By Dr Kevin Shepherdson, CEO, Straits Interactive
When organisations talk about “AI transformation”, the phrase can quickly become too broad to be useful. One company may introduce an AI assistant to help employees draft emails and call that transformation. Another may redesign entire workflows around autonomous agents. A third may create new AI-enabled products or fundamentally change its operating model.
These are very different changes.
A more useful way to think about AI transformation is to look at it through several connected lenses: how deeply AI changes the organisation, how mature the organisation is in its use of AI, how work is divided between humans and machines, what business outcome the organisation is pursuing, and whether the organisation has the capability to manage the resulting opportunities, risks, obligations and economics.
Together, these lenses provide a practical way for leaders—and particularly the Chief AI Officer - to decide where the organisation is today, where it wants to go, and what needs to change to get there.
1. Start with the AI Change Impact Ladder
The first question is not simply, “Are we using AI?”
It is:
What is AI actually changing?
The AI Change Impact Ladder describes an increasing depth of organisational change:
Task → Workflow → Function → Product/Service → Business Model → Organisation
At the task level, AI may help an employee summarise a report, draft a proposal or analyse a spreadsheet. The work itself has not fundamentally changed; one activity has simply become faster or easier.
At the workflow level, AI begins to change how several tasks fit together. An employee may no longer write a document, send it for review and manually incorporate comments. An AI-supported workflow could conduct research, produce the first draft, check it against requirements and send exceptions to a human reviewer.
Move another level up and an entire function may start operating differently. HR, customer service, marketing, legal or learning and development may redesign roles, processes and decision points around AI.
At the product or service level, AI becomes part of what customers actually buy or experience. A training company might turn its expertise into an AI-augmented tutoring service. A professional services firm might provide an AI-enabled advisory service. A manufacturer could embed intelligence into maintenance or customer-support services.
Beyond that lies a change to the business model itself - how the organisation creates, delivers and captures value - and eventually organisation-level transformation, where structures, capabilities, roles and ways of working are redesigned around intelligent systems.
This distinction matters because a company can be doing a great deal with AI without actually transforming itself. AI transformation becomes meaningful when it changes priorities, workflows, products, decisions and eventually the organisation itself.
2. Overlay the AI Maturity Framework
The Change Impact Ladder tells us what AI is changing.
The AI Maturity Framework tells us something different: how capable the organisation is of using AI responsibly and sustainably.
The progression is:
Stage 0 – Freemium → Stage 1 – Use → Stage 2 – Create → Stage 3 – Deploy → Stage 4 – Govern → Stage 5 – Manage
At Stage 0, employees are often experimenting with free or personal AI tools. There may already be considerable activity, but little organisational visibility or control.
At Stage 1, the organisation formally allows or provides AI tools such as copilots and assistants.
At Stage 2, people begin creating prompts, applications, workflows, agents and other AI-enabled capabilities rather than merely consuming generic tools.
Stage 3 represents an important transition. AI is deployed into real organisational workflows, where failures may now affect customers, operations, employees or business decisions.
Stages 4 and 5 progressively institutionalise governance and management: ownership, controls, monitoring, assurance, costs, dependencies, performance and continuous improvement become part of normal management practice.
This creates an important insight.
Impact and maturity are separate dimensions.
A marketing department may be at the Use stage while applying AI mainly to tasks. The technology team may already be at Create, building internal agents. Customer service may be moving towards Deploy, while HR may deliberately remain at a lower level because its decisions involve greater human and regulatory sensitivity.
There is therefore no single organisational AI maturity number that tells the whole story. Different departments—and different AI applications within the same department—can sit at different stages.
3. Understand the Movement from Augmentation to Automation
A third dimension concerns the relationship between humans and AI.
Work can move broadly through:
Manual → AI-Augmented → AI-Automated/Agent-Executed → Human Review or Human Re-entry
The first transition is augmentation. AI helps a person perform the work, but the person remains actively involved.
The second is more consequential. With automation and increasingly with agentic AI, the machine may execute several steps itself—planning, retrieving information, using tools, making intermediate choices and initiating actions.
This is where leaders need to be careful about confusing technical automation with economic or organisational automation. AI may technically perform a task yet still require significant human checking, exception handling, integration, monitoring and accountability. Evidence from early enterprise experience repeatedly shows that speeding up one step can simply move the bottleneck somewhere else.
