
Sign up to save your podcasts
Or


The pilot can make enterprise AI look further along than it really is.
Once systems hit live workflows, organisations start dealing with messier user behaviour, weaker ownership, and economics that no longer match the original case.
In a conversation with Anthony Saba, Partner & Managing Director, Transformation Services at ADAPT, Vijayan Seenisamy, Transformation Lead at an ASX 30 enterprise, argues that these failures are rarely about the model itself.
They come from trying to scale AI inside operating structures that were built for a different kind of technology.
Key takeaways:
Most AI programs are framed too narrowly
Vijayan argues that many organisations get the framing wrong from the start.
Once AI is labelled a technology transformation, responsibility narrows too quickly to the CIO or CTO. That misses the bigger issue.
In his view, AI is a business transformation that affects workflows, leadership, workforce design, and operating model choices far more than most executive teams are prepared to admit.
Technology may be essential, but it is only one part of the change.
That is also why so many organisations are seeing a gap between executive expectation and financial outcome.
Teams may have bought copilots, pushed adoption, and reported productivity gains, but the CFO still sees little movement in the P&L.
The local wins are real, but they do not become enterprise value unless the business changes how work is structured around them.
Production is where the business case starts to unravel
Vijayan describes a familiar pattern.
The pilot is built in controlled conditions, with curated data, known prompts, and the builders close enough to keep things working.
Production removes those protections.
Real users behave differently, data gets messier, and systems start to wobble under conditions the demo never had to survive.
He also points to unit economics as a major blind spot.
If an agent costs more to complete a task than the human process it was supposed to replace, the business case weakens fast.
That gets worse in multi agent systems, where orchestration overhead can erode the performance that individual agents appeared to have in isolation.
What looked efficient in a pilot can become expensive and unstable once usage grows.
AI needs a different economic and ownership model
A core argument in the conversation is that AI should be measured as a business asset, not as another software rollout.
Vijayan says the most important metric is cost per successful outcome, meaning the cost of producing an outcome a user or stakeholder would actually accept.
Token costs, task costs, and adoption rates are useful, but they do not tell leaders whether value is really being created.
He also argues that ownership has to widen.
Most AI initiatives still sit under a one signature model, usually with the CIO holding the budget and delivery burden.
His alternative is a three signature model: the CIO for delivery feasibility, the CFO for unit economics, and a business leader for workflow change.
That spreads accountability across the people who control whether AI can work in practice, not just in demo conditions.
Boards need enough cyber detail to judge exposure, understand whether key controls are working, and decide where intervention or investment is required.
Cyber lands better when it is framed around governance, business risk, and investment choices rather than technical complexity.
Jamie Rossato, CISO at Orica Australia, shares a practical view on how security leaders build alignment with boards and executives by connecting cyber outcomes to risk appetite, operational priorities, and the cost of leaving gaps unresolved.
Key takeaways:
Board engagement demands clarity, not complexity
Communicating with boards requires direct, concise messaging focused on controls, risks and outcomes, not technical detail.
Boards operate under time pressure and broad accountability, so cyber leaders must translate security into governance, risk and fiduciary impact.
Jamie emphasises avoiding overly dense materials, instead focusing on whether controls are effective, aligned to risk appetite, and what actions are in place to address gaps.
Executive alignment is where investment is won
Cyber funding conversations are secured with executive management before reaching the board.
Executives own implementation and operational risk, making them critical partners in shaping, supporting and advocating for cyber investment.
Jamie notes that by demonstrating control effectiveness and value for money, leaders guide executives to recognise unaddressed risks, naturally building the case for further investment.
A threat-led strategy outperforms compliance-first approaches
Focusing on real threats ensures security measures are meaningful, with compliance emerging as a by-product rather than the objective.
Compliance alone can drive checkbox behaviour, whereas threat-led strategies prioritise actual risk reduction and resilience.
Jamie explains that understanding internal and external threats allows organisations to meet regulatory requirements organically, while also making a stronger case for proactive investment, even when threats have not yet materialised.
Cyber leaders create influence not by amplifying risk, but by demonstrating control, value and foresight in a language the business already understand.
Customer experience, data strategy, and resilience do not operate the same way across every business model.
At Village Roadshow, the challenge is sharper because cinemas, theme parks, and film distribution each run on different customer expectations, planning cycles, and technology needs.
