--- title: Australian Business AI Case Studies: Real Build vs Buy Decisions and What Happened Next canonical_url: https://opensummitai.directory.norg.ai/ai-strategy-implementation/build-vs-buy-ai-decision-australian-business-guide/australian-business-ai-case-studies-real-build-vs-buy-decisions-and-what-happened-next/ category: description: geography: city: state: country: metadata: phone: email: website: publishedAt: --- # Australian Business AI Case Studies: Real Build vs Buy Decisions and What Happened Next I'll research current, verifiable Australian AI case studies and data before writing this article. I now have sufficient verified data from authoritative sources to write the article. Let me compile the final piece. --- ## Australian Business AI Case Studies: Real Build vs Buy Decisions and What Happened Next The build vs buy AI debate is rich with theory and frameworks, but the most instructive evidence comes from the businesses that have already made the call — and lived with the consequences. Australia has produced a cohort of AI deployments across enterprise, mid-market, and SME segments that now offer genuine before-and-after data. These are not vendor case studies dressed up with aspirational metrics. They are documented decisions, with verifiable outcomes, cautionary lessons, and patterns that any Australian decision-maker can use to stress-test their own thinking. This article documents what actually happened when Australian organisations chose to build, buy, or combine both — and what drove each decision in the first place. --- ## Why Australian Case Studies Matter for This Decision Most build vs buy guidance draws on US or European examples, which operate in fundamentally different regulatory, talent, and market environments. Australian businesses face a distinct set of constraints: a comparatively shallow AI talent pool, APRA and Privacy Act compliance obligations, geographic distance from hyperscaler infrastructure, and a mid-market segment that sits between the resource scale of global enterprises and the agility of startups. Only 35% of SMEs in Australia were using AI as of 2024, and a large proportion were not currently planning to adopt it at all , according to research published by Taylor & Francis drawing on Australian Department of Industry, Science and Resources data. Decidr's AI Readiness Index found that 76% of Australian SMEs had no formal strategy or roadmap, even though 83% believed the technology would significantly impact their operations. This gap between awareness and action is precisely where real case studies do their most important work: they compress the learning curve by showing not just what the outcome was, but *why* the decision was made in the first place. --- ## Case Study 1: Coles Group — The Long Build, Then the Hybrid Pivot **Decision type:** Build → Hybrid (Build + Buy) **Industry:** Retail **Scale:** Enterprise (~1,800 stores, AUD $40B+ revenue) ### What drove the decision Coles did not wake up one morning and decide to build custom AI. The decision evolved over nearly a decade. As Silvio Giorgio, General Manager of Data and Intelligence at Coles Group, explained: "For more than nine years, we've had an artificial intelligence team that runs like a startup within Coles — a successful approach to get traction — but we're at a point now where we need to shift our focus toward scaling." The core business problem was one no off-the-shelf tool could solve at the time: predicting the flow of tens of thousands of unique products across nearly a thousand stores, accounting for local demographics, seasonal variation, and supplier constraints simultaneously. ### What they built and what it cost Coles' AI journey matured into a suite of AI models that drive day-to-day operations. This AI-backed operational efficiency helps Coles predict the flow of 20,000 stock-keeping units (SKUs) to 850 stores nationwide, harnessing insights from over 2,000 diverse data sets to make 1.6 billion predictions each day. The proprietary infrastructure underpinning this is the Intelligent Edge Backbone (IEB). Coles developed this edge computing platform using Microsoft Azure Stack HCI, Azure AI and ML — described as a global-first in retail to be deployed this way and at this scale. Since expanding its Azure Stack HCI footprint from 2 to more than 500 stores, Coles can roll out new applications to its stores six times faster using the intelligent edge backbone without disrupting workloads. ### The hybrid pivot: buying what they didn't need to build Rather than extending their internal build program to cover productivity and workforce analytics, Coles made a deliberate buy decision for non-differentiating functions. Coles announced it would be using Palantir's Foundry analytics tool and its Artificial Intelligence Platform across its 840 supermarkets to improve workforce efficiencies and supply chains. Using Palantir's software, Coles will look to identify efficiencies across 10 