Walk into almost any boardroom today and you will hear a version of the same conversation. The organisation has run a generative AI proof-of-concept. A few departments have built chatbots. Someone has commissioned a data platform. Budgets have been approved, vendors engaged, dashboards built.
"But what is actually different about how the company runs before the chatbots, dashboards and platforms?" someone asked. The room goes into a deafening silence.
An MIT study cited found that around 95 per cent of large organisations are struggling to show measurable returns on their AI spend, with most initiatives parked permanently at the pilot stage, rather than scaling across the enterprise.
Almost everyone uses AI now, but very few have results.
Digital transformation asked organisations to digitalise processes. AI asks that organisations rethink how work itself is designed, how decisions are made, and how humans and machines share responsibility.
The real challenge is the business operating model. How we are making the decision rights, and the capability of leaders to think and act differently when new technology emerges. The leaders who succeed be those who treat AI as a business transformation agenda that happens to involve technology.
Here are five things that separate the companies that get somewhere in their AI transformation, from the ones that are just running in place.
1. The technology is the last question, not the first
"We need an AI strategy" is not a strategy. It's anxiety with a budget attached.
The companies making progress start somewhere less exciting. They pick three or four problems that genuinely hurt. The claims process that takes eleven days when competitors do it in four, the customer segment quietly churning at 18 per cent, the month-end close that eats a week of finance's life every month. They then ask whether AI helps. Sometimes, the honest answer is no, and they fix the process instead. That's a win too, though it never makes the annual report.
A useful gut check before approving anything: if this works brilliantly, which number moves, by roughly how much, and who will be embarrassed if it doesn't? If nobody can answer, you don't have an initiative. You have an experiment, which is fine. Just fund it and govern it like one, and stop pretending it's on the critical path.
There's a related discipline worth borrowing. Internal hackathons and innovation sprints tend to produce far more durable outcomes when they start from a real business problem owned by a real person, rather than from a demo of what the technology can do.
2. Stop running AI like an IT project
Traditional IT governance assumes you can specify a thing, build it, test it, ship it and eventually retire it. AI breaks most of those assumptions at once.
Models are probabilistic, not deterministic. They depend on data that drifts. They often get better, or quietly worse, months after go-live. And the value frequently shows up not as a single deliverable but as thousands of small improvements in speed and judgement that no capital approval form was designed to capture.
Force that into a classic ROI model and you get two predictable failures. Good initiatives die in year one because the benefits aren't legible yet. Bad ones survive for years because too much has already been spent to admit otherwise.
What works better is unglamorous: small tranches of money released against evidence, a genuine willingness to kill things, and metrics that go beyond cost savings to include adoption, error rates, decision quality and what people are doing with the time they got back. Every model in production needs a named owner responsible for how it behaves next quarter, not just how it performed at launch.
3. Your managers need to speak the language, not write the code
The most common failure we see isn't technical, it's translation. Business leaders can't say precisely what they want AI to do, so they ask for something vague. Technical teams can't see which constraints actually matter commercially, so they optimise for the wrong thing.
Both sides are competent, but neither can understand the other properly. The initiative dies somewhere in the gap, usually without a post-mortem. Fixing this doesn't require executives to learn Python. It requires enough fluency to ask uncomfortable questions in the room: What's this trained on? What does a false positive cost us here? Which decision are we actually automating, and who's accountable when it's wrong?
This is why AI literacy is increasingly being treated as a leadership skill. For companies, the practical version is to build capability at three levels at once. Fluency at board and C-suite, applied capability for middle managers, and depth for the specialists.
Most organisations do the first and third and skip the second, then wonder why nothing lands. Middle managers are the people being tasked to redesign how their teams work with the addition of AI. They need something practical. Sending them a policy document is not a plan.
4. The boring foundations are the whole game
Before an organisation can run intelligence across its enterprise, it needs clean, connected, governed data. And standardised processes generate that data consistently.
You cannot run intelligence across an enterprise that can't agree on what a customer record looks like. Fragmented systems, inconsistent master data, processes that differ by team and country. Layer AI on top of that and you get greater inconsistency.
SMU's own experience makes the point. Working with SAP and implementation partner ABeam Consulting, the University spent considerable effort on 'Project Optimus', modernising its corporate backbone and integrating finance, HR and procurement workflows onto a connected cloud foundation. It won the Line of Business Transformation Award at the SAP Customer Excellence Awards in August 2026.
The modernisation was positioned as building the foundation for future innovation, not as an endpoint. Organisations that skip this stage find that their AI ambitions collide with fragmented systems, inconsistent master data and processes that differ by team.
5. Be honest about what happens to people's jobs
Every serious AI deployment redesigns somebody's work. Pretending otherwise doesn't reassure anyone; it just moves the resistance underground, and a polite "we tried it, it didn't suit our team."
Be clear what's changing. Say what's being automated. Tell people what should be done instead, and commit to the reskilling in the same breath as the deployment.
Governance on responsible use, data privacy, bias, transparency, escalation paths when the system gets it wrong should be established before scale, because retrofitting trust is far more expensive than designing for it.
And keep accountability human. No regulator, customer or employee has ever been satisfied by "the model decided." Having clarity about who owns which decisions is what allows an organisation to move quickly without moving recklessly.
AI Transformation is a Leadership Shift
Taken together, these five describe a shift in posture. They're about how capital gets allocated, how progress gets measured, who gets trained, and what leaders are willing to say out loud.
From technology-led to problem-led. From project delivery to continuous capability. From centralised innovation theatre to distributed, accountable adoption.
But these are genuinely difficult, because it asks leaders to change how they allocate capital, measure progress, and redefine new roles. The organisations pulling ahead in Asia are typically not the ones with the most impressive models. They are the ones where leadership has done the harder work of redesigning the conditions in which those models can create value.
So, the real question for every executive team is no longer whether to invest in AI. It is whether the organisation is ready to convert a significant investments into something that lasts.
Explore AI related courses by SMU Executive Development
Explore bespoke SMU Executive Development's suit of AI programmes designed to equip leaders with the mindsets and toolkits to thrive in today’s AI era.
- Applying AI to Sustainability: From Compliance to Value Creation
Combining sustainability strategy, AI fluency, business value, and execution discipline, participants will use AI to structure complex information, challenge assumptions, compare options, stress-test initiatives, and design action plans. - Chief AI Officer Programme
The programme equips you with the strategic acumen, technical literacy and leadership capability to drive AI transformation and develop an AI Transformation Roadmap in your organisation. - Strategic Leadership with AI
The programme lays out the roadmap for AI for business leaders who are ready to move beyond adoption toward reinvention, building intelligent enterprises, redesigning operating models, and leading confidently into the algorithmic decade. - AI Governance
The programme addresses governance, compliance, risks, and ethical considerations in AI adoption, equipping directors with the tools and knowledge needed for informed, data-driven decision-making about AI strategy and oversight at the board level.