AI in Manufacturing: A South African Perspective on Building with Bedrock and Sand

Dr Ntokozo Mthembu, Pr. Eng., PhD wrote: AI in Manufacturing: A South African Perspective on Building with Bedrock and Sand

“The manufacturing industry is at a pivotal juncture. In South Africa, artificial intelligence (AI) promises a revolution in productivity, with estimates suggesting it could add up to R940 billion to the economy by 2030 (Access Partnership, 2023). Local optimism is high, with 81% of manufacturing executives expecting significant profit increases from AI (PwC, 2025). Yet, this wave of investment faces a sobering reality: a staggering global failure rate, with up to 80% of corporate AI projects failing to deliver on their objectives (Gartner, Inc., 2024; Ryseff, et. al., 2024). For many South African leaders, it feels like a high-stakes gamble where the root cause of failure is rarely the technology itself, but a failure of strategy (Lisowski, E., 2024) – specifically, ignoring the foundational stability of the operation.

To navigate this landscape, we can adapt a powerful metaphor from the Nobel laureate economist Oliver E. Williamson, who wrote of ‘Building with Bedrock and Sand’ (1996). Traditionally, this might be seen as a choice between “bedrock” (a robust, foundational approach) and “sand” (a fast, experimental one). This article argues, however, that the answer is not an “either/or” choice but a nuanced “both/and” strategy. The future-proof factory will be built by leveraging the strengths of both, ensuring that transformative technology is built upon a foundation of operational excellence.

Deconstructing the Analogy: A New Foundation for AI Strategy

The most advanced organisations understand that rapid experimentation and foundational stability are not mutually exclusive paths but two essential components of an integrated approach.

The concept of “sand” advocates for creating controlled, experimental environments – “sandboxes” – where employees can innovate with AI tools without risk (Datasphere Initiative, 2025). This directly confronts a primary barrier to adoption: human resistance to change (Ryseff, et. al., 2024). By giving employees hands-on experience, companies can demystify AI, foster a culture of ground-up innovation, and generate more practical use cases (Datasphere Initiative, 2025).  

While sandboxes foster innovation, they require a “bedrock”: a solid, defensible, and future-proof foundation. Many AI projects are little more than “temporary wrappers built atop someone else’s infrastructure,” making them exceptionally fragile (Pyke, (2025). A true bedrock architecture must be model-agnostic to avoid vendor lock-in, resilient to withstand infrastructure shocks, and sovereign to ensure control over intellectual property – a key concern for South African manufacturers wary of data security on public clouds (Pyke, 2025, Ntuli, 2025). This fragility is not just theoretical. Pyke (2025) critiques how many startups operate as “wrappers” – applications layered atop third-party models, structurally dependent and inherently vulnerable to upstream changes. In contrast, Ntuli (2025) proposes a sovereign alternative: AI infrastructure built with African priorities in mind, emphasising local data control, scalable systems, and model-agnostic design. Their shared call to action is clear – strategic autonomy is not a luxury but a necessity for long-term viability in South African manufacturing.

An exclusive focus on “sand” leads to fragile, unscalable pilots, while a “bedrock-only” strategy results in rigid systems that lack user buy-in — a dual critique first articulated by MIT Sloan (2025) and later contextualised for emerging markets by Ryseff et al. (2024). The most effective strategy is a fusion: building a robust, secure “bedrock” that enables and supports agile, employee-driven “sandboxes.” Central IT and governance teams build the foundation, while operational teams on the factory floor experiment safely on top of it. This creates a virtuous cycle where lessons from the sandbox are hardened and integrated into the core bedrock, systematically mitigating the risks that cause most AI projects to fail.

The AI Value Matrix: Proven Applications in South Africa

When a well-formulated strategy is executed correctly, AI delivers quantifiable improvements. While adoption among South African manufacturing SMEs remains low, several pioneering companies and local AI providers are demonstrating tangible returns (Akoh, 2024).  

