Most discussions about artificial intelligence focus on tools:
Which model should we use?
Should we build an AI chatbot?
Can we automate this workflow?
But many organisations run into trouble long before those questions matter. The real issue is that AI projects often begin without proper foundations.
The numbers are hard to ignore. MIT research published in 2025 found that 95% of generative AI pilots in enterprise settings produce zero measurable return — not low return, but zero. A separate S&P Global Market Intelligence survey of more than 1,000 organisations in North America and Europe found that 42% of companies abandoned most of their AI initiatives in 2025, up sharply from 17% the previous year. The average organisation scrapped 46% of AI proof-of-concepts before they ever reached production.
These are not technology failures. They are foundation failures.
Before an organisation thinks about prompts, automation, or AI agents, three fundamental questions must be answered:
Do we have the right data?
Do we have a clear strategic objective?
Will this actually create value?
I call these the Foundation Three. They represent the conditions that must exist before AI adoption even begins. When one of these foundations is weak, AI initiatives tend to stall, produce disappointing results, or quietly disappear after an initial pilot.
The Foundation Three
The three foundations are:
Data (Corpus)
Strategy (Course)
ROI (Value)
Each addresses a different dimension of readiness. Together they answer the most important early question: Should we even attempt AI here?
This framing is not unique to this framework. McKinsey’s 2025 State of AI survey identifies six dimensions essential to capturing value from AI: strategy, talent, operating model, technology, data, and adoption and scaling. AI Singapore’s AI Readiness Index (AIRI) independently maps organisational preparedness across five pillars, with Data Readiness and Business Value Readiness sitting at its core. The Foundation Three represents the minimum viable version of that readiness conversation — stripped back for organisations that are not yet ready for a full maturity assessment.
1. Data (Corpus)
Artificial intelligence systems operate on information. Without structured, accessible, and relevant data, AI cannot produce reliable outcomes. This is the most commonly overlooked foundation.
Many organisations assume AI will magically solve problems without realising that the system must first understand the environment it is working in.
What “Data” Means in Practice
Data does not always mean massive datasets or machine learning pipelines. In many small and medium-sized organisations it simply means:
documents
spreadsheets
emails
knowledge stored in staff experience
internal procedures
customer interactions
The critical question is whether this information exists in a form that AI can access, interpret, and process.
The Scale of the Problem
Research consistently points to data as the primary failure point. Gartner found that 63% of organisations lack confidence in their data management practices for AI, and a separate 2025 Gartner survey found that 57% of organisations estimate their data is simply not AI-ready. Gartner has predicted that 60% of AI projects lacking AI-ready data will be abandoned by 2026 — and given that 42% have already been scrapped, that trajectory is already well underway.
Gartner defines AI-ready data as data that is aligned to specific use cases, actively governed, supported by automated pipelines with quality gates, and continuously quality-assured. Most organisations are missing several of those conditions before they even start.
Typical Data Problems
Common issues include:
information scattered across different systems
knowledge locked in individual employees’ heads
inconsistent document formats
outdated or incomplete records
no central knowledge repository
As Lingaro’s AI Adoption Playbook notes, “AI supports daily work only when the underlying data is clean, timely, and consistent. If data is unreliable, slow to update, or spread across different systems, people will not trust the results and the technology will not scale”. In these situations AI tools struggle because they cannot find a reliable source of truth.
The Data Readiness Question
A simple diagnostic question is:
If we wanted an AI assistant to help us tomorrow, where would it find the information it needs?
If the answer is unclear, the organisation’s data foundation is weak.
2. Strategy (Course)
Even when data exists, AI projects often fail because there is no clear strategic purpose. Organisations sometimes adopt AI simply because the technology is fashionable or competitors are experimenting with it. This leads to experimentation without direction.
The Strategy Problem
Research from UQ Business School, drawn from the book Why Data Science Projects Fail, identified “choosing to fix a problem that’s not aligned with the business strategy” as one of the two biggest causes of AI project failure. MIT researchers point to a related pattern: in the rush to show quick wins, teams skip the hard work of building proper foundations, accumulating technical debt and fragmented capabilities that cannot scale.
Without strategy, organisations often ask the wrong questions:
“What AI tools should we buy?”
“Where can we apply AI?”
But the better question is: What business problem are we trying to solve?