billion rows of data, consisting of every store in the country, team members, shifts and allocations every day. For generative AI across corporate teams, Coles again chose to buy rather than build. The rollout of ChatGPT Enterprise across Coles' corporate teams is designed to automate internal processes, reduce administrative load, and surface insights from data faster than ever before. ### The measurable results Coles is wagering heavily that automation, AI and a rebuilt supply chain will protect margins and create new engines of growth. Its "Simplify and Save to Invest" program banked a record $327 million in cost savings in FY25, largely through machine learning and analytics. Coles is targeting $1 billion over the four-year program, generally targeting about $250 million per year — with FY25 delivering a record $327 million, and the halfway tally already at $560 million in benefits. On personalisation, Coles has partnered with Microsoft to personalise the shopping experience for more than four million FlyBuys customers using 19 major AI models, and has utilised computer vision technology to transform the checkout process, achieving a 94% accuracy rate in scanning items. **Key pattern:** Coles built the proprietary intelligence layer — supply chain prediction, computer vision, personalisation — where no vendor solution existed at the required scale. It bought productivity AI and workforce analytics where commoditised tools were fit for purpose. This is the hybrid architecture in action (see our guide on *The Hybrid AI Strategy: How Australian Businesses Can Build and Buy at the Same Time*). --- ## Case Study 2: ANZ Bank — Buying the Platform, Layering the Intelligence **Decision type:** Buy (platform) + Build (intelligence layer) → Hybrid **Industry:** Financial services **Scale:** Enterprise (major bank, ~8 million customers) ### What drove the decision ANZ's primary AI use case — fraud and scam detection — sits at the intersection of two non-negotiable requirements: real-time decisioning at transaction scale, and strict APRA CPS 234 compliance obligations. Building a fraud detection engine from scratch would have required years and hundreds of millions in investment, with no guarantee of matching the sophistication of purpose-built platforms already operating at global scale. ### What they bought ANZ adopted the FICO Falcon® platform as its core fraud detection engine. ANZ Falcon® technology analyses customers' banking behaviour to recognise, flag and identify suspicious transactions, monitoring millions of transactions every day, learning from thousands of data points to build a unique picture of ANZ customers' habits to help spot the difference between legitimate and fraudulent transactions. ANZ then layered purchased intelligence on top of this platform. Its partnership with BioCatch Trust — described as the world's first inter-bank financial crime intelligence-sharing network — wasn't just about technology, it was about co-designing smarter ways to protect Australians from scams. Nearly a year into the BioCatch Trust partnership, the integration of BioCatch Trust indicators into ANZ's Falcon fraud detection platform has significantly enhanced its ability to detect complex scam typologies — while reducing friction for legitimate customers. ### The measurable results Over the first six months of the past financial year, the total number of reported fraud and scam cases fell by 9% compared to the same period a year earlier, while customer losses fell by 7%. In the same period, ANZ NZ prevented around $15 million in fraud and scam transactions. Enhanced prevention and recovery measures meant 94% of reported cases resulted in no financial loss to the customer. On the specific technology investments: behavioural alerts technology helped block over $1 million in suspicious payments within eight weeks by analysing how users interact with their devices and looking for potential red flags. Since the rollout of Confirmation of Payee, mistaken payments have dropped by 30%. **Key pattern:** ANZ recognised that fraud detection was a critical function but not a source of competitive differentiation in itself — the differentiation comes from *how* the platform is configured and layered, not from owning the underlying model. This is a textbook case for the "buy the platform, build the intelligence wrapper" hybrid approach (see our guide on *When to Buy Off-the-Shelf AI: The Scenarios Where Pre-Built Tools Win for Australian Businesses*). --- ## Case Study 3: BHP — Custom Build Where Proprietary Data Is the Moat **Decision type:** Build (custom AI + strategic partnerships) **Industry:** Mining and resources **Scale:** Global enterprise, with Australian operations as the primary test ground ### What drove the decision BHP's AI build rationale is arguably the clearest example in Australian business of the "proprietary data advantage" trigger. The company has accumulated decades of geological survey data, extraction