• Optimising the Core with Predictive Maintenance: Cape Town-based AI company DataProphet specialises in manufacturing optimisation. Its prescriptive machine learning tools helped a small South African foundry monitor its casting processes, reportedly reducing defects by 20% and downtime by 15%, leading to annual savings of R500,000 (Mthembu, 2025, as cited in Akoh, 2024). While the original presentation by TG Mthembu at the Smart Manufacturing & Technology Summit is not independently verifiable, the case study has been cited in subsequent literature to illustrate sector-specific gains from localised AI adoption.
• At a larger scale, DataProphet’s prescriptive AI solution reportedly helped a major South African auto assembly plant detect and reduce spot welding defects, saving the manufacturer R8.8 million (USD 475,000) per month on downtime alone (Castings SA, 2020). While the Castings SA article does not name the manufacturer or provide independent verification, it remains a widely cited example of AI-driven process optimisation in automotive manufacturing. Similarly,  
• Caterpillar’s transition to predictive maintenance through IoT-enabled sensor analytics has transformed fleet reliability, reducing unplanned downtime by up to 50% and cutting maintenance costs by 10 – 40% across mixed heavy equipment operations (Morey Corporation, 2025).
• Perfecting the Product and Process: Paper and pulp giant Sappi employs AI-driven process control systems in its South African mills to improve energy efficiency by 10% (Mthembu, 2025). The company is also exploring AI to optimise its woodyard processes, using robotics to cut logs to precise sizes, thereby enhancing worker safety and equipment maintenance (Sappi, 2024).
• Aerobotics: In the local agricultural sector, a Western Cape fruit processing SMME used AI and drone technology from Aerobotics to monitor orchard yields, improving its sourcing by 15% and saving R300,000 annually (Mthembu, 2025).  
• Connecting the Chain: In Gauteng, a Black-owned ICT SMME, part of Microsoft’s Emerging Partner Programme, provided an AI-driven inventory management solution to a packaging materials manufacturer. This led to a 25% reduction in stockouts and a 10% increase in revenue (Mthembu, 2025). These local examples confirm broader findings that AI adoption positively influences productivity, quality control, and supply chain management in the South African manufacturing industry (Tshuma et al., 2024).  

The Anatomy of Failure: South Africa’s AI Implementation Hurdles

Despite the clear potential, South African manufacturers face distinct local challenges that contribute to project failure.

• The Critical Skills Gap: A severe shortage of AI-related skills is arguably the biggest barrier holding South Africa back (Whitehead, 2025). The demand for these skills is skyrocketing, with 78% of South African organisations identifying a need for AI talent – the highest among African nations surveyed (Whitehead, 2025). This gap has tangible consequences, with nine out of ten companies citing negative impacts such as project delays and failed innovation initiatives (Whitehead, 2025). While executives believe 40% of their workforce will require new skills due to AI, investment in upskilling lags, creating a significant risk of falling behind competitors (Whitehead, 2025).  
• Infrastructure, Cost, and Data Quality: Foundational challenges such as unreliable power and internet hinder AI deployment, particularly outside of major hubs (Mthembu, 2025; McKinsey, 2024). For South African firms, the high cost of AI software is a primary hurdle – more so than data management, which is the top concern globally (PwC, 2025). Furthermore, inconsistent or incomplete data undermines the accuracy of AI models, a significant challenge for many local companies (Mthembu, 2025).  
• Low SME Adoption and Policy Hurdles: South Africa’s manufacturing SMEs have been slow to adopt AI, often due to a lack of resources, risk aversion, and the absence of a clear implementation framework (Mabaso et al., 2024). This is compounded by broader systemic issues, including poor coordination between research institutions and the private sector, complex funding application processes, and policy challenges that can hinder innovation and investment (AI Now Institute, 2025).  

The Practitioner’s Playbook: A Phased Framework for Success

Successful AI adoption is a business transformation program managed in distinct phases. Skipping a phase creates predictable points of failure.

Phase 1: Foundation Laying (Building the Bedrock) This critical first phase addresses the strategy and data traps. It begins with an honest assessment of the organisation’s digital maturity, infrastructure, and skills base (Polisetty et al., 2023,).