Keyur Lavingia, Head of Cyber Security at Village Roadshow, outlined how the organisation is trying to tailor experience and innovation to each part of the business without creating fragmented governance or inconsistent trust.
Key takeaways:
Different business models create different resilience demands
Village Roadshow operates across businesses with very different customer journeys.
Cinema attendance is often spontaneous, while theme park visits are planned further in advance, with families saving and organising around a bigger commitment.
Film distribution brings another operating model again, with its own systems, partners, and timing pressures.
That changes how transformation has to be approached.
A single digital model cannot be applied evenly across the group.
The task is to adapt data, engagement, and operational decisions to each business without letting governance drift or the customer experience become disjointed.
Personalisation gets more valuable as data responsibility rises
With around 13 million customers each year and roughly 90% of interactions happening digitally, Village Roadshow relies heavily on data to shape customer experience.
That creates obvious opportunities to personalise moments more effectively, whether through loyalty recognition, targeted rewards, or more relevant engagement.
It also raises the stakes on trust.
The more customer data an organisation uses to improve experience, the more carefully it has to handle privacy, consent, and protection.
Trust becomes part of the experience customers have with the brand.
AI adoption works better when guardrails are clear and usable
Village Roadshow’s approach to AI has focused on making adoption easier rather than trying to control every use case upfront.
The view is that AI only creates value when people actually use it, but that wider use still needs boundaries people can understand and follow.
Keyur described an internal AI assessment model built around a traffic light system, green for approved use, amber for limited use, and red for blocked use.
Designed with input from legal, business, and technology teams, the framework gives employees a simple way to understand what is acceptable without forcing them through heavy process each time.
The goal is to let people use AI without exposing sensitive data or creating unnecessary risk.
As AI moves from specialist teams into everyday business use, data leaders have to focus less on scarcity and more on safe, scalable adoption across the organisation.
Artak Amirbekyan, Chief Data Officer at EBOS Group, argues that this shift should not be resisted. Wider use is where the value comes from.
The real task is making that use safe, deliberate, and broad enough that AI becomes part of everyday capability rather than a specialist function sitting off to the side.
Key takeaways:
Cheaper AI changes who gets to create value
AI becomes more powerful inside organisations when access stops being limited to specialists.
As costs fall and tools become easier to use, more people can experiment, solve problems, and create value directly in their own roles.
That is how Artak explains the current moment.
He says large language models existed before tools like ChatGPT, but they were expensive, specialist, and largely inaccessible.
Once the technology became cheaper and easier to use, adoption surged. For him, that is the pattern leaders should expect.
When AI gets cheaper, more people use it, and that wider use is where much of the return starts to appear.
Guardrails make wider AI use possible
Democratised AI only works inside a business when the organisation is clear about where risk sits and what cannot be compromised.
The answer is not to lock everything down. It is to put enough guardrails in place that more people can use the technology safely.
Artak says companies need practical controls around what data can be shared, what must stay inside the organisation, and which protections need to sit around the models being used.
He talks about restrictions on data leaving the business, firewalls, and protections such as indemnity requirements, but the broader point is more important.
Guardrails should increase safe usage, not reduce it. In his view, organisations get more value from AI when they enable more people to use it within clear boundaries.
AI fluency is becoming everyone’s responsibility
AI capability is moving out of the specialist domain and into general business capability.
That changes both workforce expectations and the role of technical experts.
Artak argues that AI is now everyone’s domain because the people with AI skills are starting to outpace those without them.
He also notes that this changes the shape of data and AI roles themselves.
Where companies once searched for rare technical specialists who could both build models and speak to the business, they are now increasingly enabling business users to work with AI directly.
In that environment, data scientists and machine learning engineers spend more time enabling use than building everything themselves.
Learning how to work with AI is becoming part of staying effective at work, not a side topic for a small technical group.
The interesting shift is not simply that AI has become mainstream.
It is that AI is becoming cheap enough, broad enough, and useful enough that companies now need to govern widespread use, not specialist scarcity.
Enterprise AI programmes stall when they are treated as technical capability rather than organisational responsibility.
AI adoption is a governance and ownership challenge.
In conversation with ADAPT’s Head of Analytics & Insights, Gabby Fredkin, David Scott argues that lasting impact comes from aligning AI to strategic outcomes, business accountability and existing performance measures.
