A regional bank's CFO told us something fairly blunt during a planning conversation last year: AI spend had roughly tripled across the organisation over two years, and when asked to point to where that spend had shown up in measurable cost savings or revenue, the honest answer from his team was "it's hard to say precisely." That's not a unique situation. It's close to the norm we've seen across a number of GCC enterprises, and it's worth being honest about why, because the explanation usually isn't "the AI didn't work."
The actual pattern we keep seeing
Most of the AI spend in question went toward a collection of disconnected pilots and point solutions — a chatbot here, a document processing tool there, a recommendation engine somewhere else — each individually reasonable, none of them tied to a specific, pre-agreed financial outcome that anyone measured before and after. You can spend meaningfully on AI for two years and have very little to show in hard numbers, not because the AI underperformed, but because nobody set up the measurement in a way that would let you know either way.
ROI requires a baseline, and most projects skip it
This sounds almost too obvious to write down, and yet it's the single most common gap we find: before deploying an AI system, did anyone rigorously measure what the current process actually costs, in time and money, with enough precision to compare against afterward? Frequently, no. The pre-AI baseline was a rough estimate, or didn't exist at all, which means even a genuinely successful deployment has no real way to prove its value, because there's nothing solid to measure against.
The pilots that get funded aren't always the ones that matter
There's also a selection problem. AI pilots in a lot of organisations get chosen based on which department has an enthusiastic champion and some spare budget, rather than which process has the highest-value, most measurable opportunity for AI to actually move a number leadership cares about. An enthusiastic team running a moderately useful pilot will often get funded before a less glamorous, higher-value opportunity that nobody's championing internally. The result is a portfolio of AI projects that are individually fine and collectively underwhelming relative to spend, because the prioritisation logic was about internal politics rather than actual financial opportunity.
Integration costs that never get counted
A subtler issue: the sticker price of an AI tool is rarely the real cost. Integration with existing systems, change management for the staff who now work differently, ongoing monitoring and retraining — these costs are real, often substantial, and very frequently excluded from the original ROI case because they're harder to estimate upfront. When the full cost picture only becomes visible eighteen months in, the apparent ROI looks a lot worse than the original pitch promised, not because the AI underdelivered, but because the original cost estimate was incomplete from the start.
What actually fixes this
The organisations getting real, measurable ROI from AI in this region share a specific habit: they measure the baseline rigorously before touching anything, they pick the use case based on quantified opportunity size rather than internal enthusiasm, and they count the full cost — integration, change management, ongoing operations — in the original business case rather than discovering it later. None of this requires better AI. It requires more discipline in how the investment case gets built before a single line of code is written, which is a less exciting story than "we deployed cutting-edge AI," but it's the actual difference between spend that shows up in the numbers and spend that doesn't.