AI-ready data: 87% of leaders think they have it, 43% are stuck on it
Nearly nine in ten business leaders say their data is ready for artificial intelligence. In the same breath, nearly half of them name data readiness as the number one obstacle to their AI projects. This contradiction, measured in late 2025 among 505 data leaders, says something precise about how companies approach AI today: confidence comes before the work. We suggest looking at what "AI-ready data" actually means, and where to start when you have a team of thirty rather than three thousand.
The confidence paradox: 87% feel ready, 43% are stuck
According to the "2026 State of Data Integrity and AI Readiness" study, published in January 2026 by Precisely and Drexel University's LeBow Center for Applied AI and Business Analytics, 88% of the leaders surveyed consider their data ready for AI and 87% judge their governance sufficient. A year earlier, only 12% gave the same answer. Yet in the same survey, 43% point to data readiness as the main barrier to aligning AI with business objectives, and 69% say they struggle to connect the performance of their AI projects to business outcomes.
The study covers large enterprises with more than a thousand employees. But the mechanism is identical in a Swiss or European SME, with one difference: large organisations have data teams to absorb the shock, while the SME discovers the gap the day the AI assistant gives a wrong answer because the customer file contains three spellings of the same name. In our view, the important point is not the percentage. It is that the perception of being "ready" forms before any real test. This is exactly what we see in the diagnostics run through our AI Barometer: the question "is your data reliable?" gets a quick yes, while the question "who checked, and when?" gets silence.
What "AI-ready data" means for an SME
The notion is often confused with owning a data warehouse or a dashboard. For an AI project in an SME, we use four criteria that are more modest and more useful.
Accessible: the information the AI needs lives in a system that can be queried (CRM, ERP, document management), not in mailboxes or personal Excel files. If the standard quote template lives on Mrs Perret's laptop, it does not exist for the machine.
Consistent: a customer, a product, a project carries the same identifier everywhere. A duplicate is not a cosmetic detail; it becomes a wrong answer the moment you ask an assistant "what is this customer's revenue?".
Dated and attributed: every record says who created it and when. Without this, the AI blends a 2022 price and a 2026 price with the same assurance.
Governed: someone is accountable for the quality of each dataset, with a verification ritual. The Precisely-Drexel study measures this without ambiguity: 71% of organisations with a governance programme have high trust in their data, against 50% without one.
These four criteria require neither a heavy budget nor new hires. They require a decision: treating data as an asset to be looked after, in the same way as cash flow or the order book.
The example of a 45-person industrial SME in the canton of Vaud
Take a precision engineering company on the shores of Lake Geneva, 45 employees, an ERP in place for eight years and a CRM added three years ago. Management wants an assistant that prepares answers to requests for quotation from the company's history. The pilot is launched on a generative AI tool connected to both systems.
First finding after two weeks: out of 1,200 customer records, 180 are duplicates, with different billing addresses. Out of 3,400 historical quotes, 40% have no final status recorded (won, lost, abandoned), so the assistant has no idea which offers actually convinced anyone. Finally, negotiated discounts are written as free text in the comment field, unreadable for a model without prior work.
The project was not abandoned, it was reordered. Three weeks went into deduplicating accounts, making a status mandatory on every quote and creating a structured "discount granted" field, with a named owner to maintain it. The pilot resumed on a clean base. Measured result after three months: the time needed to prepare a response to a tender fell from around four hours to an hour and a half, and above all, the error rate on quoted prices dropped sharply. The AI changed nothing about the trade. It made visible what the company had tolerated for years, then rewarded the correction.
Our four-week method to move from "we think" to "we know"
We rarely propose a six-month data programme to an SME. We propose a short cycle, tied to one precise use case, that produces a decision at the end.
Week 1, frame the use case and list the data it requires. No more than five to seven sources. A quoting assistant does not need the whole company; it needs customers, products, quotes and price lists.
Week 2, measure. For each source: duplicate rate, share of empty fields on critical attributes, average age, presence of an owner. A one-page table is enough. This is the moment when declared confidence meets the numbers.
Week 3, fix what blocks the use case, and only that. Deduplication, mandatory fields, entry rules in the CRM or ERP. Everything else goes into a dated backlog.
Week 4, name the owners and set the ritual: a thirty-minute monthly review of quality indicators, and one simple rule, no new data source without an owner.
At the end of the cycle, management has an honest verdict: the use case is ready, or it is not yet and everyone knows why. This ability to link a piece of data to a business result is what 69% of the organisations surveyed lack, and it is what separates a pilot that pays from a pilot that decorates.
What we take away
Confidence in your data is healthy, provided it is verified. The gap between 88% of leaders who say they are ready and 43% who stumble on data will not be closed by a more powerful model. It is closed with entry rules, named owners and a control ritual, in other words with governance, the least spectacular and most profitable lever of any AI project.
For an SME, the right question is therefore not "are we ready for AI?" but "for which use case, with which data, and who is accountable for it?". That is the question we ask first, before any choice of tool.
Want to see where your company stands on this scale? The AI Barometer gives you, in a few minutes, a sector-based reading of your data and AI maturity and one concrete next step. And if you would rather talk it through directly, our contact page is open.
Comments