What 133 Malaysian SMEs Tell Us About AI Adoption — and Why Open Source Isn’t Enough Alone

Malaysian SMEs want AI to grow the business without growing headcount. Turning that ambition into working systems is still the hard part.

In October 2025, research firm Ecosystm published Accelerating SME AI Adoption Through Open Source in Malaysia’s Digital Future, developed in collaboration with Red Hat and supported by Malaysia’s National Artificial Intelligence Office (NAIO). The study surveyed 133 SMEs in MDEC’s network and interviewed NAIO stakeholders.

This post summarises the findings that matter most for operators — with attribution to Ecosystm and Red Hat — then adds Oxydata’s practical view of what SMEs should do next. For the full charts and methodology, read the whitepaper or the Red Hat / NAIO press release.

Why this study matters

SMEs are nearly 97% of Malaysian businesses, employ about 48% of the workforce, and contribute roughly 38% of GDP (figures cited in the whitepaper). National AI ambition without SME execution stays a corporate story.

The Ecosystm study sits beside Malaysia’s broader policy push (MyDigital, NAIO, the National AI Action Plan). Policy sets direction. This research shows where SME capability actually is.

Interest is high. Scale is rare.

SMEs do not treat AI as a toy. Top expected benefits include:

• 44% — support business growth without increasing overhead
• 41% — create new products, services, or features with AI

Awareness and experimentation are real. Roughly 36% are researching, experimenting, or running early proofs of concept. But only about 21% have moved meaningfully beyond limited pilots into broader deployment with measurable impact. Around 30% report no active exploration yet.

Oxydata take: This matches what we see on the ground. Demos and pilots are easy. Production systems that answer real customer questions, respect approved knowledge, and hand off to humans are harder — and that is where value sits.

The two barriers that stall most projects

The whitepaper’s execution gap is blunt:

• 60% cite a lack of in-house technical skills
• 52% cite high implementation and maintenance costs

That combination creates a loop: limited skills → more vendor dependence → higher cost → less budget to build internal capability. Other friction includes unclear strategy, change and confidence issues among staff, data quality and access problems, and thin AI ethics or governance structures. The study notes that a large majority of SMEs still lack a formal AI ethics structure.

Oxydata take: Buying another chatbot licence does not fix a skills or data problem. Neither does waiting for a perfect in-house AI team. Most Malaysian SMEs need a partner who ships a narrow use case, documents the knowledge base, and leaves the client able to operate — not locked into opaque black boxes.

Where SMEs are applying AI

Product innovation, customer service (including chatbots and virtual assistants), and process automation show up as early focus areas. Generative AI is also appearing in IT operations and product work. Lead generation and scoring remain earlier for many.

That pattern is familiar: customer-facing answers and repetitive ops first; deeper analytics later.

Open source is attractive — and still underused

Open source shows up as a practical leveller for cost and lock-in. Among reasons SMEs embrace open-source AI technologies:

• 54% — cost-effectiveness / lower licensing fees
• 51% — flexibility and easier customisation

Other motivators include faster innovation cycles, data control (including on-prem options), transparency, and avoiding vendor lock-in.

Adoption is uneven. Many SMEs still lean on proprietary tools or keep open source in non-critical experiments. About 21% already use open-source tools and frameworks in production workflows. Open models (language, vision, audio) are approached cautiously — a large share prefer proprietary options or remain in early testing. Engagement with open datasets and community initiatives is limited for more than half of respondents.

Security, performance, support, and internal expertise remain the main hesitations around open source — perceptions the paper argues are often manageable with governance and the right partners.

Oxydata take: Open source lowers the licence line item. It does not remove the need for integration, data readiness, Bahasa Malaysia / English conversation design, CRM handoff, or PDPA-aware deployment. Model-agnostic delivery — open weights where they fit, commercial APIs where they fit — usually beats ideology.

What Oxydata recommends SMEs do next

Use the whitepaper as a diagnosis. Then pick one production path:

1. Choose one revenue-linked use case — for example WhatsApp enquiry handling (IRIS), FAQ / product Q&A on approved content (RAG), or CV screening (OPAL) — not a vague “AI transformation.”
2. Ground answers in your knowledge — pricing ranges, catalogues, policies, SOPs. If the AI invents, you lose trust faster than a slow human reply.
3. Fix data enough for that use case — completeness, definitions, and PII boundaries before you scale. See our data quality checklist.
4. Design human handoff on day one — AI for repeats; people for negotiation, exceptions, and brand-sensitive promises.
5. Measure something simple — reply time, qualified leads, hours saved, or escalation rate. If you cannot measure, you cannot defend ROI to the board.

Open source can sit under several of these steps. A delivery partner who understands Malaysian SME operations usually matters more than which logo is on the model.

Bottom line

The Ecosystm / Red Hat / NAIO study confirms what operators already feel: Malaysian SMEs see AI as a growth lever, but talent and cost keep most work stuck in pilots. Open source is a real enabler for cost and flexibility — not a substitute for strategy, data, and working use cases.

If you want help moving from interest to a scoped production pilot, talk to Oxydata.

Source: Ecosystm, “Accelerating SME AI Adoption Through Open Source in Malaysia’s Digital Future” (October 2025), in collaboration with Red Hat and supported by NAIO. Statistics above are summarised from that research; please refer to the original whitepaper for full context and charts. Reuse of the original material requires attribution to Ecosystm and Red Hat.