Why AI models aren't the issue - data structure is
In our third survey results review before the year’s end, we consider the results of a CData study, which finds that only 6% of participants believe their data infrastructure is AI ready and the key reason for this low figure is a disconnect between data infrastructure maturity and AI maturity. What we find most interesting is that AI models are no longer considered the constraint; it is AI data.
But first, some background.
Data infrastructure maturity is emerging as one of the foremost impediments to AI adoption and the report entitled The State of AI Data Connectivity: 2026 Outlook starts by addressing the results of the August 2025 MIT report, The Gen AI Divide: State of AI in Business 2025, which concluded that 95% of generative AI pilots at companies are failing. CData does not debate the accuracy of the 95% statistic but rather asks why a large number of companies are failing to realize meaningful ROI from their AI investments.
CData’s report is based on survey data from over 200 data and AI leaders at software providers and enterprise companies that embed AI copilots and agents into their products. And it concludes that ‘enterprise AI is no longer being limited by models. It’s constrained by data infrastructure and enterprise context.’
We should also add that the report’s insights are the combined results of two complementary surveys, one that gives the perspective of enterprise implementation leaders and the second from product leaders at software providers. And with that in mind, we will only focus on the views of enterprise implementation leaders and all the data we refer to is from the results of the survey of 100 enterprise data and AI leaders across industries.
Let’s dig in.
Who was surveyed?
CTOs and CDOs made up the bulk of participants
What size were the companies?
Percentage of company respondents categorized by company size.
How are organizations using AI technology?
A key element that comes into play for us when reviewing or summarizing a survey is its relevance to our client base. And this is no different with CData, where we’ve identified three highly-relevant areas that organizations are targeting with GenAI or agentic AI. The report highlights the use of AI in code generation (60%) and marketing/content creation (55%), stating that their adoption shows how the technology is embedded in technical and creative workflows. We would also add intelligent document processing to the list (see table below).
The AI tool sprawl
Organizations are having to use multiple tools to reach desired outcomes, what is referred as the AI tool sprawl. To do meaningful work, companies are having to use an entire raft of AI applications and platforms. For example 76% of respondents stating enterprise LLM platforms such as OpenAI and Claude) are the most important to their organizations use cases to 54% using business intelligence applications such as Tableau AI and Microsoft Copilot).
The AI tool sprawl
The tools needed to be deployed to result in meaningful outcomes
The crux of the matter
This leads to the state of data infrastructure where only a woeful 6% of enterprises said they were “very satisfied” with their integration strategy when asked the following question: how satisfied are you with your current data connectivity approach for AI initiatives, including ingestion of data from source systems, context injection for GenAI models and real data integration?
Satisfaction levels
How satisfied are respondents when it comes to their data connectivity approach.
This 6% statistic shows that a significant pain point for most organizations is their current integration strategy and infrastructure. And if we are to conclude, as the results the survey point to, that the issue lies with the data, companies must invest in building a centralized data access layer that delivers contextualized data from enterprise systems to AI models or agents.