Agentic AI needs workflow redesign before it can create value

Agentic systems can automate flawed processes as easily as sound ones.

Workflow design, data quality, and clear risk boundaries determine whether that autonomy improves performance or compounds existing problems.

At CIO Edge in Melbourne, Lynden Roberts, CMIAIO at Monash Health; Simon Moorfield, Group Executive, Customer and Technology at Transurban; and Mansi Hasabnis, Director, Digital Technology Solutions at Swinburne University of Technology, joined Peter Hind, Principal Research Analyst at ADAPT, to examine how organisations can redesign work, strengthen data foundations, and govern experimentation as agents take on more execution.

Key takeaways:

• Start with the value pool, then redesign the workflow: Select a clear business problem and rebuild the process around the outcome before deciding where agents should operate.
• Tie data quality to operational and commercial outcomes: Show how poor data affects decisions, customer experience, risk, retention, or revenue to strengthen the case for investment.
• Create clear boundaries for experimentation: Give teams ownership, frontline input, and permission to test while setting guardrails around risk, accountability, and autonomous action.

Start with the value pool before redesigning the workflow

Technology should enter the conversation after the business problem and desired outcome are clear.

At Monash Health, Lynden’s team targets specific value pools such as inefficient inpatient workflows, then co-develops solutions using AI and ambient technologies.

That keeps investment anchored to a measurable operational problem rather than a tool looking for a use case.

Workflow design then determines how much value the technology can create.

Simon expects the structure of work to change as models process more context and take on more complex execution.

Existing roles, hand-offs, and decision points need to change with that capability.

Mansi links poor process design directly to risk. Automating a flawed workflow allows agents to execute the same weaknesses faster and across greater volume.

Start with a small number of high-value problems, map how work and decisions move today, then redesign the workflow around the outcome required.

Data quality needs to be tied to operational consequences

Agents raise the cost of poor data because they can act on inaccurate information before a person reviews it.

Much of the work behind a seamless experience at Monash Health sits in getting data into the right shape. Lynden’s experience shows why model performance alone cannot compensate for weak foundations.

Fragmented technology estates add another layer of complexity.

Simon points to data spread across on-premise systems, cloud platforms, and SaaS applications, which makes consistent context harder to provide across workflows.

Boards need the consequences expressed in business terms.

Mansi frames data quality through operational risk and opportunity: an inaccurate report creates one level of exposure, while an autonomous agent acting on the same information can create a far greater one.

Linking data investment to outcomes such as student experience, retention, revenue, and operational risk gives leaders a clearer basis for prioritisation.

Semantic layers can also provide agents with more consistent business meaning across fragmented sources.

Guardrails should make experimentation easier to govern

Teams need room to test new ways of working and clear boundaries around autonomy, risk, and accountability.

Simon builds organisational understanding through practical use cases, then shares the results with executives and the board.

Visible evidence gives senior leaders a stronger basis for deciding where further investment and autonomy are warranted.

Frontline involvement also improves the quality of experimentation.

Lynden uses frontline champions and co-development to bring operational knowledge into solution design rather than separating technology development from the people doing the work.

At Swinburne, Mansi uses product-centric teams with clear ownership of outcomes and a safe-to-fail environment.

Employees are encouraged to surface edge cases and friction, giving the organisation more information about where controls or workflows need to improve.

Disciplined experimentation combines clear ownership, practical testing, frontline knowledge, and defined risk appetite.

Safe experimentation gives leaders the evidence to expand autonomy without losing control.