Key takeaways:
Strategy works best when it follows the institution’s priorities
Large organisations rarely run on a single operating reality.
Different parts of the business have different objectives, stakeholders, and constraints, which means AI strategy has to hold together without flattening those differences.
That is how David describes the University of Sydney’s approach.
The university’s work is anchored to its broader Sydney in 2032 strategy, while data, analytics, and AI initiatives are directed towards specific institutional priorities such as student experience, research, and operational effectiveness.
He is clear that this does not produce one standalone AI strategy.
It produces a set of aligned strategies tied to the areas where the institution most needs improvement.
Assurance gives people the confidence to use AI well
AI governance only becomes useful when it helps people believe the technology is being applied in ways that are responsible, credible, and fit for the setting they are working in.
David points to the university’s investment in AI readiness and its AI assurance framework as an example of that.
The framework draws on learnings from other organisations, then adapts them to the university’s own complexity.
He also highlights Cogniti, a classroom tool that lets academic staff shape how agents engage with course content, as an example of where controls are designed into the use case itself.
That matters because the scrutiny around AI governance is rising quickly, and confidence depends on visible assurance rather than generic policy statements.
Ownership and value both need to stay close to the work
Enterprise AI becomes harder to sustain when ownership is separated from the process it is meant to improve.
Local context shapes how agents are configured, how people use them, and whether outcomes hold up across different parts of the organisation.
David is direct on that point.
He says ownership should sit with the business process owner because they understand the local context and carry responsibility for the result.
Data and technology teams still matter, but their role is to enable, support, and help industrialise what works.
He also grounds value in the university’s existing metrics.
The two he points to most clearly are student satisfaction and student success, including early intervention that has helped reduce student fail rates.
That makes AI easier to evaluate because it is being measured through outcomes the organisation already recognises.
AI strategy stays aligned to institutional priorities, assurance is built to create confidence, and ownership sits with the people closest to the outcome.
In a complex organisation, that is what makes AI more likely to hold once it moves beyond early use cases.
Most organisations approach AI through an efficiency lens and are surprised when value fails to scale.
According to Dr Jon Whittle, ADAPT Advisor & former Managing Director of Data61 at CSIRO, AI must be a purpose led transformation.
He argues that sustainable impact comes from aligning AI to organisational intent, leadership capability, and human outcomes rather than narrowly defined productivity gains.
Key takeaways:
Productivity is a weak primary driver for AI
Task efficiency is easy to see.
Organisational value is harder to earn. AI can save time at the individual level while failing to lift performance across the business, because that time is often absorbed by more meetings, more tasks, or higher expectations.
That is why productivity on its own is a weak organising principle for AI strategy.
Jon makes this point directly, arguing that the evidence so far shows personal productivity gains do not necessarily aggregate to the organisational level, and can even contribute to burnout when leaders respond by pushing more work into the space AI creates.
Purpose-led AI unlocks different use cases and outcomes
The quality of AI use cases depends on the quality of the question leaders ask at the start.
Organisations that begin with speed and savings tend to optimise existing activity.
Organisations that begin with purpose tend to identify barriers that are stopping them from delivering what they exist to do.
Jon argues that this shift has to come from the CEO and board. His dental practice example shows the difference clearly.
An efficiency lens focuses on getting more patients through the chair. A purpose lens focuses on reducing fear, increasing early intervention, and improving health outcomes.
As he puts it, the use cases become “quite different” once purpose comes first.
AI adoption is a leadership and culture challenge
Most organisations do not stall on AI because the models are weak.
They stall because the organisation is not aligned strongly enough to absorb the change.
Governance, stakeholder management, executive backing, and change management determine whether AI moves beyond experimentation or gets trapped in committees and theatre.
Jon is explicit that adopting AI is “fundamentally not a technical problem” and instead goes back to culture and change management.
He also warns that committees without a clear sense of purpose and top down buy in will just keep spinning the wheels, and that leaders in charge of AI need fluency across technology, governance and ethics, and business rather than depth in only one of those areas.
Pilot mode persists when three groups lose confidence at the same time: the CFO cannot justify the return, the chief risk officer cannot map the risk, and the board cannot govern what it does not understand.
Simon Kriss uses that breakdown to explain why AI scale in Australia is being held back less by tooling than by trust, governance maturity, and cost discipline.
Key takeaways:
Trust collapses when AI is forced into old business models
Most organisations do not stall because the technology is missing.