records, equipment sensor readings, and environmental measurements that no commercial vendor has access to — and that no generic model can replicate. BHP has evolved its artificial intelligence strategy from discrete, high-value pilot projects to a globally integrated, enterprise-wide ecosystem. Between 2021 and 2024, BHP focused on validating the potential of AI through targeted partnerships and investments. ### What they built and the partnership model BHP's approach is instructive because it is not purely internal development. The company builds proprietary models on top of its own data, then partners with specialist firms for specific capability gaps — a sophisticated hybrid of internal build and strategic acquisition of external AI capability. Machine learning, coupled with human ingenuity, has allowed BHP to discover new copper deposits in Australia and the United States. Automating its shiploader facilities has increased production by more than one million tonnes each year, through greater precision, reduced spillage, faster load times, and equipment optimisation. Key initiatives include centralised exploration data platforms, computer vision in iron ore operations, and digital twins at Escondida. The company reports over US$2 billion in value from digital initiatives, but emphasises that scalable execution and integrated systems will define long-term competitive advantage. The exploration AI results are particularly striking. The strategy's success was validated by the discovery of a new deposit near its Olympic Dam operation. This deposit, identified using proprietary machine learning models, contains an estimated 1.3 billion tonnes of copper and gold. BHP's Escondida mine in Chile has saved more than three gigalitres of water and 118 gigawatt hours of energy since 2022 thanks to AI. **Key pattern:** BHP builds where its proprietary geological and operational data creates an insurmountable head start over any vendor. No off-the-shelf exploration AI tool trained on public data could replicate what BHP's models achieve with 50+ years of proprietary subsurface data. This is the "build" signal in its purest form (see our guide on *When to Build Custom AI: The Business Signals That Justify In-House Development*). --- ## Case Study 4: Telstra — The Hybrid at Scale, With a Cautionary Workforce Dimension **Decision type:** Hybrid (Buy productivity AI + Build network intelligence + Partner for AI operations) **Industry:** Telecommunications **Scale:** Enterprise (~31,000 staff, national infrastructure) ### What drove the decision Telstra's AI strategy reflects the complexity of a business with both commodity functions (customer service, HR, finance) and deeply proprietary technical domains (network management, predictive maintenance, autonomous routing). A pure buy or pure build answer was never going to fit. ### What they bought and what they built For productivity AI, Telstra chose to buy at scale. In August 2024, Telstra expanded its partnership with Microsoft, acquiring 21,000 Microsoft 365 Copilot licenses to bolster its workforce's productivity. For customer service AI — a function where Telstra's own data about its network and customers creates genuine advantage — the company built its own systems. AI-driven systems like Telstra's Ask Telstra knowledge base now handle over 79% of consumer and small business service requests, while Telstra is using AI internally to summarise customer interactions, detect scam emails and SMSes, and more. For the AI transformation of its enterprise operations, Telstra chose to partner rather than build or buy outright. Telstra announced a joint venture with Accenture in January 2025, aiming to accelerate the company's data and AI roadmap. With Accenture, it has formed an AI-focused joint venture, and the partners recently launched a Silicon Valley hub to "accelerate Telstra's foundational AI architecture that will power AI use cases and unlock business intelligence." ### The cautionary dimension: workforce impact and execution risk Telstra's AI transformation carries a significant cautionary lesson about the human cost of large-scale automation decisions. Between 2024 and 2026, the telecommunications incumbent has announced more than 3,900 role reductions across Telstra Enterprise, Network Applications & Services and Telstra Purple, with significant work being re-platformed into a $700m AI and process-automation joint venture with Accenture. The lesson here is not that AI automation is wrong — it is that the workforce planning dimension of an AI strategy must be scoped, communicated, and managed with the same rigour as the technology decision itself. When AI deployment is announced simultaneously with large headcount reductions, it creates organisational resistance that can slow implementation and damage the cultural conditions needed for AI to succeed. This is a pattern that Australian decision-makers — particularly in mid-market businesses where team