Sutowo Wong wants to tell you something uncomfortable. The managing director of AI & data at Temus — a Temasek Holdings digital transformation consultancy — stood before a room of data and digital officers at the 9th CDDO Asia Summit in Singapore and laid out a chart. A blue line, a gray line and a chasm between them.
The blue line is what AI should be delivering to your organization, given the pace of technological progress. The gray line is what it is actually delivering. The gap between them, Wong estimates, is roughly ten trillion dollars. Globally, unrealized, and sitting there like an unread email. “There’s so much that AI can do,” he told the audience, “but we are hardly scratching the surface.” He’s right. And if you’re a CDO, this is squarely your problem.
The tempo problem
Here’s the paradox that should keep every data officer awake. AI model breakthroughs now arrive every few months. GPT-4 to GPT-4o. Claude 3 to Claude 3.7. DeepSeek rattling cages out of nowhere. Meanwhile, the average enterprise AI deployment? Wong puts it bluntly: “Sometimes it can be five years.”
In AI time, that’s geological.
Wong calls it a “tempo mismatch,” and it’s not a technology failure but a management one. The models are ready. The infrastructure is increasingly mature. What lags is everything else: operating models, workflow redesign, governance structures, and the all-too-human tendency to commission another pilot instead of shipping something real.
“It is not about the technology,” Wong said, with the tone of someone who has said this too many times. “It is a problem of change management. It is a problem of integration — not just in a technical sense, but in the sense of how do you integrate that into your workflow?”
The AI triangle: Value, speed, sovereignty
Wong’s framework for closing the gap has three legs. He calls it the “AI Triangle,” and he emphasizes that you need all three to score.
Value. Achieve measurable, organization-wide outcomes. Not a science experiment in one division. Not a productivity stat no one can convert to revenue. Hard savings, or don’t bother.Speed. Prototype in hours. Ship to production in weeks. Wong’s team can build a working prototype in six hours. That same project? In production in under 12 weeks. “Gone are the days when you can spend years on the transformation program,” he added.Sovereignty. AI contextualized to your organization — your data, your architecture, your existing investments. Not a generic model answering generic questions about your highly specific, very real business.
Wong spent additional time on sovereignty. Most enterprises — especially in regulated industries — cannot simply hand their data to a cloud-based LLM and call it a day. Your competitive edge lives in proprietary data. Your compliance obligations live there, too. It means a capable model is the starting point, not the solution.
The agent economy is already here
Wong’s bigger argument isn’t about LLMs but what comes after them. “We have entered the AI agent economy,” he declared, tracing AI’s evolution from computer vision (AI can see) to transformers (AI can understand) to diffusion models (AI can create) to agentic systems (AI can act).
The shift, as he frames it, is from conversation to execution. AI is no longer a search engine you type questions into. It’s an autonomous actor that plans, maps tasks, executes them, and reports back. For a CDO, this goes beyond an incremental upgrade; it’s a category change. The governance implications alone could occupy a full offsite.
Wong offered a telling governance example: imagine two AI agents in your organization — one in finance, one in HR. What happens when the finance agent queries the HR agent for compensation data? Should that be allowed? Who approved it? What’s the fallback if it fails? These are not theoretical questions. They are the operational reality of deploying agents at scale, and most enterprises have answered exactly none of them.
He also drew a sharp line between probabilistic and deterministic AI — a distinction critical in financial services. “You want to execute a transaction — wire an amount from one account to another. You want it to be ‘about right’? A million dollars, plus or minus? No way.” Determinism matters. Especially when the stakes are real.
Proof points that aren’t vaporware
Wong is refreshingly numbers-oriented. He doesn’t offer transformation in the abstract; he offers receipts.
At a genomics lab, his team cut processing time from over 60 hours to under 90 minutes. Compute costs dropped 85 percent. At a cybersecurity client, AI agents now handle the full vulnerability management loop — scanning, severity triage, remediation recommendations and stakeholder communications — autonomously, eliminating the budget line.
At an insurance firm, AI-powered training avatars doubled agent activation rates. The context matters: only three to five agents survive out of every hundred hired. Training them is expensive and mostly wasted. Using AI to simulate demanding (and sometimes unreasonable) customers, then scoring and coaching agents, makes the economics suddenly sensible.
The common thread? None of these are pilots. They’re production deployments. “Experiment once, deploy many times” is Wong’s mantra. If you’re still running your fourth proof of concept on the same use case, you already know who you are.
The CDO imperative
Wong’s prescription for CDOs is uncomfortable in the best way. Stop waiting for the data foundation to be perfect before you deploy AI. Start with your most valuable, most ready-to-use cases, then work backward to identify exactly which data you need, and prepare only that. Multi-year data transformation programs launched in anticipation of future AI projects are a form of organizational procrastination. “Don’t wait for two years. Don’t wait for five years.”
The operating model is equally non-negotiable. If every AI use case requires a bespoke negotiation with IT to extract and clean data, you will never move fast enough. Speed demands structure — clear decision rights, clear escalation paths, and clear ownership. Without that scaffolding, AI transformation becomes a series of heroic one-off sprints that don’t compound into anything.
On workforce, Wong’s recent work at Temus emphasizes that AI literacy is now table stakes — as fundamental as knowing how to use Excel was in 1995. But literacy isn’t enough. He argues for a deeper cognitive shift: developing the instinct to delegate. Knowing which tasks to hand to AI entirely, which to do solo, and which to do in collaboration. “How do you start to build an instinct to delegate work with AI?” is the question he wants CDOs asking their whole organization — including themselves.
His parting shot was precise. Don’t outsource your judgment to AI. Use it as a sparring partner, a pressure-tester of your thinking. But keep your craft. “AI is trained on old things. You are there to create new things — and if you rely on AI entirely, you will never be able to create.”
Ten trillion dollars says it’s worth paying attention.
Image credit: iStockphoto/esilzengin