Gartner: 60% of AI Projects Without AI-Ready Data Will Be Abandoned by 2026
Gartner predicts 60% of AI projects lacking AI-ready data will be abandoned through 2026. Inde argues most firms invest in ERP, cloud and AI without understanding the data estates beneath them.

Why it matters
- Gartner predicts that through 2026, organisations will abandon 60% of AI projects unsupported by AI-ready data.
- A Gartner survey of 1,203 data management leaders found 63% of organisations lack or are unsure they have the right data management practices for AI.
- Gartner expects more than 70% of recently implemented ERP initiatives to fall short of their original use-case goals, as reported by TechTarget.
- Inde's analysis was published on 17 November 2026 and ties investment readiness to understanding undocumented dependencies and workarounds.
Gartner predicts that through 2026, organisations will abandon 60% of AI projects unsupported by AI-ready data. The figure comes from a survey of 1,203 data management leaders, and it frames the argument made by data consultancy Inde in an analysis published on 17 November 2026: most enterprises do not understand their own data estates well enough to justify the technology investments they are about to make.
The stakes are concrete. The same Gartner survey found that 63% of organisations either lack, or are unsure whether they have, the right data management practices for AI. As reported by TechTarget, Gartner also expects more than 70% of recently implemented ERP initiatives (enterprise resource planning systems that run core business functions) to fall short of their original use-case goals. The pattern, Inde argues, predates the current AI wave and shares one root cause: significant investment decisions made on an incomplete picture of the current environment.
Why does this matter before an ERP, cloud migration, or AI programme?
Inde's author, drawing on a career building business intelligence reports, opens with an observation about organisational confidence. Leadership teams can usually name the technology their organisation runs. Ask what would break if one of those systems changed tomorrow, and the answer is usually far less certain. That uncertainty rarely matters day to day. It matters when a company commits significant investment to a new ERP, a cloud migration, a modern data platform, or an AI programme.
The author writes that when organisations ask whether they are ready for a major technology investment, the first question back is whether they understand their data estate well enough: not which systems they have, but how data move between them, who relies on them, and what happens downstream when something changes. "Every major technology decision is built on the current estate, whether or not anyone understands it," the analysis states.
Gartner's research supports that framing. In many organisations, the survey notes, how data are used is poorly documented and spread across a combination of siloed systems and individual, institutional knowledge. Practitioners cited by Inde point to the same causes for ERP disappointment: an incomplete understanding of current-state processes, and underestimated integration and data work.
What does 'understanding your data estate' actually mean?
Most organisations hold a reasonable inventory of applications, platforms, and integrations. Inde calls that inventory necessary but insufficient. It is "a map of the buildings, not of the traffic between them, and it's in that traffic that hidden risk exists."
The article lists what that hidden traffic typically contains:
- Integrations known by their endpoints, not by the hard-coded business rules that transform data in transit.
- Finance running its forecast from a spreadsheet maintained on the side.
- An Access database built 20 years ago still performing a critical function.
- Institutional knowledge sitting in a few people's heads — a risk Inde says is particularly pronounced in New Zealand, where many businesses depend heavily on individuals.
The author's test is blunt: if this person left, what would fall over?
None of this accumulates by design. People find ways to do their jobs, and Inde's author offers a personal measure of the phenomenon: "I'd be a very rich man if I had a dollar for every time someone asked, 'Can I get that in Excel?'" Each workaround adds an undocumented dependency, and over time those workarounds introduce cumulative risk — spreadsheet formula errors that flow downstream for years, personal information circulating where it should not, and business logic embedded in processes few people fully understand.
The symptoms are familiar to most enterprises: figures that change depending on who you ask, business terms that mean different things across systems, and projects that repeatedly uncover dependencies nobody knew existed. Inde notes that none of these is necessarily a problem in isolation. Different parts of a business can legitimately define or use the same data differently. The risk arises when nobody knows that they do, or can explain why.
The goal is not a single source of truth for every number. A healthy data estate needs to support a shared understanding of what the numbers mean, how they are derived, where they are used, and why legitimate differences exist. When that understanding is missing, there is usually undocumented logic somewhere in the estate.
How does poor data readiness break AI specifically?
Inde's argument is that AI amplifies the quality of whatever it is built on. Models are only as good as the data beneath them and the shared business definitions that give those data meaning.