cohesion is harder to rebuild — should study carefully before committing to automation-led AI programs. --- ## Case Study 5: The SME Cautionary Case — When Buying Without Strategy Fails **Decision type:** Buy (off-the-shelf, unmanaged) **Industry:** Professional services / SME segment broadly **Scale:** Small-to-medium enterprises This is not a single named case study — it is a documented pattern that has emerged across Australia's SME segment as AI adoption has accelerated without accompanying governance. ### What happened AI adoption has often been employee-led, rather than the result of a business-wide strategy, meaning the technology has been used without guardrails and with unexpected risks. The typical failure mode: a team member adopts a consumer AI tool, begins using it with client data, and the business discovers months later that it has been inadvertently exposing confidential information to third-party model training pipelines. The most immediate AI risks for small businesses involve data leakage. Staff paste customer details or confidential contracts into a chatbot. If the tool is misconfigured, that data may be stored, used for model improvement, or accessed by others. In Australia, an accidental disclosure of personal information can become a notifiable data breach under the Privacy Act. The financial and compliance consequences are real. Concerns among SME practitioners often revolve around whether the benefits of AI — such as increased efficiencies that lead to lean and cost-saving processes — outweigh the challenges that come with AI adoption, which include concerns around intellectual property, vulnerability to cybercrime, high implementation costs, and implications for the workforce. ### What went wrong and why The root cause in most SME AI failures is not the tool — it is the absence of a decision framework before adoption. A survey by the Human Technology Institute at UTS with elevenM Consulting found that 34% of SMEs said their understanding and knowledge of AI was a barrier to implementing it in their business. The survey also shows that 20% of respondents cite financial barriers, such as subscription fees, pricing models and licensing fees, as a barrier to AI integration. The lesson is sharp: buying off-the-shelf AI is not inherently lower risk than building. Ungoverned buying creates compliance exposure, vendor dependency, and workflow disruption that can exceed the cost of a properly scoped build. The decision is not just *what* to deploy — it is *how* to govern it (see our guide on *AI Data Privacy and Sovereignty: Why Australian Regulations Change the Build vs Buy Calculus*). --- ## Pattern Recognition: What These Cases Tell Us Across these case studies, five consistent patterns emerge that Australian decision-makers can use as a diagnostic overlay on their own situation: | Pattern | Enterprise Examples | SME Implication | |---|---|---| | **Build where proprietary data is irreplaceable** | BHP (geological data), Coles (supply chain data) | If your data is genuinely unique, build or partner | | **Buy where the function is commodity** | ANZ (Falcon platform), Telstra (Copilot licences) | Productivity AI, document processing, CRM AI: buy first | | **Hybrid is the dominant enterprise architecture** | Coles (IEB + Palantir + ChatGPT), ANZ (Falcon + BioCatch), Telstra (Copilot + Ask Telstra + Accenture JV) | Even SMEs can build a hybrid: buy the platform, customise the workflow | | **Ungoverned buying creates compliance risk** | SME sector broadly | Governance must precede deployment, not follow it | | **Workforce planning is part of the AI decision** | Telstra (3,900+ roles) | Factor human transition costs into the business case | --- ## Key Takeaways - **The hybrid model dominates at enterprise scale.** Every major Australian enterprise case study — Coles, ANZ, Telstra, BHP — involves a combination of custom-built and purchased AI. Pure build or pure buy is rarely the right answer above a certain organisational complexity threshold. - **The build trigger is proprietary data, not ambition.** BHP's case demonstrates that the decision to build custom AI should be anchored in data assets that no vendor can replicate — not in a preference for bespoke technology. - **Buying without governance is not the "safe" path.** The SME cautionary pattern shows that ungoverned off-the-shelf AI adoption creates Privacy Act exposure, vendor lock-in, and workflow risk that can exceed the cost of a properly scoped build program. - **Results take time to materialise.** Coles' AI team operated for nine years before the current wave of measurable savings. Decision-makers should model 18–36 month payback horizons for custom builds, not quarters. - **The workforce dimension is non-negotiable.