Most leadership teams believe their proprietary data will give them an AI advantage. That is comforting. It is also usually wrong. What most organisations call a data asset is a mixture of outdated records, conflicting definitions, duplicated files, undocumented processes and knowledge trapped inside people’s heads. It is less a data moat than a data swamp.
Before asking what AI can do with your data, ask what your data will do to your AI.
Garbage in, chaos out
“Garbage in, garbage out” was a warning built for simple systems. A traditional system fed bad data produces bad reports,the numbers look wrong, someone investigates, the error gets found and fixed. The system is forgiving because it’s stupid.
An AI system fed bad data produces confident nonsense. The chatbot answers with authority. The agent acts on stale context. The recommendation engine suggests products based on data mislabelled three years ago. The output looks right. It reads well. It’s also wrong, in ways that may take months to discover and years to undo. That’s not the chaos of an obvious error. That’s the chaos of a system that scales blindness.
The five data diseases
Most organisations suffer from some combination of five recurring problems.
- Fragmentation: Sales, finance, marketing and customer service all have different versions of the customer.
- Staleness: Records, policies, product information and process documents quietly decay while AI continues to treat them as current.
- Opacity: Nobody knows where all the data lives. It is scattered across spreadsheets, inboxes, shared drives and unofficial software platforms.
- Semantic confusion: The same word means different things in different systems. Ask how many “customers” you have and every department gives a different answer.
- Context loss: The data remains, but the meaning disappears. Nobody remembers why it was collected, what assumptions shaped it or whether it still reflects reality.
AI does not eliminate these diseases. It gives them speed, reach and authority.
Even clean data is incomplete
There is a deeper problem. Data is not reality. It is a partial record of reality. Revenue tells you what sold, not what almost sold. Churn data tells you who left, not why they left or what they told their friends. A customer satisfaction score tells you what respondents selected, not what the angriest customers refused to say.
Every dataset contains what someone decided to measure. Everything else remains outside it: hesitation, emotion, politics, culture, informal workarounds and unspoken needs. That is why future-fit organisations need both analytics and anthropology. Dashboards and conversations. Machine intelligence and people close enough to the work to recognise when the numbers do not tell the full story. AI can process what you captured. It cannot recover what you never noticed.
Do the audit
Before you deploy an agent, train a model, or announce an AI strategy to the board, ask the brutally honest questions. Can you produce a single, consistent list of your customers — agreed on by sales, marketing, finance, and service — in under an hour? If not, your data is fragmented. Does “customer” mean the same thing in every system? If not, no amount of AI will fix that.
Your data is always worse than you think. That’s not a criticism — it’s a universal condition. The question is whether you’re willing to look, and willing to treat data infrastructure with the same seriousness you treat financial infrastructure.