The article gives a specific example. Even within a single widely used ERP platform, the same concept can appear as product code, item ID, item number, or product number, because different teams built different modules. A person familiar with the organisation or platform may know these mean the same thing. A model cannot be expected to know that reliably. Without clear definitions and context, it can interpret the same data differently and produce inconsistent answers.
The fix, Inde writes, is unglamorous but effective: agree business definitions and logic, document them, and maintain them in an authoritative reference point, such as a data governance platform or metadata layer that both people and AI can draw on. That is a modest claim, and it carries an implicit caveat — governance work delivers no visible capability on its own, which is precisely why organisations skip it.
What is the business case for doing this work?
Executive teams care most about two things, according to Inde: risk mitigation and return on investment. The case for understanding the data estate is usually made on the first. The analysis argues that this undersells the second.
Too many technology investments start with assumptions — that a particular platform will solve their problems, that the data needed already exist, or simply that the organisation needs to move quickly to keep up in the market. Understanding the data estate tests those assumptions. It shows what is actually available, where dependencies lie, what the data can readily support, and what needs to be addressed first.
The return also shows up in a different place. An assessment of the data estate may not deliver a return on its own. Its value is realised through every investment that follows, because each one is scoped against what is actually there rather than what is assumed. For AI in particular, Inde argues, the value compounds: once questions of ownership, lineage, and governance are answered, each new use case is faster and cheaper to deliver.
The longer-term shift the article describes is from a reporting culture to an asking-questions culture, which works only when information is timely, accurate, and accessible — when anyone in the business can ask a question and get an answer they trust, without commissioning a report or stitching together twenty spreadsheets. Treated this way, understanding the data estate is a strategic asset, not a pre-project chore.
What does good readiness look like in practice?
Inde is explicit that a full map of the estate is not a prerequisite. Good can start with the investment already on the table. An AI use case needs a deep understanding of the specific data it draws on: where the data come from, who owns them, how they are maintained, and what else relies on them. Discovery can be scoped around that initiative and extended with each one that follows.
Readiness is also not one-size-fits-all, even within AI. Sometimes a second agent checking the first one's work is worth it; sometimes, where the cost of a wrong answer is higher, a person in the loop is the better choice. Those decisions depend on understanding the use case, the risk, and the data underneath it.
Done this way, Inde claims, understanding the data estate delivers four things:
- Surfaces risk before it becomes cost. Dependencies found in discovery are a planning input; found during delivery, they become overruns and rework.
- Improves sequencing. It may show that modernising a few applications, retiring shadow spreadsheets, and putting governance in place first will make the major investment more predictable and its return more reliable.
- Recognises that a platform is not a process. A new platform will not fix weak governance, data management, or validation. Inde's verdict on skipping those foundations is blunt: "it's lipstick on a pig: impressive capability on a foundation that is wobbly at best."
- Informs architecture. Whether an organisation needs a central data platform, a semantic layer, better integration, or simply clearer ownership should depend on what the estate requires, not on what the market is promoting.
On method, the article recommends a data audit as a first step to establish what exists and highlight technical debt. Genuine understanding, Inde argues, comes from tracing how data move and are used, talking to the people who work with them every day — not only the technology team — and connecting that picture to the decision in front of leadership.
How should leadership act on this?
Inde closes with a distinction and a question. Knowing your systems is not the same as understanding your data, and neither is the same as understanding how those data are used. That gap, the article states, is where budgets blow out, pilots stall, and returns erode.
The way to close it is to tie the effort to the investment leadership already cares about. Before committing significant money to the next AI tool, cloud initiative, or ERP, Inde advises one question: how confident are we that we understand what this investment depends on, and what will depend on it? In the author's experience, that question can open a lot of eyes, and the answer is rarely as straightforward as people expect.
It is not a one-off exercise. Knowledge kept current informs every investment decision that follows — a point worth weighing against Gartner's 60% abandonment prediction, since the analyst's forecast implies most of that avoidable loss will not be avoided.
Source: itbrief.com.au
More from James Calloway
Show full bio
News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.
722 articles
Related articles
- SAS: Corporate AI's 2027 Test Will Be Trust, Not Deployment Speed
- 51% of AI projects miss targets, yet 85% of firms raise budgets
- Why 80 Percent of AI Projects Fail Before the Model Is Ever Wrong
- Only 8% of Australian Firms Have Data Ready for AI at Scale
- Over 40% of Agentic AI Projects Face Cancellation by 2027