Why most AI projects fail, and how we start
AI use is now common in the UAE. A report by Strand Partners for AWS and the UAE AI Office found that 72% of UAE businesses use AI in 2026, but 47% are still at basic uses.
The wider evidence is blunt. MIT NANDA's 2025 "GenAI Divide" study found that about 95% of generative AI pilots show no measurable P&L impact. Gartner expects 30% of generative AI projects to be abandoned after proof of concept, and over 40% of agentic AI projects to be cancelled by 2027. The causes repeat: no baseline, a chatbot with no clear job, scattered data, no owner, and unplanned running costs.
So we start where a finance team would:
- Measure first. We record today's cost, hours and response times before we build anything.
- One workflow at a time. The most painful, highest-volume process goes first.
- A named owner on your side, and a running-cost budget agreed up front.
- Rules where rules will do. We use AI only where it beats a form or a simple rule.