AI should always be tied to a specific operational objective. Examples might include:
reducing time spent on administrative work
improving customer response times
extracting insights from internal documents
translating or summarising information quickly
When the objective is unclear, AI initiatives become disconnected experiments.
Strategy Creates Boundaries
A clear strategy defines:
where AI should be used
where AI should not be used
which outcomes matter
This prevents organisations from chasing every new tool that appears. A useful framework for grounding this is SMART objectives — ensuring AI goals are Specific, Measurable, Achievable, Relevant, and Time-bound. This converts vague intention into a testable project scope.
The Strategy Readiness Question
A useful question is:
If this AI project succeeds, what measurable problem will it solve?
If that question cannot be answered clearly, the strategic foundation is not yet in place.
3. ROI (Value)
The third foundation is economic reality. Even if data exists and strategy is clear, AI initiatives must produce real value.
That value does not always mean direct revenue. It can also include:
time savings
reduced operational errors
faster decision-making
improved customer experience
However, the value must be tangible and observable.
The ROI Illusion
Many organisations invest in AI because it appears innovative or impressive. But innovation alone does not justify a project. McKinsey’s research found that while GenAI has produced productivity gains of 10–15% for some clients, less than 10% of enterprises have deployed it at scale, and less than 5% of scaled use cases are integrated across entire workflows. McKinsey now describes the coming decade as primarily about “value realisation” — converting AI investment into demonstrable business outcomes, not just pilots.
Without clear value, AI initiatives often become:
pilot projects that never scale
internal experiments that lose momentum
technology demonstrations without practical benefit
Gartner’s move of Generative AI into the Trough of Disillusionment in its 2025 Hype Cycle reflects exactly this pattern — initial enthusiasm colliding with the gap between expectation and practical return.
Where Value Usually Appears
In practice, the strongest ROI opportunities tend to occur in areas where organisations experience:
repetitive information processing
document-heavy workflows
large volumes of customer communication
time-consuming administrative tasks
These are environments where AI can reduce friction. The principle is simple: the more structured and repetitive the task, the more predictable the return.
The ROI Readiness Question
A helpful question is:
If this AI solution works, how will our daily work actually change?
If the answer is vague, the expected value may not yet be clear.
How the Foundation Three Work Together
The three foundations interact closely. AI initiatives succeed when all three are present.
Data → provides the information
Strategy → defines the objective
ROI → justifies the investment
This is why many AI projects fail quietly after the first stage. They were launched before the foundations were stable. MIT’s research identified this as a “learning gap” — not a failure of the AI model itself, but a failure of enterprise integration and organisational readiness. The technology works. The foundation does not.
Why the Foundations Matter for AI Adoption
The Foundation Three represent the starting point for any serious discussion about AI adoption. Only after these foundations are addressed does it make sense to explore:
how employees will interact with AI systems
how organisations will govern their use
how AI initiatives will gain traction internally
These later stages involve human factors such as training, culture, and organisational support. But none of those issues can be solved if the initial foundations are weak.
This mirrors what AI Singapore’s AIRI framework calls Organisational Readiness — which includes AI literacy, management support, and employee acceptance. These are the second-order conditions. The Foundation Three are the first-order conditions that must come before them.
A Simple Way to Think About It
When organisations begin exploring AI, they often jump directly to tools. But a better approach is to ask three grounding questions:
What information will the AI use?
What problem are we solving?
Why will this matter to the business?
If those questions have clear answers, the organisation has established its AI foundations. If not, the priority should be strengthening those foundations before any AI system is deployed.
The Role of the Foundation Three
The purpose of the Foundation Three is not to limit experimentation. Instead, it helps organisations focus their energy where AI can actually work.
The statistics are clear: organisations that skip foundations do not just slow down — they fail at rates that would be unacceptable in any other area of technology investment. By examining data, strategy, and value first, businesses can move beyond hype and begin asking a more practical question:
Where can AI genuinely help us work better?
Only after that question is answered does it make sense to move into the next stage of AI adoption.
Key sources drawn on: MIT Project NANDA (2025), S&P Global Market Intelligence (2025), Gartner AI Data Readiness (2024–2025), McKinsey State of AI (2025), UQ Business School / Dr Evan Shellshear (2025), AI Singapore AIRI Framework, Lingaro AI Adoption Playbook (2026), Thomson Reuters SMART AI Strategy Framework (2026).