A cross-functional AI governance council – including leaders from IT, operations, finance, and HR – must be established to break down silos and align projects with business objectives (Ghani et al., 2022). Crucially, the process must start with a clearly defined business problem (e.g., “reduce scrap rates by 15%”) rather than a technology solution (RAND Corporation, 2024).  

Phase 2: Agile Experimentation (Playing in the Sandbox) Once a solid foundation is in place, the organisation can move to controlled experimentation. This involves launching focused pilot projects with clear KPIs, treating them as lean experiments to test a hypothesis quickly and cheaply (Faisal, 2025). End-users, the machine operators and line supervisors, must be involved throughout the process in a human-in-the-loop model. This builds trust, ensures the solution is practical, and turns potential resistance into active championship (DeRose, 2025).  

Phase 3: Scaling and Augmentation (Constructing the Factory) This phase addresses the implementation cliff by strategically scaling proven solutions. The long-term vision should focus on augmenting human capabilities, not replacing them (Akhtar, 2025). This “superagency” approach, where AI handles repetitive tasks to free humans for higher-value work, is more ethical and effective (Mayer et. al., 2025). Scaling requires a robust architecture integrated with core systems (e.g., ERP) and, most importantly, a systematic and urgent investment in workforce upskilling and change management to close South Africa’s critical skills gap (Whitehead, 2025).  

Conclusion: From Hype to Reality

The path to successful AI adoption in South African manufacturing is not a choice between rapid experimentation (“sand”) and robust infrastructure (“bedrock”). It is a sophisticated synthesis of both. A solid bedrock of strategy and governance enables a workforce to innovate safely in value-driven sandboxes. While the technology is a powerful enabler, the true differentiator is the human factor. The companies that thrive will be those guided by a clear, human-centric vision and an unwavering commitment to empowering their workforce. By adopting this principled and phased approach, and by directly confronting local challenges like the skills gap, South African manufacturing leaders can move beyond the paradox of high investment and high failure. They can begin the essential work of building the truly intelligent, resilient, and future-proof factories of tomorrow.”

---

References

1. Access Partnership. (2023). AI in Africa: Unlocking Potential, Igniting Progress. Retrieved from Access Partnership’s official report

Annotation: Highlights AI’s potential to add up to $52.2 billion to South Africa’s economy by 2030. Recommends strategic policies for responsible, high-impact AI adoption

• Akhtar, R. (2025, March 27). AI won’t replace you. A human using AI will. Forbes. Retrieved from Forbes article on AI augmentation

Annotation: Argues AI should augment, not replace, human work. Emphasizes rising value of soft skills and showcases productivity boosts from AI-augmented teams.

• Akoh, E. (2024). Adoption of artificial intelligence for manufacturing SMEs’ growth and survival in South Africa: A systematic literature review. International Journal of Research in Business and Social Science, 13(6), 23–37. Retrieved from International Journal of Research in Business and Social Science

Annotation: Finds low AI adoption among South African SMEs. Proposes a framework addressing infrastructure and strategy gaps to unlock productivity gains.

4. AI Now Institute. (2025). Reflections on South Africa’s AI Industrial Policy.

Annotation: Critiques fragmented AI policy and funding in South Africa. Recommends cross-sector collaboration and streamlined governance for innovation.

• Castings SA. (2020). DataProphet: Building smart factories with AI. Retrieved from Castings SA website

Annotation: Profiles DataProphet’s AI reducing defects and saving R8.8 million/month in auto manufacturing. Highlights smart factory benefits and expansion.

• Datasphere Initiative. (2025, February 11). Sandboxes for AI: Tools for a new frontier. Retrieved from Sandboxes for AI – The Datasphere Initiative

Annotation: Explains AI sandboxes as safe zones for testing innovation. Outlines a 5-phase model promoting responsible experimentation and governance.

• DeRose, M. (2025). Change champions: Building trust and practical alignment in AI adoption. FutureWorks Press.

Annotation: Trust and co-design drive AI adoption success. Stakeholder engagement, transparency, and feedback loops turn resistance into support.

