This is the central tension in almost every digital program I have seen. The vision is compelling, and the data reality is brutal. Here is an honest assessment of what that means in practice.
The problem is real and underappreciated.
Most organisations discover too late that their data is fragmented across legacy systems, inconsistently defined, poorly governed, and held in formats that were never designed to work together. For a Digital Twin, you need asset registers, maintenance histories, sensor feeds, spatial data, and operational records to line up. In government and infrastructure, that alignment rarely exists at the start.
The instinct is to treat this as a precondition. Fix the data first, then build the capability. That approach has caused more program failures than any technology ever has.
Where current AI genuinely helps
Things have shifted meaningfully in the last couple of years. There are four areas where AI is already doing useful work on this problem, not just promising to do so.
Schema matching and data integration. Large language models are surprisingly good at inferring what data means from field names, sample values, and context, then suggesting how to map between inconsistent systems. Work that used to take a data architect weeks can now be drafted in hours and checked by a person. It is not perfect, but it sharply lowers the cost of integration.
Data quality detection. AI can scan large, messy datasets and surface anomalies, gaps, duplicates, and inconsistencies far faster than rule-based tools. More usefully, it can prioritise which quality issues matter for a given use case, rather than producing a remediation list nobody acts on.
Synthetic data. When real data is sparse, sensitive, or missing, AI can generate plausible synthetic datasets that preserve the statistical shape of the real thing. For planning and simulation, this is increasingly viable. You can train and test before the full pipeline is in place.
Natural language access to fragmented data. Rather than waiting for full integration, you can put an AI layer in front of several disconnected systems and have it reason across them to answer questions. It is not the same as proper integration, but it brings value forward and strengthens the case for investing in the underlying foundation.
What AI cannot yet fix
There are limits worth being clear about. AI does not solve the governance problem. Who owns the data, who maintains it, what the definitions mean, and who is accountable when it is wrong are still people-and-process questions. AI also cannot manufacture data that was never collected. Sensor gaps, broken asset registers, and missing historical records are not things a model can reliably infer for high-stakes decisions.
There is also a real risk. The ability of AI to paper over poor data can tempt organisations to defer the hard work of fixing the foundation. Getting plausible-looking answers from poor inputs is more dangerous than getting no answer at all, because people act on it.
The honest answer
Current capability genuinely helps, enough to unblock a program that would previously have stalled. The economics of integration have changed. The barrier to showing value from imperfect data has dropped. You can now build credible proofs of concept that create the appetite to fix the data properly, rather than needing perfect data before you start.
But it is not a solved problem. The shift is from “we cannot proceed until the data is fixed” to “we can proceed in parallel while fixing the data.” That is a meaningful change in how you sequence a program and structure a business case. It is not a bypass.
For government clients, the practical implication is simple. Data and AI strategies need to be written together, not one after the other. The AI ambition justifies the data investment, and the AI deployment is shaped by what the data can support today against what it could support in two or three years.
GC3 Digital helps government and infrastructure clients turn data and AI ambition into deliverable programs. Get in touch.
