Watch part one of our conversation with Tobias Leong, Chief Technology Officer, Axium Industries. The full interview is available through AIM Asia Circle.
Most manufacturers no longer need convincing that AI matters. The pressure is already coming from the board, the CEO, or the competitor down the road who’s moved faster. The harder question is what comes after the pilot: how do you turn an AI project into a measurable production gain, without over-investing in the wrong workload?
Tobias Leong, Chief Technology Officer, Axium Industries Group, works on exactly this problem. Axium deploys industrial AI for companies across manufacturing, energy, maritime, and physical retail, with a strong base of manufacturing customers in Malaysia. His view, shaped by deployments across these sectors, is that AI ROI has less to do with model selection and more to do with data readiness and use-case prioritisation, decisions companies make long before an AI agent ever touches the factory floor.
AI delivers measurable ROI in manufacturing when companies first structure their operational data, then prioritise AI use cases based on what humans cannot do, should not do, and do not want to do. According to Tobias Leong, CTO, Axium Industries Group, manufacturers who follow this sequence, rather than starting with the AI model, see faster deployment, clearer returns, and less resistance to scaling.
Why Manufacturing AI Is Harder Than Simply Choosing an AI Model
A common assumption is that AI adoption starts with picking a model or a vendor. Leong’s experience points the other way: the real bottleneck is understanding the problem and the data well enough to apply AI to it.
He points to a frequently cited industry figure that a large share of AI projects fail and attributes this, based on Axium’s own deployment experience, to three recurring gaps:
Insufficient problem definition: teams aren’t curious enough about the actual operational issue before reaching for AI.
Insufficient data readiness: companies haven’t structured their data well enough for an AI agent to use it.
Unaddressed real-world deployment constraints: questions like whether the solution can run on-premises, whether it’s tied to a specific model provider, and whether sensitive data can leave the company at all.
The 3 Vs of Manufacturing Data: Variety, Volume and Velocity
Before any AI agent can be useful, Leong argues the underlying data has to be usable. Axium frames this around three characteristics common to manufacturing environments:
Variety: Data is scattered across ERP systems, SharePoint folders, Excel spreadsheets, PDF documents, and structured tabular data such as sales and transaction records. Most manufacturers have all of this living in silos rather than one place.
Volume: Industrial processes generate large volumes of sensor and process data. Leong notes that many companies already have this data but avoid using it simply because of its scale “I don’t want to even touch it because it’s way too big.”
Velocity: High-tech manufacturing environments, from semiconductor fabs to automotive and aerospace lines, generate data at high frequency. Leong cites sensors running at roughly 10 kilohertz, producing more than 10,000 data points every second.
Importantly, Leong is direct about one common misconception: companies don’t need a data warehouse or a full ERP system to begin. In Axium’s experience, some customers started with nothing more than a handful of Excel sheets and PDF files, which was a sufficient base to build an initial knowledge base and begin structuring data for agentic AI.
Why Data Foundations Matter for Agentic AI
Axium’s approach follows a two-phase sequence: first consolidate scattered data into what Leong calls a data fabric, then apply agentic AI for intelligent automation and tasks that go beyond current human capacity.
The logic is straightforward: data → knowledge base → AI agents → intelligent automation → business outcomes. Skipping the first step is, in Leong’s framing, one of the more common reasons AI initiatives underdeliver: the agent is only as useful as the structured knowledge it can draw on.
From AI Pilots to Real Manufacturing ROI: A Prioritisation Framework
Rather than treating every process as an equally good AI candidate, Leong recommends a simple filter for identifying where to start. He frames it around three categories of work:
What humans cannot do: Complex forecasting models, for example, where companies lack the right data, talent, or depth of understanding to solve the problem manually.
What humans should not do: Work that is potentially dangerous or unsafe, where AI-supported automation is a better starting point than continued manual handling.
What humans do not want to do: Repetitive, tedious tasks such as document checking or manual data entry, which drain time without adding much value.
Leong also frames this as a growth narrative rather than a headcount reduction story: freeing employees from repetitive or unsafe work allows them to be redirected toward higher-value activities. Once these candidate workloads are identified, he recommends prioritising by expected ROI: starting with whichever use case is likely to deliver the clearest return first.
