According to a recent report1 only 7% of enterprises say their data is completely ready for AI, and more than a quarter describe their data as not very, or not at all ready.
For most organisations, the gap between ambition and readiness has become the primary factor holding back their AI strategies.
Unprepared data is often treated as an absolute stop sign; a problem that must be solved before an AI strategy can launch. Data readiness sits near the top of almost every list of barriers to AI adoption, and it’s easy to see the reasons behind this. Messy, siloed, incomplete data feels like something that must be fixed before building any capability on top of it.
But treating readiness as the entry ticket to AI is a strategic mistake, and increasingly an expensive one. Below, we lay out why the organisations pulling ahead are the ones who have understood that the data will never be “perfect”, and have stopped waiting for it to be.
The problem with waiting for “ready”
It’s important to recognise that data readiness isn’t a milestone that gets reached and then left behind. As soon as new systems get introduced, customer behaviour shifts, or markets move, what looked like clean, complete data six months ago suddenly isn’t either.
As Stelia’s VP of Applied AI, Paul Heathcote, put it at Cannes Lions this year, the organisations waiting until they have perfect data are waiting for something that is never going to arrive. The ones that lead will not be those with the cleanest datasets, but those that worked out how to build with the data they already hold.
That runs against a deep instinct in most data teams, but reflects how these systems actually behave. AI is far more tolerant of imperfect inputs than traditional, deterministic software, provided the system built around it is designed to make effective use of the information that is genuinely available.
The value in proprietary data
For many organisations, proprietary data is one of the most valuable competitive assets they hold, as outlined in our recent article. While foundation models are becoming increasingly capable, the insights, relationships and operational knowledge contained within an organisation’s own data remain difficult for competitors to replicate.
The organisations pulling ahead are those finding practical ways to put that data to work now – whether through fine-tuning, retrieval-based systems or other approaches – rather than waiting for it to meet an ever-shifting standard of perfection.
Doing so successfully, however, depends not only on the data itself, but on the architecture around it: the governance, security and infrastructure required to keep proprietary information controlled while making it available to AI systems effectively.
In that context, with the right technical foundations in place, imperfect proprietary data is often more strategically valuable than well curated public data, because it reflects knowledge that is uniquely their own.
Building a data flywheel
When used within the right system, proprietary data can become the starting point for a data flywheel – where each interaction can generate new information that improves the system over time. Predictions can be compared against real business outcomes, user feedback captured, and gaps or inconsistencies identified, creating the signals needed to improve both the model and the quality of the information available to the system over time.
This reframes the sequence many organisations assume they are locked into. Rather than treating data quality as a prerequisite for deploying AI, a well-built AI system itself helps expose opportunities to improve the data as it is used. Readiness becomes something that develops through continuous use and learning, rather than a fixed standard that must be reached before work can begin.
Seeing the flywheel in practice
Our recent collaboration with Monks is a direct example of how a system like this works.
The solution deploys fine-tuned prediction models, trained on brand-specific performance and historic campaign data, to pre-validate an ad creative’s projected algorithmic delivery before media spend is committed, improving return on ad spend in the process.
Crucially, it isn’t designed to produce a single prediction and stop there. As content delivery algorithms evolve and consumer behaviour shifts, the system continuously captures new performance data and compares predictions with real-world outcomes. That creates an ongoing feedback loop, allowing the models to be refined over time while steadily enriching the underlying knowledge base. Rather than waiting for perfect data before getting started, the system is designed to put existing data to work and improve through use.
The cost of waiting
Today, “we’re waiting until our data is ready” is a strategic liability. The organisations gaining an advantage are not necessarily those with the cleanest datasets, but those building the capability to put their existing data to work.
Doing so requires the right foundations: an architecture that keeps proprietary information secure, governed and within the organisation’s control while enabling AI systems to learn and improve over time. Without those foundations, imperfect data remains a constraint; with them, it becomes an asset that compounds in value.
Get that right, and the advantage builds on itself – each cycle improving on the last.
Because every quarter a competitor spends building is a quarter of compounding advantage that a team waiting cannot buy back. The organisations moving decisively now – using the data they already have and the architecture to make the most of it – are the ones that will still have a choice later.