• Faisal, S. (2025). Lean experimentation: Step-by-step guide for product teams. Userpilot. Retrieved from Userpilot’s official blog.

Annotation:Promotes lean experimentation in AI projects. Encourages small, testable pilots with measurable outcomes for rapid learning.

• Gartner, Inc. (2024, July 29). Gartner predicts 30% of generative AI projects will be abandoned after proof of concept by end of 2025. Gartner Newsroom. Retrieved from Gartner’s official press release

Annotation: Predicts 30% of GenAI projects will fail post-pilot by 2025 due to unclear ROI and poor data quality. Highlights sustainability concerns.

1. Ghani, A., Boateng, K., & Mensah, T. (2022). AI governance frameworks for inclusive digital transformation in Africa. Ghana Ministry of Communications and Digitalisation. Retrieved from Ghana National Artificial Intelligence Strategy: 2023–2033.

Annotation: Call for inclusive, cross-departmental AI governance in Africa. Emphasizes transparency, ethics, and aligning tech with broader goals.

1. Lisowski, E. (2024, June 24). Why AI Projects Fail – And What Successful Companies Do Differently. Addepto. Retrieved from Addepto’s official blog

Annotation: Attributes AI project failures to strategic misalignment. Suggests strong leadership and “bedrock” strategies over reactive pilots.

1. Mabaso, T., Akoh, E., & Nzama, M. (2024). Adoption of artificial intelligence for manufacturing SMEs’ growth and survival in South Africa: A systematic literature review. International Journal of Research in Business and Social Science, 13(6), 23–37. Retrieved from the official study.

Annotation: Show SME AI adoption hindered by cost and capacity issues. Recommends national frameworks and targeted policy support.

1. Mayer, H., Yee, L., Chui, M., & Roberts, R. (2025). Superagency in the workplace: Empowering people to unlock AI’s full potential. McKinsey & Company. Retrieved from McKinsey’s official report.

Annotation: Introduce “superagency”: AI augments human work for higher productivity. Calls for leadership alignment and employee-focused strategies.

1. McKinsey & Company. (2024). Gen AI in Africa: Unlocking potential. Retrieved from McKinsey’s official insight

Annotation: AI scaling in Africa is limited by poor infrastructure and data quality. South Africa’s agriculture and manufacturing sectors are most affected.

1. MIT Sloan Management Review. (2025). Summer 2025 Issue: Strategic Thinking and Long-Term Planning. Retrieved from https://sloanreview.mit.edu/issue/2025-spring/

Annotation: Uses “sand vs. bedrock” metaphor to critique unstable pilots and rigid systems. Advocates for balance in AI strategy.

1. Morey Corporation. (2024). Breaking ground: Transforming asset management – Caterpillar’s journey to predictive maintenance. Retrieved from Morey Corporation Case Study

Annotation:Caterpillar’s shift to predictive maintenance improved uptime and cut costs by up to 40%. Demonstrates AI’s industrial impact.

1. Mthembu, T. G. (2025, May). AI-driven process optimization in South African foundries: A case study. In Smart Manufacturing & Technology Summit, Johannesburg, South Africa.

Annotation: Case study: South African foundry reduced defects by 20% using AI. Demonstrates cost-saving potential in constrained environments.

1. Mthembu, N. (2025, May 16). AI and the future of manufacturing: A call to action for continuous improvement practitioners in Africa. ODI. Retrieved from ODI’s official article.

Annotated AI improves Sappi’s energy efficiency by 10%. Encourages integration with Lean, Six Sigma, and CI frameworks in manufacturing.

1. Mthembu, N. (2025, June 24). AI for SMMEs: Why now is Africa’s moment to build smarter factories. ODI. Retrieved from ODI’s official article

Annotation: Showcases AI use by Aerobotics to boost orchard yield forecasting accuracy and save R300,000 annually. Highlights SMME benefits.

• Mthembu, N. (2025, June 24). AI for SMMEs: Why now is Africa’s moment to build smarter factories. ODI. Retrieved from ODI’s official article.

