Picture a data team that answers a vendor survey and ticks “mostly ready” for its enterprise data. The following week, the support agent it shipped still cannot tell a refund from a chargeback.
Data readiness for AI is whether a dataset can be used for one specific AI task and has been shown to work for it: the fields mean what the task assumes, the exceptions the task must handle are present, and the result was checked on an agreed evaluation set in a way that can be repeated when the data or the task changes.
By Ho Bae, Founder & CEO, CUBIG Corp. · Updated November 2026.
Two surveys published in October put numbers on how organizations rate their own data. Dun & Bradstreet asked 10,000 businesses in 32 countries in its quarterly survey (released October 28, 2026): 8% said their enterprise data is fully ready to support AI at scale, 33% said mostly ready and 44% partially ready.
Further surveyed 515 U.S. AI decision-makers (released October 13, 2026). In that survey, 29% had completed a formal data audit and remediation plan, yet only 5% believed their data was largely AI-ready.
Key takeaways
- In both surveys, readiness is self-reported. Neither release defines what “ready” means, so the numbers measure how confident teams feel, not whether a task works.
- An audit can finish without the team believing its data is ready: in the Further survey, 29% had completed a formal data audit and remediation plan, while 5% believed their data was largely AI-ready.
- Readiness only becomes checkable once you name the task. Then it is a before-and-after result on an agreed evaluation set, not a label.
What do these surveys mean by AI-ready data?
Both surveys ask people to rate their own data. The Dun & Bradstreet release reports respondents choosing between fully, mostly and partially ready, and the Further release reports what respondents believe. Neither release states the criteria behind those labels, and neither says whether any task was run to check them.
Self-ratings still carry a signal. When fewer than one in ten organizations call their data fully ready, the ratings suggest a large gap between AI plans and the data underneath them. The rating cannot say which task the data would fail, or why, because the question was never asked about a specific task.
Why does a finished data audit not make data ready?
A typical data audit checks data against rules: completeness, formats, duplicates, ownership, retention. Those checks matter, and a remediation plan that follows them makes data cleaner and easier to govern. The Further numbers show the limit: completing an audit and believing the data is ready are different things, and the second happens far less often.
The release does not report whether those remediation plans had been carried out. One reason the two numbers can still differ is that readiness depends on the task. A customer table can be complete and still be unfit for a collections workflow if the field the workflow depends on means something else in one region, or if the hardship cases the team handles by hand are missing. Data quality for AI walks through the checks a rules-based audit does not cover.
| Question | Survey self-rating | Readiness for one AI task |
|---|---|---|
| Scope | All enterprise data | The fields, records and exceptions one task depends on |
| Evidence | How respondents feel | A result on an agreed evaluation set, before and after preparation |
| Unit | A label: fully, mostly, partially | Pass or fail against the success criteria the business set |
| When it changes | The next survey wave | When the data, the model or the task changes |
How would you know your data is ready for a task?
Start by naming one task and working through five checks in order. The checks turn a self-rating into something a second team could repeat.
- Name the task. Write down what the model or agent has to decide, for whom, and what counts as a correct answer.
- List what the task depends on. The fields, the relationships between records, the units, and the exceptions the business handles today.
- Diagnose the data against that list. Check usability, integrity, context, consistency, reproducibility and traceability for those fields, not for the whole warehouse.
- Run the task before and after preparation. Same model settings, same evaluation set, so the comparison shows whether preparation changed the result.
- Record the version and conditions. Keep the data version, the preparation applied and the evaluation conditions with the result, so the check can be repeated when the data changes.
What does the ROI gap in the survey tell you, and what does it not?
In the Dun & Bradstreet survey, 81% of businesses that rated their data fully ready reported broad or strong AI ROI, compared with 12% of those that rated it partially ready. The release does not define broad or strong ROI, and both sides of the comparison are self-reported.
The comparison shows that the two ratings move together. It does not show that preparing data caused the returns, since organizations that rate their data highly may differ in budget, skills or the tasks they chose. A team that wants to know whether its own data preparation pays off has to measure it on its own task, with the before-and-after comparison above.
How does CUBIG check readiness for a specific task?
Syntitan keeps the original Dataset as it is and starts each use case from a copy in its own Project, so readiness is checked per task. Ready for Raw Data diagnoses that copy along the six axes in the checklist above and records the prepared result as a project Version.
Ready for Model runs the task on the data before and after preparation under the same evaluation conditions, and shows the two results side by side. It also exports the files needed to rerun model training and scoring.
A readiness result does not replace the business’s own definition of success; it is the evidence behind that decision. Try it on your data.
