Start small, win big: How early adopters approach AI in healthcare.

Ray Delany
2 days ago
5 min read

The narrative surrounding Artificial Intelligence often feels polarised: a dizzying "back-and-forth tennis rally between hype and disillusionment". On one side, we hear about major tech companies pouring hundreds of billions of dollars into sprawling data centres and chasing "super-intelligence". On the other, we see reports of underwhelming new products and organisations spending $US30-40 billion on generative AI with little to no return.

For most organisations, especially mid-sized health services and non-profits, this high-stakes game is irrelevant.

The biggest mistake leaders can make today is believing that AI success requires competing on scale or conquering the universe. The truth, proven by early adopters across Australia and Aotearoa New Zealand, is that winning with AI doesn't require massive transformation; it requires a smart, tightly scoped approach.

The most successful organisations are moving past obsession with technical power and concentrating on achievable, high-impact utility.

You don't need to go big to get results

While big tech continues to funnel billons into proprietary models, the technical progress in established large language models (LLMs) has begun to slow. GPT-5, for instance, did not make the same impression as its predecessors, despite costing ten times more to train. Attention is shifting away from inventing super-intelligence and toward less exciting, but deeply valuable, practical applications, such as organisational workflows and coding software.

This shift is fantastic news for ordinary organisations, as it means:

Diminishing returns on scale: Open-source AI models are rapidly closing the performance gap with proprietary rivals on basic tasks like drafting, transcribing, and summarising. A large language model with 700 billion to one trillion parameters may be "more than enough" for many applications.
Proven financial prudence: Early AI implementations are already delivering verified returns on investment (ROI). Documented evidence shows that for every $1 invested, a return of $3.20 can materialise within 14 months. Furthermore, individual AI solutions costing less than $1 million annually are preventing malpractice cases worth up to $1 million each.

The momentum in this region is driven by practicality: 82% of SMES in New Zealand are already experimenting with AI (significantly above the global average), and 45% of healthcare SMEs in Australia are using it, all driven by real operational needs, not just curiosity.

The patterns of successful early adopters

The organisations leading the way with AI in healthcare aren't the biggest, flashiest, or best funded. They're the ones taking a pragmatic, people-first approach. Across New Zealand and Australia, early adopters are proving that success with AI comes from starting small, focusing tightly, and learning fast.

Focus on one pain point - Rather than attempting a sweeping transformation, early adopters begin with a single, high-friction workflow - often one that staff complain about most. Automating appointment scheduling, reducing documentation load with AI scribes, or using chatbots for after-hours queries are examples of small, contained pilots that deliver quick wins and visible value.
Involve your team from the start - Clinicians and operational staff are key to AI success. When frontline teams help select and test tools, adoption rates rise and the solutions work better. Early pilots that included nurses, administrators, and clinicians in design and evaluation saw faster uptake and fewer issues.
Choose low-risk, high-impact use cases - Early benefits are emerging from non0clinical and assistive AI - the systems that streamline operations rather than make life-and-death decisions. Automating stock ordering, rostering, or patient communications saves time, boosts efficiency, and builds organisational trust before expanding into more complex clinical applications.
Keep humans-in-the-loop - The most effective AI tools don't replace people - they extend them. Whether it's a digital assistant supporting patient engagement or a radiologist verifying AI-generated reports, human oversight remains central. AI should stretch your workforce, not substitute it.
Measure outcomes, not algorithms - The standout adopters judge success by real-world improvements: shorter wait times, few missed appointments, earlier diagnoses. Features are secondary to results.

Why "start small" works so well

Starting small by focusing on administrative and operational challenges works because it creates clear, measurable wins while building internal expertise and trust.

It mitigates data risk: AI systems are only as good as the data they are trained on. Focusing on limited, well-understood data sets for administrative tasks reduces the risk of bias or inaccurate outcomes compared to complex clinical diagnostics.
It addresses equity from the ground up: If the training data is incomplete or biased (e.g. under-representing certain groups), the AI tool may perpetuate or reinforce existing disparities. In the Aotearoa New Zealand context, governance frameworks must specifically prioritise Māori perspectives and uphold Māori data sovereignty (Māori data is a taonga, or treasure) to address long-standing health inequities. Starting small allows organisations to embed equity evaluation early in the process.
It aligns effort with strategy: Instead of facing "decision paralysis" around massive, risky systems, focusing on Stage One (Administrative Efficiency) provides immediate ROI and establishes infrastructure before moving to Stage Two (Clinical Enhancement).

How to take your first step

If you are ready to move beyond the hype and begin realising tangible value, start with a systematic, safety-first implementation framework. Healthcare transformation experts suggest a four-step framework: problem identification, workforce empowerment, responsible testing, and scaled deployment.

Before you deploy any AI solution, the leadership team must align on five essential questions:

What is the problem we're solving? AI is a tool, not a strategy. The goal is what matters.
Do we have enough data? AI relies on data quality. Is your data clean, digitised, and representative of the population you serve?
Who will monitor and review AI decisions? Clinician oversight is critical, and protocols must be in place to handle errors or ambiguous recommendations.
What are our ethical guardrails? You must have a framework for responsible AI, addressing potential bias and demanding transparency from vendors.
How will we evaluate success? Define clear metrics, such as a reduction in call centre load, higher staff efficiency, or earlier diagnoses.

Begin where the need is real

AI adoption is no longer a question of "if", but "how well". The competitive advantage window for early, thoughtful implementation is narrowing.

The evidence from successful organisations shows that AI does not demand a sudden expensive leap into the unknown. It demands focus: finding the single, high-friction workflow, like clinical documentation or scheduling, and deploying a measured solution to support the people doing that work.