort summed from the linked recipes; systems and hosting written per dimension

Hosting shown for
StepWhoOne-off effortWork that recursSystems that have to exist
A to BFoundational recipesDeveloper or data engineer, Subject-matter staff, Data curator, Catalog or data manager, and 3 moreabout 5 person-daysPer item: about 30 minutes per indicator; an hour per questionnaire; half an hour per documentPer period: minutes after that
  • The Metadata Editor and a catalog (NADA or equivalent)
  • The World Bank metadata schemas and a completeness profile
  • Version control for records and profiles
B to CAI-ready recipesDeveloper or data engineer, Data curator, Catalog or data manager, Subject-matter staff, and 2 moreabout 4 person-daysPer item: half a day per source; half an hour per document; half an hour per tablePer period: minutes per period; part of each round; an hour per period to review
  • A controlled vocabulary (SKOS) and its mapping script
  • Published code lists and crosswalks
  • The Metadata Reviewer or an equivalent review step
C to DAI-native recipesDeveloper or data engineer5 to 10 person-days—
  • A concept scheme and question bank shared across surveys
  • Records regenerated from production at each release
Hardware and hosting, medium office
The catalog host and a small server for the review pipeline
Kinds of cost
  • Staff time of curators (the largest cost at every level)
  • No licence costs: the editor, the catalog, and the schemas are open source
Depends on
No other dimension; this is where the data side starts.

Effort is the sum of the time lines of the recipes the dimension links, written for a first implementation on the running examples; a large catalog multiplies the per-item work. Money is given as kinds of cost because prices differ too much between countries and years for a figure to hold. Pillar I has no such table: its investments are institutional decisions the program does not cost.

Relevant cookbook chapters

Progression between maturity levels

  1. A B

    Document each product in the World Bank schema for its type with the Metadata Editor or the CSV template, and run the completeness check at the foundational level.

  2. B C

    Publish code lists and crosswalks, record breaks and provenance as fields, and emit schema.org or DCAT from the catalog (NADA does this on every page).

  3. C D

    Reach the AI-ready profile across the catalog, link concepts to shared vocabularies, and use AI-assisted review under a curator to keep records consistent.

Levels A to D are the maturity scale of the framework. The actions listed are those supported by program resources; the framework's suggested actions for each transition also apply.

Evidence for the assessment form

  • Completeness report at the target level using the organization's profile
  • A validated schema.org or DCAT record
  • Published code lists and crosswalks
  • Provenance block in records

Recommended sequence

This companion orders the work for an early-stage organization as data quality control, metadata standards, a searchable catalog, and API access under a clear licence first. These correspond to dimensions 2.2, 2.1, 2.3, 2.4, and 2.6 on this page and to chapters 1 to 3 and 5 of the cookbook. Agentic and generative AI access (dimension 2.5) follows.

CookbookContact the program

Question numbering follows the reference guide draft v0.0.8 (June 2026). The mapping is maintained with the site; corrections are welcome as GitHub issues.