There is a scene I have watched play out, in different rooms, in different industries, for over two decades. A leadership team announces an ambitious AI agenda. Budgets are discussed. Vendors are shortlisted. Then someone asks a quiet question. Which version of the customer master are we using?
And the room goes silent.
That silence is the real story of enterprise AI today. Not the demos or the keynotes, and certainly not the breathless predictions, but the pause when someone asks where the data actually lives, who owns it, and whether anyone in the building trusts it.
I have spent a corporate career of more than two decades inside that silence. Now, as a founder building a data intelligence platform, I watch the same pattern repeat across banks, telecom operators, manufacturers and government departments. The ambition is genuine. The foundation underneath it usually is not, and almost nobody wants to look too closely at the gap.
So let me say the uncomfortable part plainly. AI rarely fails in the model. It fails in the plumbing.
Consider what a large enterprise actually looks like underneath the strategy deck. An ERP customised so heavily that the original product is barely visible. A planning system bolted on a decade ago by people who have since left. Several CRMs inherited through acquisitions and never properly merged. Spreadsheets that were supposed to be temporary and have quietly become systems of record. And a data lake that was meant to unify all of it, which instead became one more silo, only larger and more expensive than the rest.
Now ask that landscape a simple question. What was our true cost last quarter? What is our real exposure to this counterparty? Which customer is this, across the five systems where she exists under slightly different spellings of her own name? In too many organisations, different teams return different numbers, and each will defend its version in the next review meeting with complete sincerity.
This is the environment into which companies are now pouring AI. And AI has one defining characteristic that everyone forgets in the excitement. It amplifies whatever you feed it. Feed it clean, governed, well-understood data and it amplifies insight. Feed it contradiction and it amplifies contradiction, confidently, at scale, in fluent and persuasive prose. A hallucinating chatbot is merely embarrassing. A chatbot trained on your own inconsistent data, answering your customers and advising your executives, is something closer to a liability.
There is a phrase I keep returning to for the layer where all of this goes wrong. The boring middle. The unglamorous space between raw data and shiny dashboards, where lineage and catalogues and quality rules and ownership and definitions are supposed to live. Nobody puts data governance on a conference banner. Yet this is precisely where AI programmes live or die.
Let me make that concrete with something my team actually found. While building column-level lineage for a large enterprise system, we traced a single identity field through the database and discovered it stored at three different field lengths across multiple tables. Any join across those tables risked silently dropping matches. No error. No warning. Just records quietly falling through the cracks, invisible until someone went looking. Now picture an AI model trained on that data, making decisions about real people. The model would not be wrong, exactly. It would be faithfully reproducing a broken foundation.
I have seen variations of this everywhere I have worked. The same KPI name carrying different formulas in different departments. The same metric drawn from different source tables on different refresh schedules. These disagreements look analytical, as though more analysis might settle them. They are actually definitional, and no algorithm settles a definition. Someone has to sit in a room and decide, once, what the number means and who owns it.
This is why I have grown wary of the phrase AI strategy. Most companies do not need an AI strategy. They need a data strategy that AI can sit on top of, and they need it before the budget is signed off rather than after the first project quietly stalls.
In my experience it comes down to four habits, none of them glamorous. Know what you have, through a living catalogue rather than a one-time audit that goes stale within a quarter, because you cannot govern what you cannot see. Trace where it flows, with column-level lineage from source to report, so that when a regulator or an auditor or your own model asks where a number came from, the answer takes minutes rather than a forensic investigation. Fix definitions before you fix technology, by agreeing what a customer is, what revenue means, what counts as an active asset. Those conversations are political and tedious and they are the highest-leverage work in the entire programme. And assign ownership, a real name against every critical data element, because a committee is not an owner, and data without an owner decays quietly and predictably.
There is a regulatory dimension threaded through all of this too. India’s DPDP Act, Europe’s GDPR and the emerging wave of AI-focused regulation worldwide rest on the same quiet assumption. That you know what data you hold, where it came from and who is allowed to touch it. The companies that built the boring middle will experience compliance as a documentation exercise. The ones that skipped it will experience it as a recurring crisis.
I understand the impatience. Boards watch competitors announce AI initiatives and feel the clock ticking loudly. But the organisations that win this decade will not be the ones that shipped a chatbot first. They will be the ones whose data was ready when the models arrived. Infrastructure compounds over time, and so do shortcuts, only in the wrong direction.
So before your next AI investment, ask the quiet question in the boardroom. Which version of the customer master are we using? If the room goes silent, you have just found your real AI project. It will never make a headline. It will make everything else possible.
To attributed to:- Rohit Kumar, COO ( Chief operational officer), SCIKIQ
Click Here to get the latest EV news and exclusive updates from EV Mechanica on WhatsApp!