They stall because finance, risk, and governance are still trying to judge AI through models built for more predictable systems.
Simon says corporate Australia has been “very, very, very slow to adopt” because “we simply don’t” trust AI, even though “we know the tech” and “we get it.”
He then points to two practical blockers that show up when organisations try to move from proof of concept to production: the CFO cannot see where the ROI comes from, and the chief risk officer struggles to apply the same risk logic to very different use cases, like summarising phone calls versus helping synthesise a human drug.
His conclusion is direct: organisations “can’t just shoehorn AI into an existing business model” because the business model, the risk model, and the governance model all need to change with the technology.
AI economics replace cloud economics as the next leadership shock
The cost problem changes shape once AI moves beyond contained pilots and into agent driven work.
Simon separates simple exploitation use cases, where the ROI calculation looks like any other business case, from exploratory and agentic use cases, where the cost profile becomes much harder to predict.
He says the issue intensifies with agents because they do not complete a task once, they iterate, backtrack, and retry, sometimes “20 or 30 times,” which can blow out token consumption rapidly.
In his example, something that started at “$28 per month for five people in proof of concept” can turn into a “half $1 million bill” once lag billing catches up.
He is also clear that the market is still “figuring it out,” so leaders cannot assume AI cost controls are already mature in the same way cloud FinOps became.
Sovereignty shifts from risk control to economic advantage
Sovereignty becomes more valuable when leaders stop treating it as a narrow hosting question and start treating it as a control and value capture question across inference, models, and cultural fit.
Simon says sovereignty is “about protectionism more than isolationism” and argues that organisations still mostly see it as a way of controlling risk, even though it is starting to become “an economic enabler.”
He points to dependence on offshore inference, the US Cloud Act, limited control over how foreign foundation models are built, and the fact that token spend continues to flow offshore.
He also gives a more operational argument for local relevance: a chatbot that better understands Australian language and nuance will use fewer tokens and create better customer loyalty because it reaches understanding faster.
In his view, that makes selective sovereignty a practical commercial decision as much as a governance one.
Rolling out AI access across a large organisation is the easy part. Building the judgement, safeguards, and leadership momentum needed to use it well at scale is far harder.
That is the challenge CommBank faced.
With 50,000 employees and responsible AI established as a long-term strategic priority, the bank offers a useful example of what it takes to move AI beyond isolated capability and into enterprise-wide execution and impact.
Jen French explains how CommBank is approaching that shift through leader led adoption, responsible AI frameworks, practical skill building, and reviewing adoption data to understand where confidence is genuinely taking hold.
Key takeaways:
Scaling AI through people, not just platforms
CommBank's AI journey began nearly a decade ago with machine‑learning‑driven customer engagement, initially built by specialist technical teams.
What quickly became clear, however, was that value only scaled when the broader workforce understood and used AI.
Rather than limiting AI to technical experts, CommBank focused on equipping employees with AI assistants and foundational education: what AI is, how it’s already used across the bank, and how individuals can apply it safely in their day‑to‑day work.
This democratisation of AI capability required a mindset shift, not just new tools.
Guardrails that enable, not restrict
Empowering tens of thousands of employees naturally raises risk, but CommBank addressed this by embedding responsible AI principles directly into how work gets done.
A group‑wide Responsible AI Framework, supported by clear policies, standards, and toolkits, ensures AI use is ethical, secure, and compliant with regulation.
Importantly, governance is not treated as bureaucracy: legal, risk, compliance, HR and technology teams are brought together early to shape AI solutions before they are deployed, whether for customer experiences or internal productivity.
Leadership, momentum and measuring value
Leadership visibility has been critical, with senior executives actively role‑modelling AI use to normalise adoption and build trust.
Usage data helps to understand how AI tools are being used and where confidence is growing, while a strong learning culture allows teams to iterate on what works and adjust what doesn’t.
Rather than relying on one‑off business cases, CommBank sustains momentum through continuous learning, shared ownership, and embedding AI into everyday operating rhythms.
This is a #paidpartnership with Commonwealth Bank.
AI pressure is hitting universities differently from most organisations.
It is being driven from the ground up by students, academics, and researchers already testing where AI helps and where it starts to distort learning.
Kerry Holling, Interim CIO at the University of Sydney, explains how that pressure is changing governance, teaching, and trust across the institution.