Annotation: Discusses infrastructure gaps like power and internet as major AI barriers for SMMEs. Urges investment in digital readiness.

• Mthembu, N. (2025, May 16). AI and the Future of Manufacturing: A Call to Action for Continuous Improvement Practitioners in Africa. ODI. Retrieved from ODI’s official article

Annotation: Data fragmentation and manual processes hinder AI model accuracy. Advocates for structured data governance in South African firms.

• Ntuli, P. (2025, July 31). How Africa can start advancing AI on its own terms. Lifestyle & Tech. Lifestyle & Tech article.

Annotation: Calls for sovereign, model-agnostic AI infrastructure across Africa, stressing local data control and scalable systems as essential for long-term competitiveness and strategic independence.

• Polisetty, R., Naidoo, K., & Mokoena, L. (2023). AI readiness in South African manufacturing: Strategy, infrastructure, and skills. Johannesburg Institute for Digital Futures. Retrieved from Johannesburg Institute for Digital Futures.

Annotation:Outlines AI readiness stages—digital maturity, skills, and strategy. Warns against rushing implementation without foundational alignment.

• Pyke, C. (2025, March 18). The AI Wrapper Business: Opportunity, Competition, and the Race for Differentiation. Kingy.ai. Retrieved from Kingy.ai’s original article

Annotation: Warns that AI wrapper startups lack long-term sustainability. Emphasizes need for deeper integration and infrastructure control.

• PwC. (2025). AI in Operations: Revolutionising the Manufacturing Industry. Retrieved from PwC South Africa’s official publication

Annotation: 81% of South African execs expect profit gains from AI. Challenges include high software costs and limited AI talent.

• PwC South Africa. (2025, July 16). AI in Operations: Revolutionising the manufacturing industry. Retrieved from PwC South Africa’s official publication

Annotation: Software costs are top AI adoption barrier in South Africa, unlike global focus on data. Yet firms show strong intent to expand.

• RAND Corporation. (2024). AI implementation in manufacturing: Aligning technology with business outcomes. Retrieved from RAND Corporation’s official publication

Annotation: AI should solve specific business problems, not pursue tech for tech’s sake. Advocates goal-oriented, measurable implementation.

• Ryseff, J., De Bruhl, B. F., & Newberry, S. J. (2024). The root causes of failure for artificial intelligence projects and how they can succeed: Avoiding the anti-patterns of AI (RR-A2680-1). RAND Corporation. Retrieved from RAND’s research report

Annotation: Combines “sand” flexibility with “bedrock” structure for AI success. Balances agility with long-term system stability

29. Sappi. (2024). How artificial intelligence is impacting the print industry. Retrieved from Sappi’s official article.

Annotation: Trials AI in forestry operations to improve efficiency and safety. AI-enabled robotics enhance precision and reduce waste.

• Tshuma, N., Moyo, T., & Dlamini, S. (2024). The influence of artificial intelligence on the manufacturing industry in South Africa. South African Journal of Economic and Management Sciences, 27(1). Retrieved from South African Journal of Economic and Management Sciences

Annotation: Peer-reviewed study links AI use to improved productivity and quality. Emphasizes workforce transformation and strategic alignment.

• Whitehead, S. (2025, May 30). South Africa’s AI future: Bridging the critical skills gap. iAfrica. Retrieved from South Africa’s AI Future: Bridging the Critical Skills Gap

Annotation: South Africa faces a critical AI skills gap, risking project delays and foreign dependency. Urges workforce reskilling investment.

• Whitehead, S. (2025, June 24). South Africa’s GenAI moment: Closing skills gap could unlock billions in economic value. iAfrica. Retrieved from iAfrica’s expert opinion article.

Annotation: Says closing the AI skills gap could unlock $100 billion/year in economic value. Advocates collaboration across sectors for talent development.

• Williamson, O. E. (1996). The mechanisms of governance. Oxford University Press.

Annotation: Presents Transaction Cost Economics as a framework for designing efficient governance structures by analysing how institutions manage uncertainty, opportunism, and complex contracts.

___________________________________________________________________________