The most powerful narrative of the past two years in the ERP world is clear: AI will finally fix project failure rates that have not changed for decades. Sales presentations, product launches and industry events repeat this promise every day. Yet the data published by research organisations points to a different picture: a significant share of AI-supported ERP projects will fail too. This article is not opposition to technology; on the contrary, it is an effort to separate a genuinely powerful technology from the hype. Because the problem discussed in the first article in the series — making commitments without measuring — does not disappear under an AI label; it merely gets new packaging.
Why Are the Numbers So Quiet?
The louder the marketing, the quieter the research data. According to Gartner’s prediction, more than 40% of agentic AI projects will be cancelled by the end of 2027; the reasons cited are escalating costs, unclear business value and inadequate risk controls. The same organisation’s ERP-specific prediction is even more striking: according to the "Predicts 2025: Revisit ERP Strategies" report, fewer than 10% of organisations that implement agentic AI in their ERP will have achieved measurable value by 2027.
The picture is no different in the pilot arena. Deloitte’s 2025 research shows that 38% of organisations are trialling AI agents at the pilot stage, but only around 11% are able to use these systems actively in production — the picture the industry calls "pilot purgatory": an impressive demo, a project that never reaches production.
An honest note: most of these figures are predictions and survey findings, not measured facts — they are not definitive results. But the fact that they all point in the same direction should not be dismissed as coincidence: the AI label alone does not change the success rate.
So Why Do They Fail?
The answer is not that the technology is bad; it is that a good technology is built on bad ground. The three root causes on which AI-supported ERP initiatives stumble form a single chain in which each feeds the others.
1. Intelligence Built on Dirty Data
AI cannot exceed the quality of the data it is fed. Duplicate customer account records, inconsistent stock codes, master data that is never updated — a model built on top of these does not make smarter decisions; it makes wrong decisions faster. In the same ERP report, Gartner predicts that 70% of organisations will lack AI-ready ERP data by 2027. IDC research shows the other side of the coin: in the AI era, the real differentiator is not algorithms but data readiness and infrastructure — those with the cleanest data will stand out.
2. Automation Built on Unmeasured Ground
In most organisations, processes are known "as they should be": the flow written in the procedure document, the approval chain on the organisation chart. Yet the "as they actually work" state of the business — orders taken by phone, approvals circulating in Excel, person-dependent exceptions — has often never been measured. Automation blows up precisely on this gap: an agent built on the process on paper does not know the real process. Automating a faulty process does not eliminate the error; it scales it.
3. Unaudited Trust
The third cause is not technical but managerial. Forrester’s concept of the "trust tax" sums up the situation clearly: every autonomous decision must be loggable and defensible to an auditor — and today that cost is high for most organisations. There is no trust in a decision that cannot be traced; without trust there is no use either. The name of the solution is human-in-the-loop: where AI stops and where a human steps in to give approval should not be a feature added to the architecture later, but a boundary drawn from the start.
The three causes share the same common denominator: the problem is not AI itself but the ground it is built on.
Would You Operate Without an X-Ray?
A proposal made without measuring us is a prescription written without a diagnosis. In this equation, AI is a superb X-ray machine: it scans fast, gives a sharp image and catches the pattern the human eye misses. But an X-ray machine on its own heals no one — it takes a physician to read the film, make the diagnosis and approve the treatment. Today, a significant share of the AI-supported promises on the market propose selling the X-ray machine and laying off the physician. The right order has not changed: X-ray first, then diagnosis, treatment last — and at every step the decision stays in human hands.
4 Questions to Ask a Firm That Says "We Use AI"
When you hear the phrase "AI-supported" at the proposal table, the following four questions genuinely separate the hype. If the answers are clear, you are facing a serious approach; if they are evasive, the label is nothing more than packaging.
- Is our data of a quality to feed this AI? If not, how do we fix it first? (A "we can start straight away" answer given without measuring data quality is an invitation to the first root cause.)
- Do you know the processes as they should be, or as they actually work? Without a measurement of the current state, automation is built on the process on paper.
- Who will audit the AI’s decisions, and how? Are logging, traceability and an audit mechanism defined from the start?
- Where does the AI stop and where does a human step in — can you show us? If the human-in-the-loop boundary has not been drawn, you are the one who pays the trust tax.
Conclusion: An Accelerator, Not a Saviour
AI is not a saviour but an accelerator: it drives a badly designed project into the wall faster, and makes a well-designed project stronger. What is decisive is not the technology but the ground it is built on — the cleanliness of the data, the measuredness of the processes and the design of oversight. The third article in the series will look at the other side of the coin: what AI genuinely changes in ERP projects, and through which mechanism — not in percentages, but in how things work.
This article is the second in the growing «From Man-Days to Value» series.
1. Is Consulting That Sells Man-Days Coming to an End?
2. Let’s X-Ray the Promises of AI (this article)
3. What Has AI Made Cheaper in ERP, and What Has It Made Valuable?
4. Diagnosis First: What Does the New Path in ERP Look Like in Practice?
5. The Unmeasured Layer of Digital Transformation: Leadership
6. Why Do Reports
Stay on the Shelf in Digital Transformation?
7. (coming soon)
To make sure you do not miss the rest of the series and new analyses, subscribe to the MindDX-Digital Excellence newsletter.