Key takeaways:
Universities need guardrails that people will actually use
AI governance in universities has to work in the real world. If the controls are too rigid, staff and students will route around them.
If they are too loose, privacy, data sovereignty, and research integrity are exposed.
That is why the University of Sydney has focused on practical guardrails developed jointly across IT and Legal, with self assessment tools that help staff judge use cases without turning governance into a bottleneck.
Kerry’s point is that balance matters more in a university environment because academic work depends on openness and experimentation, while the institution still has to protect sensitive data, intellectual property, and research quality.
The goal is to create enough structure to support safe use without shutting down the value AI can bring to research, teaching, and operations.
The bigger teaching challenge is no longer access to AI, it is assessment
The hard question for universities is no longer whether students will use AI.
The real issue is whether assessment still measures understanding, judgement, and learning in an environment where AI can generate convincing outputs quickly.
That is where Kerry sees the pressure building.
He argues that AI can improve learning when it helps students deepen their understanding, but weak assessment design will invite shortcuts instead.
The stronger institutional response is to rethink how knowledge is tested so students still have to demonstrate real comprehension.
He also points to examples where AI is improving access to teaching rather than undermining it.
One University of Sydney academic built Cogniti to replicate parts of one to one support at scale, and Kerry says it is now used by more than 5,000 academics to help develop curriculum material and provide more personalised tuition to students.
For him, that is what useful AI in education looks like, expanding learning support in places where human access is naturally limited.
Trust grows when AI is used to augment people, not displace judgement
Universities will get more value from AI when they treat it as a tool for augmentation, not as a substitute for human thinking, academic rigour, or institutional accountability.
Kerry is optimistic about AI’s potential, but he is careful about how far that optimism should go.
He supports AI for personal productivity and sees genuine value in tools that accelerate research, improve learning outcomes, and reduce friction in university work.
At the same time, he is wary of over-dependence, concerned about the concentration of power in large technology companies, and clear that institutions should be selective about how they deploy AI.
He describes it as something that should be used with respect, not submission, and that framing matters.
The trust challenge in higher education is not only about policy.
It is also about making sure AI strengthens human capability, protects academic judgement, and earns confidence as adoption becomes more embedded.
What does safe scale look like when the cost of getting AI wrong is measured in public trust?
In this ADAPT Insider podcast episode, Daniela Polit, Public Sector Transformation Executive, outlines a clear test for government AI. It should help solve complex public problems, reduce friction for citizens, and improve services at scale, while operating within guardrails that protect sovereignty, accountability, and trust.
Key takeaways:
Public value has to come before the technology
Strong AI strategies begin with the problem being solved.
In government, that means asking whether a tool can genuinely improve a service, shorten wait times, reduce bureaucracy, or make support easier to access for citizens.
That is the lens Daniela applies throughout the conversation.
She describes AI as a way to solve complex public sector problems, especially where service delivery involves scale, complexity, and large volumes of information.
The value, in her view, comes from helping people deal with government faster and with less friction, whether that means reducing unnecessary touchpoints, improving transparency, or tailoring services more effectively across very different citizen needs.
If AI can clearly improve the outcome, it has a case. If the likely value is marginal and the risk is higher, it should not be forced in.
Trust holds when sovereignty and accountability are protected
Public sector AI needs trust built into where models run, how data is handled, and who remains responsible for the outcome.
Daniela makes that standard explicit.
She says government models are hosted in closed internal environments, with the same rules and authorisations that already apply to public data carried through into AI use.
She is equally clear that accountability stays with a person.
A tool can support a decision, accelerate a process, or structure information more effectively, but it cannot replace the accountable decision maker.
That combination, sovereignty, transparency, and human oversight, is what allows AI to be used in sensitive environments while preserving public confidence.
Safer scaling depends on risk based governance and training
AI becomes easier to scale when governance gives teams a clear way to assess value, manage risk, and move suitable use cases forward with confidence.
That is how Daniela describes the public sector approach.
She points to frameworks and assurance checks that run from ideation through implementation, testing, and evaluation, with policies evolving as use cases become more complex.
She also makes clear that training should reflect the level of public impact.
More structured education and tighter oversight are used for public facing or higher risk applications, while lower risk internal tools can be handled more flexibly.
Even in a large and federated system where collaboration across agencies is still often informal, that discipline creates a stronger filter around value, accountability, and safe deployment.
From the publisher's feed