Data Readiness for AI

Intro & scope

Use cases that make an organisation's data, documents and knowledge ready for AI before any model is deployed. Patterns: PREP.INVENTORY (know what data exists, who owns it and how to reach it: PREP-001..005), PREP.DIGITIZE (get paper and legacy formats into machine-readable form: PREP-006..008), PREP.STRUCTURE (organise files, metadata, tables and definitions: PREP-009..012), PREP.KNOWLEDGE (curate content, terminology and examples for assistants: PREP-013..017), PREP.LABEL (build evaluation sets, categories, labels and feedback loops: PREP-018..021) and PREP.PROTECT (pseudonymisation, permissions, classification and legal basis: PREP-022..025).

Scope boundaries: - PREP-001 (organisation-wide readiness and inventory) vs PREP-002 (one concrete use case) vs PREP-003 (shadow spreadsheets) vs PREP-005 (system access). - PREP-006 (scanning and OCR of paper) vs PREP-007 (digital legacy formats, mail archives) vs PREP-008 (avoid paper by capturing digitally at source); the parsing step itself is DOC-014. - PREP-009 (folder and site structure) vs PREP-010 (metadata) vs PREP-011 (spreadsheets) vs PREP-012 (definitions and semantic layer). - PREP-013 (curated RAG corpus) vs PREP-014 (rewriting for retrieval) vs PREP-015 (terminology) vs PREP-016 (gold examples) vs PREP-017 (lifecycle). - PREP-018 (evaluation sets) vs PREP-019 (category schemes) vs PREP-020 (labels from records) vs PREP-021 (user feedback). - PREP-022 (pseudonymisation of data) vs PREP-023 (permissions) vs PREP-024 (classification rules per AI tool) vs PREP-025 (purpose and legal basis). - All PREP cases are minimal AI Act risk; most are organisational work with light tooling, so the hardware tiers are low.

Overview

ID	Title	Pattern	Maturity	Path	HW tier	AI Act risk
PREP-001	AI data-readiness assessment & data inventory	PREP.INVENTORY	early-adopter	C	HW1	minimal
PREP-002	Use-case data mapping & gap analysis	PREP.INVENTORY	early-adopter	B	HW1	minimal
PREP-003	Discovery of Excel shadow systems & personal data stores	PREP.INVENTORY	early-adopter	C	HW1	minimal
PREP-004	Data ownership & stewardship roles for AI-relevant data	PREP.INVENTORY	mainstream	A	HW0	minimal
PREP-005	Getting data out of ERP, SaaS & legacy systems (exports, APIs, connectors)	PREP.INVENTORY	mainstream	C	HW1	minimal
PREP-006	Back-file scanning & OCR of paper archives for AI use	PREP.DIGITIZE	mainstream	C	HW1	minimal
PREP-007	Conversion of legacy file formats & mail archives to AI-readable formats	PREP.DIGITIZE	mainstream	C	HW1	minimal
PREP-008	Replacing paper checklists & logbooks with digital capture at source	PREP.DIGITIZE	commodity	A	HW1	minimal
PREP-009	File-share & SharePoint restructuring for AI indexing	PREP.STRUCTURE	mainstream	C	HW1	minimal
PREP-010	Document metadata & tagging for retrieval	PREP.STRUCTURE	early-adopter	C	HW2	minimal
PREP-011	Restructuring spreadsheet data into tidy, machine-readable tables	PREP.STRUCTURE	mainstream	C	HW1	minimal
PREP-012	Business glossary, KPI definitions & semantic layer for AI	PREP.STRUCTURE	early-adopter	B	HW1	minimal
PREP-013	Curating an authoritative knowledge base for RAG assistants	PREP.KNOWLEDGE	early-adopter	C	HW1	minimal
PREP-014	AI-ready authoring: rewriting documents for retrieval	PREP.KNOWLEDGE	early-adopter	A	HW0	minimal
PREP-015	Company terminology, abbreviation & synonym lists for AI	PREP.KNOWLEDGE	mainstream	A	HW1	minimal
PREP-016	Collecting gold-standard examples & templates for AI	PREP.KNOWLEDGE	mainstream	A	HW1	minimal
PREP-017	Knowledge content lifecycle: review cycles & feedback from assistant errors	PREP.KNOWLEDGE	early-adopter	C	HW1	minimal
PREP-018	Building test & evaluation sets with domain experts	PREP.LABEL	mainstream	C	HW1	minimal
PREP-019	Defining category schemes before AI classification	PREP.LABEL	mainstream	C	HW1	minimal
PREP-020	Harvesting labels from existing business records	PREP.LABEL	mainstream	C	HW1	minimal
PREP-021	Capturing user feedback & corrections on AI output	PREP.LABEL	early-adopter	C	HW1	minimal
PREP-022	Pseudonymisation of datasets & exports before AI use	PREP.PROTECT	mainstream	C	HW1	minimal
PREP-023	Permission & oversharing clean-up before Copilot / AI assistant rollout	PREP.PROTECT	mainstream	A	HW0	minimal
PREP-024	Data classification rules for AI use ('what may go into which AI tool')	PREP.PROTECT	mainstream	A	HW1	minimal
PREP-025	GDPR purpose & legal-basis check for reusing data in AI	PREP.PROTECT	early-adopter	A	HW2	minimal

Shared stack

Common layers for all PREP use cases: one source and ownership register (spreadsheet, Notion or a data catalog such as DataHub / OpenMetadata) fed by PREP-001..004, a document-parsing and OCR service (Docling, Tesseract or PaddleOCR) for PREP-006/007, a metadata and glossary store shared with PREP-010/012/015, a PII detection and pseudonymisation layer (Presidio) for PREP-022, and a lightweight evaluation and feedback log for PREP-018/021 that later feeds the DOC, KNOW and ANOM families.

HW0 (no own hardware, SaaS/office tools): Microsoft 365 / Google Workspace features, SharePoint permission reports, shared spreadsheets. Covers PREP-004, 014, 023.
HW1 (CPU/laptop): Tesseract/PaddleOCR, Docling, pandas/DuckDB, Presidio, a 3-8B quantised model via llama.cpp/Ollama for tagging and drafts, n8n. Covers most cases (PREP-001..003, 005..009, 011..013, 015..022, 024).
HW2 (single 24-48 GB GPU): 7-32B instruct model for bulk metadata tagging and consent/purpose triage at scale (PREP-010, 025).
HW3: not needed for this family.
Path guidance: A = managed SaaS or built-in office features first; B = self-host/build when data sensitivity or volume requires it; C = service-led or in-house project work with a partner (scanning bureau, integrator) and DPIA where personal data is involved. Run PREP-001/002 first, PREP-024/025 before any data leaves the organisation, and reuse one register, one parsing service and one glossary across all follow-up use cases.