Human Data Engineering: Preparing Enterprise Data for Artificial Intelligence

Most enterprises approach AI data the same way: collect data, give the data to AI, and expect AI to find the insights. This skips the human logic layer entirely.

Enterprise data does not behave as isolated tables — it behaves as a distributed knowledge network. The "Enterprise Data Octopus" describes this: a single organisation spans many arms (Customer, Finance, HR, Sales, Operations, Procurement, Manufacturing, Marketing, CRM, Support, Inventory, Compliance), and every arm carries an owner, a purpose, a creator, its users, dependencies, rules, a lifecycle, a quality level, and an authority — and every arm feeds the same brain. Although data is stored across multiple systems, departments and processes, every dataset remains connected to a common organisational purpose.

Current AI ignores this: it receives data and produces analysis without any of the organisational context that gives the data meaning. Gartner (2025) reports that through 2026, organisations will abandon 60% of AI projects that are not supported by AI-ready data — the failure is not the model but the data and its context (Gartner, 2025 — Lack of AI-Ready Data Puts AI Projects at Risk: gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk).

The gap is structural. Current AI thinks in a single step — Data → Analysis. The proposed research asks a foundational question: should AI learn the organisational context of data before analysing its contents? And if yes, how should it be done?

A second question concerns inputs. When AI is fed data or information, what inputs should it efficiently process or analyse — raw data, or context, purpose, ownership, business logic, data lineage, definitions, and decision objectives? Enterprises feed raw data to AI and treat confident output as reliable, but AI operating without context cannot truly interpret the data it processes. The problem is therefore not how to analyse data faster, but how to prepare data — and its context — before AI is introduced.

AI Logic Engineering holds that human logic must precede AI analysis. Human Data Engineering is the part of the Enterprise AI Framework that prepares data before AI touches it: data must be understood, organised logically, structured, and governed, and AI must be taught what the data represents before it analyses it.

Within this, Data Context Engineering is the specific practice of giving AI the context of a dataset before its contents. Before AI receives sales data, it should know: why the report exists, who owns it, how often it is updated, what decisions depend upon it, its known weaknesses, its missing fields, its confidence level, the business definitions behind it, its stakeholders, and the expected output.

The discipline also holds that the right tool should be used for the right stage. Existing systems (ETL, SQL, Power BI, Snowflake, SAP, Python, Tableau) already perform aggregation, transformation and statistical summarisation extremely efficiently. AI should be used as a tool to generate analysis, ideas and perspective from the pre-refined and analysed data — not to re-perform the work those tools already do better.

The research is structured around the Enterprise AI Framework as the governing architecture, comprising eight components: (1) Enterprise AI Governance, (2) Human Data Engineering, (3) AI Learning Standardisation, (4) Knowledge Management, (5) Verification, (6) Continuous Improvement, (7) Enterprise KPI Framework, and (8) AI Maturity Assessment.

Within this, the research part of Enterprise AI proceeds as: Enterprise AI Framework → Data Governance → Human Data Engineering → AI Data Interpretation → AI Learning Standard → Continuous Improvement.

The operational research proposal is: Business → Human understands the data → Data is organised logically → Data is structured → Data is governed → AI is taught what the data represents → AI analyses → Human verifies.

Data Context Engineering is investigated through a seven-stage process: • Stage 1 — Understand the data (What is this dataset?) • Stage 2 — Understand the business (Why does this dataset exist?) • Stage 3 — Understand the decision (What decision is expected from this analysis?) • Stage 4 — Identify missing information (What information is unavailable?) • Stage 5 — Evaluate data quality (Can reliable conclusions be drawn?) • Stage 6 — Recommend additional data (What else would improve confidence?) • Stage 7 — Perform analysis (Part of AI Logic Engineering is Data Context Engineering)

The proposed data flow for analysis efficiency and resource utilisation is: ERP → Warehouse → Cleaning → Transformation → Aggregation → Human review → AI reasoning → Decision.

The research contrasts the current model (Data → Analysis) with the proposed model (Context → Purpose → Structure → Data → Analysis).

Status: concept published; research in progress. This document is a proposed research outline, not the completed research.

A standardised method for preparing enterprise data for Artificial Intelligence — shifting enterprises from feeding raw data to AI toward context-engineered analysis. The contribution is a defined, repeatable Human Data Engineering practice (including Data Context Engineering and the seven-stage process) that improves the reliability of AI-assisted enterprise decisions and directly addresses the AI-ready-data failure point identified by Gartner.

Enterprise AI Framework (the Human Data Engineering and Data Context Engineering components), with supporting integration into the AI Quality Assurance Framework and the AI Verification Framework. Findings are intended to inform future editions of the Enterprise AI Framework.