The better question is therefore not:
“Can AI automate this?”
It is:
“What should AI do, what should humans continue to do, where should humans intervene, and does the redesigned workflow actually produce a better outcome?”
That last question introduces economics. Automation that saves ten minutes but creates fifteen minutes of checking is not transformation. Nor is an autonomous agent valuable merely because it completes more steps without human intervention.
The relevant measures include time, quality, error and rework rates, human review effort, customer outcomes, risk and ultimately the economics of the complete workflow.
4. So What Exactly Is AI Transformation?
Putting the first three lenses together helps clarify a term that is frequently overused.
AI transformation can operate at three broad levels.
Improvement occurs mainly at the task and workflow levels. The objective is productivity, quality, consistency or speed.
Differentiation occurs when AI begins changing functions and, particularly, products and services. The organisation is no longer merely doing existing work faster; it is creating capabilities or customer experiences competitors may not offer.
Transformation in the stronger sense appears when AI begins changing the business model or organisation itself: how value is created, how work is structured, where human expertise is deployed, how decisions are made and what capabilities the organisation needs.
None is inherently superior.
A highly profitable task-level use case may be more valuable than an ambitious organisation-wide transformation programme. The issue is knowing which kind of change you are pursuing and ensuring that the ambition does not exceed the organisation's maturity.
This also prevents executives from declaring an organisation “AI transformed” merely because hundreds of employees have access to copilots.
5. The Five Plays: How Transformation Is Actually Managed
Once an organisation understands where change is occurring, how mature it is and how human-machine work is being redesigned, it still has to execute the transformation.
This is where the Five Plays that every Chief AI Officer or AI Transformation office should consider:
Strategy → People & Culture → Technology & Data → Governance & Risk → Economics & Control.
They should not be treated as five sequential projects. They are five perspectives that must continually inform one another. They are interconnected: technology decisions create governance consequences; workforce limitations affect strategy; and cost pressures may eventually force technology and model choices to change.
Play I — Strategy
Strategy establishes what the organisation is trying to achieve and why.
The maturity framework helps determine what the organisation is ready to attempt. The Change Impact Ladder clarifies whether the ambition is improvement, differentiation or genuine organisational transformation.
But strategy also needs a broader management discipline.
OCEG's concept of Principled Performance is particularly relevant here. It defines the desired outcome as the ability to reliably achieve objectives, address uncertainty and act with integrity.
Applied to AI, this prevents strategy from becoming a one-dimensional pursuit of productivity.
The organisation must consider its objectives, the opportunities AI creates, the obstacles that could prevent those objectives being achieved, and the obligations it has to employees, customers, regulators and other stakeholders.
OCEG's broader Learn → Align → Perform → Review logic provides a useful management cycle: understand the environment first, align priorities and accountability, execute through actions and controls, and continuously review whether those arrangements still work.
So the strategic question becomes:
What should we try to achieve with AI, given the opportunities, uncertainties, obligations and capabilities we actually have?
Play II — People & Culture
Technology does not transform an organisation by itself.
People decide whether AI is trusted too much, challenged appropriately, used creatively, ignored, misused or incorporated into better ways of working.
The People & Culture Play therefore develops organisational capability across five dimensions:
Knowledge, Skills, Tools & Technology, Processes, and Transformation Mindset.
Knowledge includes understanding AI's value, risks and constraints.
Skills increasingly extend beyond prompting to judgement: framing problems, evaluating outputs, checking sources, recognising weak reasoning and deciding when AI should not be relied upon.
Tools must be appropriate to the work and risk. Processes need redesign rather than simply adding AI onto existing work. And the transformation mindset determines whether people are willing to experiment, learn, challenge assumptions and adapt without surrendering professional judgement.
The Change Impact Ladder becomes particularly important here because the required human capability rises as AI moves from task assistance towards workflows, functions and autonomous activity.
Culture sits underneath all of this. Governance culture affects accountability. Risk culture determines whether employees feel able to question AI outputs or stop a problematic deployment. Workforce culture shapes learning and adaptation. Ethical conduct determines whether the organisation's stated principles survive everyday pressures.
The objective should therefore not be maximum AI adoption. It should be a workforce capable of producing reliable outcomes with AI.
