Securing Success with AI: A Guide for Australian Business Leaders

7 min. read

In our pursuit of higher operational productivity, AI is no longer a futuristic concept—it's a present-day imperative for Australian organisations. From exploring AI's implications to deploying full-scale systems, businesses are at various stages of their AI journey. But regardless of where your organisation stands, one thing is clear: crafting a deliberate AI roadmap is crucial for extracting real value.

AI Diligence is an effective process to sidestep the AI value detractors, avoid common pitfalls, and safely apply AI to transform your operating landscape.

AI is ready sooner than expected

AI can automate business activities earlier than previous estimates projected (McKinsey 2023). Foundational AI models from Anthropic, Google, Microsoft, Meta, Mistral, and OpenAI are as easily accessible as any cloud service or app. We use them on our mobile devices and desktops as personal aids. Business AI systems in production use them to infuse machine intelligence into business activities. It’s never been easier to have AI by our side.

AI for business is gaining traction

Executives are curious how AI can improve decision-making with data-driven reasoning. AI’s cognitive abilities to see, reason, read, categorise, and generate content is redefining the operating landscape. These are new ways of transforming productivity in the workplace.

In Australia, forty four percent of AI decision makers surveyed by the CSIRO said they have implemented AI, and twenty four percent are expanding or upgrading their AI implementation (CSIRO 2023). A global McKinsey survey (McKinsey 2024) finds that adoption of generative AI has surged since 2023, with companies deploying AI in more parts of the business.

AI high-performers (survey respondents where twenty percent of EBIT in 2022 was attributable to AI use – McKinsey 2023) are much more likely than others to use both traditional AI (statistical and machine learning models) and generative AI (large language models) in product development.

Unlocking value from AI for business

Value from AI for business that automate or augment tasks in operations can sometimes be elusive and more challenging than it appears. Amid the excitement surrounding generative AI, understanding the five factors for success with AI for business operations is crucial. This knowledge helps minimise the risk of premature applications that fail to deliver.

AI has performance thresholds

Businesses typically operate processes with an accepted margin of error, backed by procedures designed to catch and correct mistakes. Valuable AI systems must not only match but exceed these existing performance standards at a viable cost. If an AI system fails to deliver improvements at a reasonable cost, its adoption can stall, leading to diminished returns and waning support from stakeholders. To avoid this misstep, it’s crucial to understand the performance limits and associated costs of AI for your specific use case before committing to full-scale implementation. The objective isn't just to integrate AI; it's to ensure that AI drives measurable improvements in your operations.

AI embraces an evolutionary design process

The optimal design of an AI system often emerges through an iterative process of trial and error, making it challenging to predict its final form at the outset. Discovering late in development that AI fails to meet functionality or economic objectives can be costly and demoralising. Validate the feasibility of your AI project against predefined cost, performance, and functional benchmarks before committing substantial resources.

AI has a sweet spot in task automation

AI excels within a specific range of tasks, nestled between those requiring pure human expertise and those easily proceduralised. Some tasks are better automated by traditional application software at lower risk and cost. AI misapplied to tasks outside its optimal range can lead to inefficiency, increased costs, and unnecessary risks. An AI-task fit assessment can help identify the right opportunities.

AI model designs must balance cost and effectiveness

Designing AI systems often focuses on training foundational models, but this approach can quickly become cost prohibitive. The complexity of execution, combined with escalating costs for labour and infrastructure, can strain the business case, making it less viable. However, in many scenarios, it is more efficient and cost-effective to adapt existing models to meet specific goals. A working prototype can help confirm which of the two paths – adapting or training – is truly your best option.

Data need to be shaped for AI

Not all data is immediately ready for AI applications. AI models require a substantial amount of diverse, high-quality data to identify patterns, minimise bias, and develop reliable insights. However, the availability of suitable data and the teams needed to prepare it are often constrained. Testing a sample dataset with AI models early in the process can provide valuable insights and highlight any gaps. Ultimately, anticipate and plan for the process of data preparation.

The importance of AI Diligence

AI Diligence is crucial before embarking on solution development. This step is essential to ensure that the foundational AI models align with the specific business tasks at hand. By doing so, you safeguard your investment while gaining vital insights that steer your AI journey, ensuring it remains in sync with your business objectives from the outset.

