AI was supposed to save time — but it’s costing millions instead

AI was supposed to be the great productivity unlock. Instead, the numbers tell a different story and organizations in Asia Pacific are spending millions for next to nothing.

92% of companies are now investing in it, yet McKinsey finds only 1% are actually seeing a return. An MIT report recently revealed that 95% of generative AI pilots fail. 

Developers report saving more than 10 hours a week with AI tools, but still lose at least six hours chasing context or waiting for human input. For a 500-person company, fragmented knowledge alone translates into nearly $7.9 million a year wasted in duplicated work and re-created documents.

Chang LingYi, Co-Founder, illumi, said: “The companies winning aren’t the ones with better AI, they’re the ones collaborating with AI better.” She believes the promise of AI is being buried under what experts call “cognitive debt”: scattered prompts, siloed experiments, and disconnected workflows that quietly drain productivity.

We found her perspectives intriguing, and decided to find out more…

Why do most AI pilots fail, and how can organizations in Asia Pacific escape the “1% ROI trap”? 

Chang: Most AI pilots fail for one simple reason: they start with a solution, not a problem.

You see this pattern everywhere — leadership declares, “We need AI!” before anyone defines what pain it’s solving. So teams scramble to “find use cases” instead of fixing real inefficiencies.

The second issue is the expectation–execution mismatch. Once leadership sets the AI vision, employees are told to “learn and deliver”. But AI isn’t magic — it’s easy to reach 70% readiness, but the effort to go from 70% to 90% could be really tough, and the last mile of 10% always needs human judgment, domain knowledge, and iteration.

And this is where most projects collapse. AI isn’t plug-and-play. It requires trial, error, and deep involvement from the people who actually understand how the business runs. Even the best AI experts can’t design solutions without that operational context.

The organizations escaping the 1% ROI trap are doing the opposite. They:

• Start with real problems identified by frontline teams.
• Build cross-functional squads that mix domain experts with technical talent.
• Engage AI specialists only after the mission statement is clear—so technology aligns with the real workflow.
• And they treat AI as a continuous learning journey, not a one-time rollout.

I recently spoke with an AI lead at a major enterprise who built a prototype in three weeks. His boss was thrilled — and immediately gave him an unlimited budget to “realize ROI by cutting people”. Within a month, he discovered the truth: AI is not good enough to replace people, and now he’s under enormous pressure.

What is “cognitive debt”, and how is it silently draining millions of dollars from companies? 

Chang: Cognitive debt is the invisible cost companies pay when people rely on AI too much — and share too little.

On the surface, everyone looks productive. Each person is talking to their own AI, generating reports, and automating tasks. But underneath, the organization is bleeding knowledge. None of that learning compounds because it’s trapped inside private chat windows. When someone leaves, that context disappears with them.

But cognitive debt isn’t just organizational — it’s also personal. When individuals start outsourcing too much thinking to AI, they gradually lose the ability to question, synthesize, and create from first principles. They become editors of AI output instead of true problem solvers.

I’ve seen this play out in meetings: everyone brings AI-polished work that looks perfect — but nobody dares to challenge it. The team nods, but the substance is thin. Over time, this erodes creativity and critical thinking.

In this end? Teams move faster on paper, but slower in reality — because they’re constantly redoing, refining, or fixing shallow work. The hidden cost adds up to millions in lost time and duplicated effort.

The solution is context engineering — connecting those isolated AI conversations so one person’s insight becomes everyone’s foundation. When AI work is shared, reviewed, and iterated together, teams start compounding intelligence instead of accumulating cognitive debt.

In other words, AI should amplify human thinking, not replace it. The moment we stop thinking critically and start accepting AI output at face value, we trade speed for depth—and that’s when cognitive debt becomes a real liability.

Please explain what you mean when you say: “The companies winning aren’t the ones with better AI, they’re the ones collaborating with AI better.”

Chang: Every company today has access to the same models or tools — OpenAI, Anthropic, Gemini. The real differentiator is how teams collaborate with AI.

Most tools were built for individuals, so organizations end up with thousands of private chats and no shared learning. But companies that build collective AI habits — where every experiment is visible, where prompts and outputs are reusable — learn exponentially faster. Their AIs don’t just get smarter; their people do too.

Collaboration turns AI from a productivity tool into a knowledge multiplier. It’s how small teams start outpacing big enterprises still stuck in silos.

What are some real use cases you have come across that demonstrate how teams are moving from fragmented AI experiments to collective intelligence in Asia Pacific?

Chang: We’ve seen this transformation firsthand.

Innovation teams at a global manufacturing company saved tremendous time using illumi because we support multiple AI models in one place. They can lay out their thoughts visually — almost like building a mind map — where AI works alongside them in that same space. Every idea, prompt, and finding now lives on a shared canvas, so others can build on it instead of starting from scratch.

Consulting teams at BIM use shared AI templates to co-create client frameworks. What used to take three separate tools — ChatGPT, Notion, and Excel — now happens in one connected workspace. The result is faster delivery and higher consistency across projects.

Higher education institutions such as Imperial Business School and SUTD are partnering with illumi to help students and researchers collaborate and test their AI workflows. They’re improving prompt and context-management skills while learning something deeper — that innovation is a team sport.

Across Asia Pacific, we’d like to facilitate the shift from chasing “better AI” to engineering better collaboration first, and that’s what unlocks real competitive advantage.