The real reason your AI project will fail

95% of procurement AI pilots stall. What makes the 5% scale?

Article originally published on LinkedIn

MIT’s State of AI in Business report put a number on something a lot of procurement leaders already suspected: 95% of GenAI pilots fail to deliver any measurable return. Despite $30-40 billion in enterprise investment, most organisations are getting nothing back for it.

And here’s the interesting part. MIT found this split isn’t down to model quality, and it isn’t down to regulation either. It’s determined by approach. A small number of teams are pulling millions in value out of AI. Nearly everyone else is stuck. Same technology, wildly different outcomes.

Twelve months on, and if anything, the picture has hardened rather than improved. S&P Global Market Intelligence found that the share of companies abandoning most of their AI initiatives climbed to 42% in 2025, up from 17% the year before. And Gartner predicts that through 2026, 60% of AI projects unsupported by AI-ready data will be abandoned. 

If AI is becoming more accurate and powerful, why are abandonment rates still rising? Clearly the problem isn’t with the technology. So, what’s actually different about the 5% that work?

Trust, mostly. Not trust in the abstract sense, but the practical, everyday kind: can a buyer see why the system scored a supplier the way it did? Can a category manager challenge a recommendation before it goes anywhere near a contract? Can someone trace what happened, and why, six months later when a decision gets questioned?

Systems that can’t answer those questions get quietly abandoned. Not because the output was wrong, but because nobody could stand behind it.

MAKE’s survey of business decision-makers backs this up from the people-side. 84% are genuinely excited about AI’s potential. But 55% are worried that losing human judgement from critical workflows will create new problems, and 39% admit they don’t feel ready for AI-enabled ways of working. That’s not due to people’s resistance to change, it’s a reasonable response to tools nobody has properly explained.

‘Black box AI’ is any system that produces an outcome without showing its working. In procurement, that shows up as unexplained supplier scores, contract terms nobody can trace back to a rationale, and forecasts nobody can interrogate when the number looks off. Over time, that opacity doesn’t just slow adoption down. It hides bias, errors, and flawed assumptions until they’ve already caused damage.

The fix isn’t complicated, even if it’s rarely done properly. Good procurement AI will:

• document its data sources and logic,
• keep a full audit trail of every decision and intervention,
• build in human review for anything with real financial or contractual weight, 
• and let someone pause or reverse an action when it’s got something wrong. 

There’s a regulatory angle here too. Frameworks like the EU AI Act are shifting the question regulators and insurers will ask after something goes wrong, from simply ‘what happened’ to ‘were adequate safeguards in place to prevent it.’ That puts governance and transparency on the same footing as performance, not as an afterthought bolted on once something’s already gone sideways.

For procurement leaders picking AI tools right now, that changes what the vendor conversation should look like. Ask less about accuracy claims and more about explainability: can I see why the system made this call? Can I override it? Who’s accountable if it gets something wrong? A vendor that can’t answer clearly is handing you their risk, not their value.

Decisions on AI vendors will increasingly be made on ROI, risk and strategic value rather than vendor positioning, because adoption has become a regulatory and governance question first and a technology question second.

Where Market Dojo stands on this

These principles are built directly into how Synergy AI Agent works, and it shapes what we choose not to build as much as what we do.

Every AI capability in the platform is designed to sit alongside a procurement professional, not in front of them. The AI Sourcing Assistant gives users guidance at the point of need rather than assuming they already know the answer. Intelligent event creation turns requirements, SoWs, and specifications into structured RFx events in minutes, cutting the admin without removing the person deciding what ‘good’ looks like. And every deployment can be configured around your organisation’s own sourcing rules, terminology, and governance controls, so the AI reflects how your business actually sources, not a generic template.

This sits inside what we call our Sustainable AI Philosophy: three commitments covering people, pricing, and planet. 

• For people, AI is built to remove admin and friction so procurement experts can spend their time on supplier relationships and commercial strategy, not to justify smaller teams. 
• On pricing, we don’t do ‘AI-washing’ or charge inflated premiums for features that should be standard; every AI capability ships as part of the plan you’re already on. 
• For our planet, we favour lean, purpose-built AI over feature bloat, because more compute isn’t the same thing as more value.

This article is based on original content from Market Dojo’s playbook on AI in procurement, covering human-centred design, AI-powered sourcing, predictive intelligence, and how to scale AI value once the pilot stage is behind you.

Get the full playbook here: Innovate: A Guide to Leveraging AI to Accelerate Procurement Excellence