Where AI Can Deliver Immediate Manufacturing Value
Three examples from Axium’s work illustrate how this plays out in practice.
Procurement Automation
In the energy sector, Axium worked with a company where reviewing tender bids was a slow, manual process. With multiple bidders each submitting 70–100 page documents, comparing technical specifications and commercial terms against requirements could take roughly six weeks per bid package. Axium’s AI agents were deployed to read the requirements, validate bidder submissions against them, flag missing or insufficient information, and generate clarification letters, complete with the company’s own letterhead automatically. According to Leong, this shortened response times to bidders from several weeks to a matter of days, to the point that bidders themselves noticed and commented on the turnaround. Axium has also applied a version of this on the bidder side, in the construction sector, helping companies prepare and adapt their own submissions to procurement formats faster.
Manufacturing Process Optimisation
Working with manufacturers in Malaysia, Axium analysed existing production processes to identify bottlenecks and redesign work combinations across a production line, reducing takt time not just at a single station, but across the entire line. The result, per Leong, was an increase in overall production capacity. Axium has also applied AI-supported defect detection to help manufacturers improve output quality alongside quantity, which Leong ties to improvements in OEE (Overall Equipment Effectiveness) metrics and, in some cases, savings in the millions of ringgit.
Inventory Optimisation
In a separate energy-sector engagement, Axium used AI agents to analyse a large, complex spare-parts inventory across offshore platforms, identifying items with duplicate codes that were effectively the same part, and helping the company set appropriate minimum and maximum stock levels. Leong states this work saved the company more than 10 million (currency as referenced in the interview) in reduced duplicate inventory and better working-capital efficiency.
The Human Factor Behind Successful Industrial AI
Leong is clear that data and models are only part of the equation, the people deploying AI matter as much as the technology itself. Axium’s engineering teams operate what Leong describes as a forward deployed engineer model (a concept he notes was popularised by Palantir): engineers work directly on the factory floor to understand real production context, rather than building solutions remotely.
He identifies three traits Axium looks for in these engineers:
Strong computer science fundamentals
Effective communication, the ability to translate technical concepts for non-technical stakeholders and ask the right clarifying questions
Intellectual curiosity, engineers who study manufacturing processes and chemical or industrial fundamentals outside their formal training, so they understand why a process works the way it does, not just what data it produces
He also points to the growing importance of roles that bridge IT and OT (Operational Technology), historically siloed functions with separate budgets and priorities, as a factor accelerating successful AI deployment inside manufacturers that are moving fastest in this space.
Tobias Leong is careful not to overstate AI’s autonomy. Even with strong deployments, he notes that full autonomy across an entire production process isn’t realistic today, human oversight remains part of the design, both as a safety gatekeeper and because domain expertise still sits with people. Axium’s agents are engineered with two components: an experience layer (broad industrial knowledge plus company-specific learning) and a memory layer (retaining company-specific facts and preferences over time), with humans typically remaining in the loop for final approval.
What Manufacturers Should Do Before Starting Their AI Journey
Based on Leong’s advice, manufacturers considering AI adoption can work through a practical checklist:
Examine existing processes end-to-end before selecting a technology.
Identify tasks humans genuinely cannot do well today.
Identify dangerous or unsafe tasks that are better suited to automation.
Identify repetitive, low-value tasks that drain employee time.
Assess what data already exists, even unstructured Excel sheets and PDFs are a valid starting point.
Prioritise use cases by expected ROI, not by novelty.
Start with one well-defined problem rather than a broad AI programme.
Measure the business outcome clearly before expanding.
Once a use case proves out, scale to additional workloads, Leong notes this is the pattern Axium sees consistently once customers see results, moving from one AI agent to several working together across the plant.
Meet Tobias Leong, Chief Technology Officer, Axium Industries Group
Tobias Leong leads technology and co-founded Axium Industries Group, which deploys industrial AI for companies across manufacturing, energy, maritime, and physical retail, sectors that make up a substantial share of ASEAN’s GDP contribution. Axium is headquartered in Singapore with a large team based in Kuala Lumpur, and works closely with universities across Malaysia to recruit engineering talent for its forward-deployed teams.
