Will your AI be built on sand?
I spend much of my time helping organisations understand why their AI projects struggle. The pattern is always the same. The model is rarely the issue. The real problems sit underneath, in the everyday systems, data flows and processes that AI depends on. When those foundations are weak, even the most impressive tools deliver fragile results.
Over time, I have learned that the smartest approach is not to start with AI. It is to start with honesty. You need a clear view of what you already have, what you rely on and what could quietly break your ambitions. AI succeeds when its foundations are stable, connected and trusted.
AI must operate within the realities of legacy platforms, shadow IT, clever workarounds and human behaviour. When those elements misalign, projects stall. Outputs lose credibility. Teams stop trusting the tools. Costs rise in places leaders did not expect.
Weak foundations have predictable consequences. Systems fail to communicate, so people compensate with manual work. Inconsistent data leads to conflicting reports that fuel doubt. Integration becomes expensive and slows momentum. Frustrated teams disengage from future initiatives. Clearing up these issues later costs far more than addressing them early.
The challenge is avoidable. A structured readiness assessment gives you clarity, reduces risk and protects your investment. It turns AI from an experiment into something that can scale.
Over the years, I have refined a simple framework: four tests that reveal how prepared an organisation truly is. Each test scores from 1 to 5. Your total score guides priorities before significant AI investment.
Every organisation believes it knows its tech stack. Few actually do. Tools proliferate quietly. Teams adopt new platforms to solve local problems. Unofficial tools become essential without leadership realising.
A strong foundation begins with visibility. You need an accurate, shared inventory of every tool, including shadow IT. You need to know who uses each system, what purpose it serves and why it was chosen. The exercise is practical, not political. The goal is clarity, not blame.
AI thrives when data flows cleanly. It struggles when systems require manual handoffs or operate in isolation. When I see staff carrying data between systems using CSVs or attachments, I know the foundations cannot yet support enterprise AI.
Mapping data flows end to end exposes the weak points. It shows where data originates, how it moves and where humans intervene. It highlights silent bottlenecks and disconnected platforms. Once visible, these issues become solvable.
AI depends on trustworthy data. If two departments report different values for the same metric, your data is not ready. When teams maintain their own spreadsheets because they do not trust the system of record, the problem is bigger than tools. It is a matter of confidence.
Testing data quality means running the same query across multiple systems and comparing results. It means naming known gaps, inconsistencies and estimates. Most importantly, it requires asking whether teams trust the data enough to act on it without manual checks. If the answer is no, AI adoption will stall.
Good governance gives AI safe boundaries. It requires clear ownership of systems and datasets. It sets expectations for access, change and quality. Without it, decisions slow, standards drift and security risks rise.
Strong governance is practical. Owners are accountable for decisions, not just usage. Change processes are documented and followed. Access standards are consistent. Timelines for approvals are visible and reasonable. When these elements are in place, AI can scale without fear of unintended consequences.
The four tests give you a score between 4 and 20. Low scores indicate foundational work that must come first. Mid-range scores suggest targeted improvements. High scores show you are ready for pilots while continuing to monitor your environment.
The scoring does not measure success. It measures readiness. It helps you decide where to focus your energy. It ensures that when you do invest in AI, the foundations can support it.
You do not need a budget to strengthen your foundations. Small steps make a meaningful difference.
Document your tech stack in a simple shared spreadsheet. Map one data flow from source to destination. Clarify ownership for your most important datasets. Consolidate duplicate tools where you can. Create a short data dictionary so teams speak the same language. Draft a RACI for a system that often causes confusion.
These improvements build trust and reduce friction. They create visible progress and prepare teams for larger changes.
The assessment works best with a cross-functional group. The aim is shared understanding, not evaluation. Walk through each test openly. Score honestly. Prioritise areas that fall behind. Start with the quick wins, then address systemic issues.
Rescore after you make progress. Seeing improvement reinforces momentum and strengthens confidence across the organisation.
AI only scales when the foundations beneath it are strong. Investing in tools before addressing the basics puts both budget and credibility at risk. With clarity, governance and trusted data, AI becomes stable, useful and ready to grow.
Start with what you already have. Fix the foundations. Then build your future with confidence.
To explore these ideas further, I recently discussed them during AI Week 2025. You can watch the full session here. https://aiexpert.digileaders.com/talks/will-your-ai-be-built-on-sand-how-to-test-your-tech-foundations/
November 2025
November 2025