Why 80 Percent of AI Projects Fail Before the Model Is Ever Wrong

More than 80 per cent of AI projects fail, roughly twice the failure rate of comparable IT projects, according to RAND Corporation estimates cited in a new analysis of enterprise AI spending. The piece, published on 10 November 2026, argues that the technology is rarely the culprit — the budget sequence is.

The supporting numbers are blunt. A 2025 S&P Global Market Intelligence survey of more than 1,000 organisations found that 42 per cent had abandoned most of their AI initiatives, up from 17 per cent the year before. The average organisation scrapped 46 per cent of its AI proofs of concept before they ever reached production.

The context matters: boards want AI, vendors demo it, and leadership teams can usually name use cases they want to pursue. What's missing, the analysis argues, is AI that runs in production daily and moves a number the business cares about.

Why do AI projects actually fail?

RAND interviewed 65 experienced data scientists and engineers to answer that question. Their answers fall into five patterns — and only one concerns the limits of AI itself.

• The wrong problem. The most common cause is a misunderstanding of what the project should achieve. Data teams optimise the metric they were given, not the outcome the business actually needed.
• Not enough of the right data. A sound idea still fails if the data to train a reliable model does not exist or is not good enough.
• Technology first. Projects chase the newest technique rather than the simplest thing that solves the user's problem.
• No path to production. Without infrastructure to manage data and deploy models, a working prototype has nowhere to go.
• Problems AI cannot solve. Some projects ask the technology to do something it simply cannot do yet.

To that list, the analysis adds two patterns seen repeatedly in Australian organisations.

The first is spending in the wrong order. Teams build heavy data pipelines and integrations before they know the idea works, or keep tuning a model for accuracy long after returns have flattened. Either way, the budget is gone before the value is proven.

The second is underestimating production. A model that performs well in a pilot still has to be fed live data, monitored for drift, connected to systems that act on its outputs, and adopted by the people whose work it changes. That is where most of the effort sits — and it is the part most business cases leave out.

The takeaway: AI projects rarely fail because the model was wrong. They fail because of decisions made long before anyone trained one.

What does the four-stage framework change?

The proposed fix splits every AI and machine learning project into four stages, each ending in a clear decision. The first three stages typically consume 20 to 40 per cent of the total investment — meaning the time and budget a project uses from idea to production. Their job is to prove the idea deserves the remaining 60 to 80 per cent.

• Strategy and ideation (typically 5 to 10 per cent). Define the problem in business terms, the expected uplift and the available budget. If nobody can say what success is worth, the project stops here, cheaply.
• Data analysis and feasibility (5 to 10 per cent). This is the stage most projects skip. It tests whether the data can support the idea, through data statistics and correlations, external variables that matter, interpretations from the people who know the operation best, and often a site visit. Teams also test critical components, map integrations and required business change, and produce a cost, feasibility and ROI estimate.
• Proof of concept (10 to 20 per cent). Build a model that directionally predicts the outcome, using the simplest data collection possible and a small, auditable set of features. No heavy pipelines yet. Once the evidence is convincing, stop tuning — chasing extra accuracy at this stage is, in the authors' words, "one of the fastest ways to burn a budget."
• Implement, test and productionise (60 to 80 per cent). Now the investment goes in, with confidence. Data is ingested directly from source, pipelines and automations are made production grade, alerts watch for drift, and the model's outputs connect to the systems and people who act on them. MLOps keeps models tested, retrained and redeployed so performance holds over time.

The shape matters more than the exact percentages. Most failed projects spend heavily in stages three and four before stages one and two have earned it. Reversing that order means a weak idea costs weeks, not a year, and a strong idea arrives at production with evidence behind it.

Why can't stage two be skipped?

Most organisations jump straight to a proof of concept. It feels like progress: something gets built, and there is a demo to show the board. But a proof of concept answers only one question — whether a model can work. It does not test whether the data, the foundations, the running costs or the business case will hold up in production. That gap is why nearly half of proofs of concept get scrapped before they get there.

A readiness and feasibility assessment asks those questions first, and the analysis calls it the one stage not worth compromising on. It covers three things:

• Readiness and feasibility, together. Feasibility asks whether a specific idea can deliver. Readiness asks whether your data, platforms, access controls, governance and people can support it. An idea can be feasible in theory and still fail on weak foundations — which is where many of the failures RAND describes begin.
• The cost of running it, not just building it. The authors estimate model consumption based on expected usage and recommend design choices that keep running costs predictable. As AI running costs become a visible line item, boards want that number before approving the build.
• "Don't proceed" is a good result. Catching a flawed assumption early is far cheaper than discovering it after months of implementation. The study is designed to test a decision, not to justify one already made.

The point is simple: a cheap early "no" beats an expensive late one.

What should you ask before funding the next AI project?

The analysis closes with three questions for anyone signing off on AI spending.

• What is it worth if it works? If the expected uplift and the available budget are not written down, the project has no way to know when to stop.
• Have we tested the data, or assumed it? A short feasibility phase is cheap insurance against the most common technical cause of failure.
• Who owns it in production? Someone needs to own the pipelines, the monitoring and the change to how people work, from day one — not after the pilot succeeds.

The stakes are straightforward. With abandonment rates climbing from 17 to 42 per cent in a single year, and the average organisation killing nearly half its proofs of concept, the pattern is now measurable at scale. AI does not need to be a gamble: the organisations getting real returns are not the ones with the most ambitious pilots, but the ones that prove value early, decide with evidence, and then commit properly to the hard work of production.