Tobias will be sharing his insights on “Building Intelligent Supply Chains by Leveraging IoT, 5G, & Digital Twins to Ensure Agility, Transparency & Responsiveness” at the 3rd Annual AIM Asia Week 2026 in Penang.
Frequently Asked Questions
What is industrial AI in manufacturing?
Industrial AI refers to the application of AI, particularly AI agents, to manufacturing operations such as production optimisation, procurement, quality control, and inventory management. According to Tobias Leong of Axium Industries Group, it typically requires structured company data before it can deliver reliable results.
How is AI used in manufacturing?
Manufacturers use AI for tasks including defect detection, production line optimisation (reducing bottlenecks and takt time), procurement document review, and inventory analysis. Axium’s deployments span these areas across manufacturing, energy, and other industrial sectors.
What data do manufacturers need before adopting AI?
Companies don’t need a full ERP system or data warehouse to start. Leong notes that even basic sources like Excel spreadsheets and PDF documents can form a sufficient starting base, provided the data is eventually structured to address variety, volume, and velocity.
What is agentic AI in manufacturing?
Agentic AI refers to AI systems that can take autonomous action on structured company data, such as validating procurement bids or flagging inventory duplicates, rather than simply generating information for a human to act on. In Axium’s deployments, these agents still typically operate with human oversight.
How can manufacturers measure AI ROI?
Leong recommends measuring ROI against a specific, well-defined use case, such as production capacity increase, procurement turnaround time, or inventory cost savings, rather than evaluating AI adoption broadly. Starting narrow makes the return easier to isolate and prove.
How can manufacturers move beyond AI pilot projects?
According to Leong, the path from pilot to production runs through prioritisation: identifying which workloads offer the clearest ROI, proving out one use case, and then expanding to additional workloads once results are demonstrated, rather than attempting a broad rollout from the start.
What are the biggest challenges in implementing AI in factories?
Leong points to three recurring challenges: not understanding the operational problem deeply enough before applying AI, insufficient data readiness (the 3 Vs), and unresolved real-world deployment considerations such as data residency, on-premises requirements, and model flexibility.
Why This Conversation Matters for Asia’s Manufacturing Sector
Tobias Leong sees AIM Asia as more than an industry event, he points specifically to the presence of government alongside industry as a distinguishing factor, describing the need for what he calls a tripartite partnership between industry, government, and capital (VCs and private equity) to move manufacturing transformation forward. He also points to Penang’s own development as state infrastructure and investment converge with this kind of industry gathering.
His practical framing for manufacturers considering whether to attend: rather than spending months independently researching AI adoption, a few days of direct conversation with peers who’ve already walked this path can meaningfully shorten that learning curve, walking away with both ideas and a network to test them against.
These are the kinds of implementation questions shaping the next phase of manufacturing transformation across Asia. At the Annual AIM Asia Week, manufacturing leaders, technology innovators, government representatives, and investors will explore how AI, innovation, and human capital translate into practical industrial transformation, including through AIMxchange Arena, where project owners connect directly with solution providers like Axium.
Register now to secure your place
Explore the Full Axium Interview on AIM Asia Circle
This article features highlights from Section 1 of our conversation with Tobias Leong, Chief Technology Officer, Axium Industries. The conversation continues in Sections 2, 3 and 4, where we explore more insights into industrial AI, manufacturing transformation and real-world AI adoption. Access the full interview exclusively on AIM Asia Circle and continue the conversation with the manufacturing and innovation community. Join AIM Asia Circle: https://aim-asias-community.circle.so/aim-asia-circle. Contact us at [email protected] to gain access to the exclusive AIM Circle platform and stay connected with industry leaders and opportunities.
Connect with AIM Asia
Join us at our upcoming AIM Asia programmes to connect with industry leaders, innovators, governments, investors, and technology providers shaping the future of manufacturing and industrial transformation across ASEAN. Connect with us to gain access to the exclusive AIM Circle platform and stay connected with industry leaders and opportunities across the advanced manufacturing ecosystem.
For AIM Circle access, speaking, sponsorship, exhibition, and partnership opportunities, contact [email protected] or +65 8868 1418 (SG) | +6011 1616 3281 (MY)