AI Diligence is more than just a check; it’s a critical process for guaranteeing the value of your AI initiatives. This process is designed to mitigate value detractors and to validate the fit between AI and your chosen business tasks.

Suppose you’ve already identified potential tasks for AI implementation, and preliminary checks confirm that the necessary data is available or can be reasonably sourced.

Here’s how to proceed:

• Establish baseline benchmarks for the current task.

• Set the performance benchmark that the AI system must achieve.

• Prepare the learning datasets.

• Build a prototype using one or more foundational models, or a combination thereof.

• Iteratively adapt, engineer and refine the AI models on your dataset until they consistently meet the required performance.

• Cease iterations when the system hits or exceeds its performance targets or when it becomes clear that the desired outcomes are unattainable.

At this stage, you will have gathered comprehensive information with metrics on the feasibility and benefits of your AI project. You’ll understand the AI system’s behaviour, expected benefits, error rates, biases, decision-making logic, and cost drivers. This data allows you to build a robust business case, and if it holds up, you can proceed confidently to implementation.

However, if the AI solution isn’t viable, you can clearly explain the reasons, extract valuable lessons for future initiatives, and control expenditure.

The Diligence process ensures that the AI system is well-suited to the task, feasible to implement, can perform effectively, is maintainable and financially sustainable. It goes to the very heart of the approach we bring – AI systems that are practical, aligned and robust.

Executive actions for AI success

Adopt a practical approach

AI is ready for business integration, but the landscape is constantly evolving. With new models and cost structures emerging regularly, staying informed is crucial. Be prepared to adapt your AI strategy as it gains momentum. Run several Diligence initiatives, promoting forward the most promising use cases. Start small, and as these use cases prove effective in production, scale up accordingly.

Use Diligence to safeguard your AI investment

The Diligence step is essential for AI value assurance and risk mitigation. Put a few use cases through Diligence and promote the practical and promising ones. By putting selected use cases through this process, you can assess feasibility, understand risks and costs, and proceed with confidence. This approach builds and maintains stakeholder trust, ensuring your AI initiatives are grounded.

Prioritise budgeting for data quality and curation

AI systems need diverse, high-quality data, which is crucial for effective learning and bias reduction. Use real datasets during Diligence to test AI performance. Plan and budget for curating and maintaining the data supply chain to support long-term AI success.

Align AI for success

Establish clear AI policies, ethics and design principles with which the systems align. Appoint AI owners responsible for outcomes. Adjust job roles in line with AI adoption and upskill your employees. Foster an AI-positive culture supporting your staff through the transition. Align your AI infrastructure investments with a roadmap informed by Diligence and when deployed, continuously monitor the system for model degradation and emerging concerns.

Embrace continuous learning

View each AI initiative as a learning opportunity. Diligence is a great way to do so. Whether use case is deployed or not, the insights gained should inform future AI strategies and implementations. Over time, this continuous learning approach will enhance your organisation’s AI maturity.

Measure and communicate effectively

Implement metrics and benchmarks from the start of your AI initiatives. Continuously measure performance against these benchmarks, adjusting your approach along the way. Diligence produces excellent insights to inform your decision-making and regularly keeping your stakeholders engaged on progress and results. Continue communicating as you deploy and operate your AI use cases.

How we can help

Irada, in partnership with Deakin University’s Applied AI Institute, assists growth-minded organisations plan, choose and deploy practical, aligned, and robust AI systems for business.

References, Resources, Readings

1.     McKinsey. The state of AI in 2023: Generative AI’s breakout year. August 2023.

2.     McKinsey. Generative AI and the Future of work in America. July 2023.

3.     McKinsey, The State of AI in early 2024. May 2024.

4.     CSIRO National Artificial Intelligence Centre. Australia’s AI ecosystem momentum. March 2023.

The author thanks Professor Rajesh Vasa, Head of Translation Research at Deakin University’s Applied AI Institute, for his inputs to this article.

Links to external websites were correct at the time of publishing. Articles may be behind a paywall. Irada is not responsible for the content of external websites.

The information in this article is general in nature. Your circumstances may vary.

This work is licensed under CC BY-NC-SA 4.0