Play III — Technology & Data
As organisations move from using AI to creating and deploying it, technology decisions become system decisions.
This is where lifecycle thinking, including ISO/IEC 5338, becomes important. The CAIO does not need to become the organisation's chief engineer, but should understand how AI moves from inception and design through development, verification, deployment, operation, monitoring and eventual retirement.
The AI Factory provides a complementary business metaphor. Instead of viewing AI simply as a chat interface, the organisation begins to understand the machinery required to turn knowledge and data into useful intelligence products. The AI Factory approach similarly emphasises moving from passive use towards the creation, deployment and operation of organisational AI capability.
As AI moves towards agents, this becomes even more important because an AI system may no longer simply generate an answer. It may retrieve information, invoke tools, call other systems and execute parts of a workflow.
Play IV — Governance & Risk
Greater capability creates greater responsibility. Governance must therefore expand as the Change Impact Ladder and maturity level rise.
This Play brings together relevant regulatory requirements, voluntary frameworks, governance standards, accountability structures, risk management and assurance mechanisms. Depending on the organisation and use case, this may include the standards and frameworks you have identified—such as ISO-related governance guidance, APSR and other applicable industry or regulatory requirements.
An AI Governance Committee can provide cross-functional oversight, but governance cannot be delegated entirely to a committee. Business owners remain accountable for the outcomes of the AI they use.
The OCEG model is helpful here because governance, management and assurance have different functions, while its Lines of Accountability distinguish operational ownership from support, monitoring, independent assurance, executive oversight and board accountability.
AI governance must also follow the data. Drawing on ISO/IEC 38505’s governance-of-data principles, organisations should establish accountability for how data is sourced, accessed, transformed, used, shared, retained and disposed of. Governance should therefore extend across every stage of the AI lifecycle, not begin only when a model or application is deployed.
The central governance question becomes increasingly important with autonomous AI:
Who is accountable for what the AI does?
Play V — Economics & Controls
Finally, AI transformation has to make economic sense.
A cheap subscription can give the misleading impression that machine intelligence itself is cheap. At enterprise scale, AI consumes tokens, compute, infrastructure, integration effort, monitoring, security, expert review and management attention.
Agentic AI can amplify these costs because one visible user request may trigger planning, retrieval, repeated tool calls, reasoning, retries and additional model interactions. The appropriate unit of economic analysis is therefore increasingly the workflow rather than the prompt.
This Play consequently covers token economics, consumption and usage disciplines such as CUDs, model selection and routing, cost controls, vendor dependency, reliability and ROI.
The economics question should not be left until after deployment.
It belongs inside the design decision:
Does this combination of human and machine intelligence produce sufficient value to justify the resources, risks and controls required to operate it?
Economic discipline must be matched by operational control. As AI scales, organisations need to sustain ethical, privacy, security and responsible-use controls throughout operation—not merely at initial approval. Controls should be monitored, tested and adapted as models, data, users, workflows and threats change, so efficiency gains do not gradually weaken trust or compliance.
Bringing the Framework Together
The value of these ideas comes from using them together rather than turning each into another isolated framework.
For any significant AI initiative, a CAIO should be able to answer five connected questions:
1. Change: Where are we on the AI Change Impact Ladder—task, workflow, function, product/service, business model or organisation?
2. Maturity: Are we using, creating, deploying, governing or managing AI at the level required for this ambition?
3. Human–AI allocation: What remains human-led, what is augmented, what can be automated or agent-executed, and where must humans review or re-enter?
4. Transformation objective: Are we trying to improve existing work, differentiate what we offer, or genuinely transform how the organisation operates?
5. Execution: Are the Five Plays—Strategy, People & Culture, Technology & Data, Governance & Risk, and Economics & Control—strong enough to support the change?
Seen this way, AI transformation is not a destination and it is certainly not a technology rollout.
It is the deliberate movement of an organisation towards deeper forms of human–AI collaboration and organisational change, while progressively building the capability, governance and economic discipline needed to support them.
And that may be the most useful way to understand the CAIO's mandate: not to maximise the amount of AI in the organisation, but to help the organisation move to the right level of AI-enabled change - at the right maturity, with the right human involvement, and with enough control to create reliable and sustainable value.
