Why SMEs are being left behind in the AI integration race

The Business Times, Why SMEs are being left behind in the AI integration race

Organisations that succeed are not those with the largest engineering teams, but those that can effectively embed institutional expertise into AI systems

KPMG’s recent partnership with Anthropic reflects how organisations are moving beyond experimentation towards deployments that integrate artificial intelligence into core workflows, and embed it within legal, compliance and operational frameworks.

Previously, public discourse around AI focused on access. Who has experimented with ChatGPT or Claude? What is the best prompt?  Today, it is about depth of integration. How can organisations use AI safely, deeply and systematically?

Although the Infocomm Media Development Authority (IMDA) reported that three-quarters of the Singapore workforce use AI at work, AI adoption is not the same as AI integration. The difference lies in whether organisations have the governance structures, data controls and operational frameworks required to deploy AI in sensitive or regulated environments.

AI adoption by small and medium-sized enterprises (SMEs) increased from 4.2 per cent in 2023 to 14.5 per cent in 2024. But this is still low compared to large enterprises, where adoption rates rose from 44 per cent to 62.5 per cent in the same period.

Over time, organisations that embed AI into core workflows may see productivity gains that direct further investment in infrastructure and systems. This creates a self-reinforcing cycle that can widen the AI divide between large organisations and SMEs.

Structural barriers

This is cause for concern. SMEs comprise 99 per cent of businesses in Singapore and may benefit the most from AI integration. SMEs may have highly specific operational problems that AI could help solve.

In many cases, they are closer to the customer pain points than large corporations. Some of the most consequential applications of AI may come from clinics, schools, professional practices and small businesses. They have the context. They have the knowledge. But they face structural barriers.

Consider a small accounting firm serving a family office or listed company. Its client contracts may explicitly prohibit third-party data processing. Or consider a school using AI to identify students at risk – cross-referencing grades, learning needs, behavioural patterns, family background and pastoral notes. Should that data be processed through a cloud system?

When sensitive information is sent to a cloud-based AI model, the exposure is not only in the output; it is in the input. Prompts reveal strategy. Uploaded files contain client data.

Analytical queries expose proprietary thinking.

Even when the final response looks innocuous, the process that created it may have passed through systems the organisation does not own or fully control.

Large organisations can increasingly pay for safe integration. They can afford private deployments, enterprise agreements, security reviews, custom workflows and dedicated support from frontier AI companies. Smaller organisations often cannot.

Policymakers have taken note. IMDA has announced an expansion of the Digital Enterprise Blueprint with two new partnerships. Grab will launch a programme targeting 10,000 F&B, e-commerce and retail SMEs – offering online training, masterclasses and a structured two-day AI course co-developed with the Singapore University of Technology and Design. RSM Stone Forest IT will offer phishing simulation exercises to 2,000 SMEs to strengthen cyber resilience.

IMDA, SkillsFuture Singapore and Workforce Singapore have also jointly released the AI for Enterprise Impact Playbook to help businesses identify where to start and what support to access.

But knowing where to start is a different problem from being structurally able to deploy AI safely.

What SMEs need to operationalise AI
For SMEs to be AI-native, they need to operationalise AI safely inside the work that matters most.

First, SMEs need hardware that supports local AI deployment. Running advanced AI models used to require specialised infrastructure, including cloud-based graphics processing unit clusters and dedicated engineering support.

With recent improvements in hardware, it has become feasible to run enterprise AI models directly on local devices. Schools, clinics and accounting firms can now operate AI systems without specialised engineering teams.

Second, software must become standardised systems. Many on-premise AI systems today are custom integration projects which can be expensive and technically complex.

Increasingly, companies will expect AI platforms that work more like software-as-a-service products – configurable, maintainable and operational out of the box.

Third, AI experiences must be designed for domain experts, not only technologists.

The most valuable institutional knowledge does not reside with software engineers. It resides with the domain experts – those with the tacit knowledge, contextual awareness and intuition to operationalise enterprise AI.

They could be teachers who understand student behaviour, clinicians who understand patient workflows, and lawyers who understand legal risk.

The next phase of enterprise AI adoption will therefore depend on whether domain experts themselves can configure, supervise and operationalise AI within the environments they understand best.

A teacher should be able to build an AI assistant for student support workflows. A clinic should be able to create a secure note-summarisation process. A law firm should be able to analyse case materials without compromising privileged information. A manufacturer should be able to interrogate operational data without exposing commercial secrets.

The organisations that succeed with AI will not necessarily be those with the largest engineering teams, but those that can most effectively embed institutional expertise into AI systems.

This matters not just for operational efficiency, but for building institutional knowledge as a long-term asset. With AI integration, workflows, client interactions and decision patterns accumulate over time into systems that continuously improve and remain fully owned by the organisation that created them.

More powerful AI models alone will not drive SME adoption. SMEs need systems that are safe, governable and affordable to operate.

That is the next frontier worth building towards.

• The writer is the co-founder of Estha, a no-code enterprise AI infrastructure platform that enables organisations and individuals to run private, secure AI agents locally on Apple Silicon devices