1. The Model at a Glance

Level	Name	What changes	Who decides	Governance demand
0	Digital Transformation Level (DTL)	Data is captured digitally at all; it is vague, isolated, incomplete.	Humans, on partial information.	Minimal — data doesn't yet exist to govern.
1	Automation Without Intelligence (AWOI) - an anti-pattern, not a milestone	A process is automated before the underlying data/logic is sound. Speed replaces judgement.	The system, on rules that ignore context.	Rising, but usually absent — this is where silent harm accumulates.
2	Automation With Intelligence (AWI)	The right data reaches the process; output becomes logical and context aware.	AI recommends, human still decides.	Moderate — explainability of recommendations.
3	Supervised Delegation	AI executes a bounded task; a human monitors and can override.	AI decides and acts; human supervises.	High — override mechanisms, audit trail.
4	Delegate to AI (DTAI)	AI agents pursue an open-ended goal across multiple steps, with minimal step-by-step instruction.	AI decides and acts toward a goal; humans set direction and boundaries.	Highest — accountability, PDPA/data exposure, liability allocation.

2. Levels, in Detail

Level 0 — Digital Transformation Level (DTL)

Data exists in fragments, if at all: paper records, disconnected spreadsheets, tribal knowledge held by individuals rather than systems. No amount of AI capability helps here — the prerequisite work is digitization and consolidation.

Level 1 — Automation Without Intelligence (AWOI)

Not a level to aim for. This is what happens when an organization automates a process using rules or sequencing that ignores the context needed to make the outcome sensible. Speed and consistency improve; judgement does not. This is where the bank queue and hospital queue examples below start.

Level 2 — Automation With Intelligence (AWI)

The right data now reaches the process, and the output reflects it. Critically, a human is still the decision-maker — AI provides the recommendation, ranking, or option set, and a person chooses. This is the level most industry AI adoption sits at today.

Level 3 — Supervised Delegation

AI both decides and executes within a bounded task, and a human monitors with the ability to override. This is the rung the original model skipped. It matters because most organizations — especially regulated ones — move here long before they are ready for open-ended delegation and treating it as a distinct stage gives learners a realistic, lower-risk entry point into agentic AI.

Level 4 — Delegate To AI (DTAI)

An AI agent, or group of agents, is given a goal rather than a task list, and pursues it with meaningful autonomy — planning, sequencing, and adapting its own steps. The human's role shifts from decision-maker to director: setting the goal, the guardrails, and the escalation conditions.

3. Worked Examples Across the Ladder

Both examples given in the original briefing illustrate the AWOI failure mode. Pairing each with what AWI, supervised delegation, and DTAI would look like in the same scenario makes the progression concrete rather than abstract.

Stage	Bank transaction queue	Hospital operation queue
AWOI (the broken version)	Single first-come-first-served line. A 3-minute balance enquiry waits behind a 40-minute account-opening case.	Waitlist ordered mainly by referral date, shaped by individual doctors' experience and judgement, with no consistent urgency scoring.
AWI (the fixed version)	Transaction type is captured at ticket issue; short transactions are routed to a fast lane; wait time becomes a function of task, not arrival order.	Cases are scored on clinical urgency, deterioration risk and resource load, then sequenced so the most time-critical cases surface — doctors still decide.
Supervised delegation	A system reassigns counters in real time as queue composition shifts through the day; staff can override the assignment at any point.	The system re-ranks the list nightly as new referrals and capacity data arrive; a clinical lead reviews and signs off before publication.
DTAI	An agent is given the goal 'minimize average wait time across all counters' and continuously reallocates staff and lanes to hit it, reporting exceptions.	An agent is given the goal 'maximize clinically safe throughput' and coordinates theatre scheduling, staffing and bed availability, escalating only edge cases.

4. The Governance Axis

Data and process maturity are only one axis of this model. As decision authority shifts from human to AI, the organization's need for trust infrastructure rises just as steeply: explainability at Level 2, override mechanisms and audit trails at Level 3, and full accountability and liability allocation at Level 4 — including, in Malaysia, PDPA exposure wherever agents handle personal data. A team can be data-ready for DTAI and still be nowhere near governance-ready for it.

5. A Familiar Analogy for Cross-Industry Audiences

Because this model asks learners to reason about degrees of autonomy, the Society for Automotive Engineering (SAE) levels of vehicle automation — a scale most audiences already have some intuition for — map onto it directly and can be used to introduce the concept before naming the AI-specific levels.

SAE driving-automation level	Maps to
Level 0–1 — no automation / driver assistance	DTL / AWOI — the vehicle (organization) is either unequipped or fitted with automation that doesn't yet understand context.
Level 2 — partial automation, driver supervises closely	AWI — AI recommends, human still holds the wheel and makes the call.
Level 3 — conditional automation, driver ready to intervene	Supervised Delegation — AI drives; human monitors and can take over.
Level 4–5 — high to full automation	DTAI — AI operates toward a goal within (4) or beyond (5) defined conditions, humans set the destination.

6. A Caution on Non-Linearity

In practice, no organization sits at a single level. A bank may be at DTAI in fraud detection, AWI in credit scoring, and still at DTL in HR records — simultaneously. The model is best taught as a lens applied per process or per decision, not as a single label for an entire organization; learners should be able to place a specific workflow on the ladder, not just their employer.

Dynamic Q

Monday, September 7, 2026