AI Adoption, Globally: The Good, The Bad & The Ugly

I read the research so you don't have to. McKinsey, MIT, Stanford, Gartner, RAND, PwC, the UN. Twelve months of global studies, one honest question: if AI adoption were a customer experience, how would it be scoring?

Here's the strangest thing about AI in 2026: it is simultaneously the fastest-adopted technology in business history and one of the least trusted. Adoption is up. Spending is up. Trust is down. Returns are wildly uneven. All true at once. That tension is the whole story.

The Good

88% of organisations now use AI in at least one business function (McKinsey). Worldwide AI spending hits US$2.59 trillion this year (Gartner). Enterprise generative AI went from $1.7B to $37B in revenue in two years, the fastest-scaling software category in history (Menlo Ventures).

And where the value lands, it lands fast: Bain found median payback of ~4 months for customer service AI. In CX specifically, the pattern is consistent, churn down, NPS up, cost-to-serve down. But only when AI is woven into core processes. Never when it's stitched on as a bolt-on.

The Bad

MIT's Project NANDA study found 95% of generative AI pilots delivered no measurable P&L impact, despite $30–40 billion invested. RAND puts AI project failure above 80%, roughly double ordinary IT projects. 42% of companies abandoned most of their AI initiatives last year, up from 17%.

And it's not the technology failing. MIT was explicit: the difference between the 5% and the 95% isn't model quality. It's approach. No defined success. Weak data foundations. AI dropped into old workflows instead of workflows redesigned around it.

"Meanwhile employees at over 90% of firms are using personal AI tools where official pilots failed. Your people aren't resisting AI, they're routing around your governance to get to it. That's not a compliance problem. It's a design problem."

The Ugly

Nearly 50,000 job cuts linked to AI this year already. Forrester predicts half of AI-attributed layoffs will be quietly rehired, offshore, or at lower salaries, because companies cut people for AI capabilities that don't exist yet. 55% of employers already regret it.

Only 46% of people globally are willing to trust AI, and trust has FALLEN as adoption has risen. Gen Z, the heaviest users, are the most pessimistic. And the divide is widening: less than 10% of global AI governance frameworks come from the Global South.

So What?

Here's the reframe I think we all need.

Every general-purpose technology in history, electricity, computing, the internet, followed a J-curve. Costs come first. Benefits come later. Factories took decades to reorganise around electricity before productivity showed up. That's not failure. That's how transformation works.

The problem is that AI moves fast, so we expect it to pay fast. We've confused the speed of the technology with the speed of the transformation. The model updates every quarter. Your organisation doesn't.

"When 95% of pilots show no return after 18 months, that's not proof AI is hype. It's proof that most organisations skipped the slow work and jumped straight to the shiny work."

The 5% aren't luckier. They just did the boring work first.

Now What?

Honestly? Pause. Breathe. Maybe even meditate on it. Because the highest-leverage AI move available to most organisations right now is not another pilot. It's the important work you've been putting off:

1. 01

   Get a sound strategy. Not an AI strategy, a business strategy that AI serves. What are you actually for? What does this make better, for whom?

2. 02

   Map the customer journey. Really map it. You cannot consciously automate an experience you've never consciously designed. Every AI winner in the CX research embedded the technology into a journey they deeply understood.

3. 03

   Set a 15–20 year vision. Not a 12-month roadmap. If AI is a decades-long transformation (it is), then your vision has to outlast the hype cycle. Decide who you want to be on the other side of this, then build up from there, layer by layer.

4. 04

   Then build. Small, embedded, measured. Define success before you start. Treat friction as feedback. Bring your people with you, they're already using this technology in the shadows; legitimise that energy instead of suppressing it.

"AI adoption is not a technology programme. It's a human transformation wearing a technology costume."

The organisations that win won't be the ones that moved fastest. They'll be the ones that were still standing, clear on their purpose, close to their customers, kind to their people, when the J-curve finally turned.

Slow is smooth. Smooth is fast. Even in AI. Especially in AI.

Sources: McKinsey State of AI 2025 · Gartner · Stanford HAI AI Index 2026 · MIT Project NANDA · RAND · S&P Global · Bain · Menlo Ventures · Forrester · Challenger Gray & Christmas · PwC AI Jobs Barometer · University of Melbourne/KPMG Trust in AI · UNDP · Brookings