** Telstra's experience demonstrates that AI deployment decisions cannot be separated from workforce planning. The human transition cost — in change management, retraining, and cultural resistance — must be included in the business case from day one. --- ## Conclusion The cases documented here are not outliers. They represent the leading edge of a wave of AI deployment decisions that Australian businesses across every sector are now navigating. The patterns they reveal — build where proprietary data creates a moat, buy where the function is commodity, govern before you deploy, and plan for the human transition — are the closest thing Australian decision-makers have to a validated playbook. What the data makes clear is that the build vs buy question is rarely a binary choice. It is a portfolio decision, made function by function, dataset by dataset, and risk tolerance by risk tolerance. The businesses getting it right are those that have mapped their own situation against real precedent — not against vendor promises or theoretical frameworks. For readers who want to translate these patterns into a structured decision process, see our guide on *Build vs Buy AI: A Decision Framework Tailored for Australian SMEs*. For the financial modelling that underpins the business case for each path, see *How to Build a Business Case for AI Investment in Australia: Calculating ROI for Build vs Buy Scenarios*. And for the regulatory layer that shapes every decision in this article, see *AI Data Privacy and Sovereignty: Why Australian Regulations Change the Build vs Buy Calculus*. --- ## References - ANZ Banking Group. "Redefining Scam Prevention Through Trusted Partnerships." *ANZ Bluenotes*, 2025. https://www.anz.com.au/bluenotes/2025/july/johnson-scam-prevention-partnership-biocatch/ - ANZ New Zealand. "ANZ NZ Rolls Out New Tech to Fight Fraud and Scams." *ANZ Newsroom*, October 2025. https://www.anz.com.au/newsroom/new-zealand/2025/10/anz-supercharges-scam-defences-with-tech/ - BHP. "Artificial Intelligence Is Unearthing a Smarter Future." *BHP Insights*, August 2024. https://www.bhp.com/news/bhp-insights/2024/08/artificial-intelligence-is-unearthing-a-smarter-future - eResearch. "BHP Advances AI Strategy Across the Mining Lifecycle." *eResearch Analyst Articles*, March 2026. https://eresearch.com/2026/03/08/eresearch-reports/analyst-articles/pdac-2026-article-bhp-advances-ai-strategy-across-the-mining-lifecycle - Microsoft Australia. "Coles Announces New Five-Year Strategic Partnership with Microsoft." *Microsoft Australia News Centre*, November 2024. https://news.microsoft.com/en-au/features/coles-announces-new-five-year-strategic-partnership-with-microsoft-to-accelerate-efficiency-and-drive-innovation/ - Mi-3 Australia. "Australia's Power Trio Retailers Chase Same Prize." *Mi-3*, August 2025. https://www.mi-3.com.au/29-08-2025/Australias-three-giant-shopkeepers-Woolworths-Coles-and-Wesfarmers - Inside Retail Australia. "Coles and OpenAI Partner to Integrate ChatGPT Across Teams." *Inside Retail*, November 2025. https://insideretail.com.au/digital/coles-and-openai-partner-to-integrate-chatgpt-across-teams-202511 - Department of Industry, Science and Resources. "AI Adoption in Australian Businesses — Q4 2024." *Australian Government*, 2025. https://www.industry.gov.au/news/ai-adoption-australian-businesses-2024-q4 - Taylor & Francis / Oldemeyer et al. "The Promise and Pitfalls of AI Adoption in Small and Medium-Sized Enterprises." *Special Issue Call*, 2025. https://think.taylorandfrancis.com/special_issues/the-promise-and-pitfalls-of-ai-adoption-in-small-and-medium-sized-enterprises-smes/ - Employment Hero / Decidr. "Why Most Australian SMEs Lack a Formal AI Strategy." *Employment Hero*, February 2026. https://employmenthero.com/news/sme-ai-adoption-strategy-divide-australia/ - Human Technology Institute, UTS / elevenM Consulting. "AI for SMEs: Overcoming Cost and Integration Barriers." *INTHEBLACK / CPA Australia*, 2025. https://intheblack.cpaaustralia.com.au/technology/ai-for-smes-overcoming-cost-and-integration-barriers - Information Age / ACS. "Coles Inks Deal with Controversial Big Data Firm." *Information Age*, February 2024. https://ia.acs.org.au/article/2024/coles-inks-deal-with-controversial-big-data-firm-.html - Information Age / ACS. "Telstra Triples Undersea Capacity as AI Demand Surges." *Information Age*, 2025. https://ia.acs.org.au/article/2025/telstra-triples-undersea-capacity-as-ai-demand-surges.html - Certified Strategic. "Telstra's Data Centre Disconnect: Headcount Out, Hyperscale In." *Australia Data Centre Index*, 2026. https://certifiedstrategic.com/insights/telstra-s-data-centre-disconnect-headcount-out-hyperscale-in - Australian Cyber Security Centre. "Artificial Intelligence for Small Business." *Cyber.gov.au*, January 2026. https://www.cyber.gov.au/business-government/secure-design/artificial-intelligence/artificial-intelligence-for-small-business - S&P Global Market Intelligence. "A Peek at AI Revolution in Mining: Promise Meets Peril." *S&P Global*, February 2025. https://www.spglobal.com/market-intelligence/en/news-insights/research/a-peek-at-ai-revolution-in-mining-promise-meets-peril