Walk into almost any small or medium enterprise (SME) in Malaysia today, and you will likely find someone using artificial intelligence. An employee might be using ChatGPT to draft a customer email. A marketer might be generating draft social media copy or doing basic online research.
While these personal efficiency wins are helpful, they barely scratch the surface of what modern AI technology can achieve. A study of 133 Malaysian SMEs by Ecosystm and Red Hat, supported by the National AI Office, found only 21% had progressed beyond AI pilot projects, while around 60% cited a lack of in-house technical expertise as a major obstacle.
Exabytes founder and CEO Chan Kee Siak says getting to agentic AI takes two things most of them are missing: reworked workflows and bosses who use AI themselves.
Founded in 2001, Exabytes began as a hosting company selling domain names, web space, and emails. Over the past two decades, the company has grown into a regional tech player with a team of over 330 people across Malaysia, Singapore, and Indonesia. With a customer base of 160,000 businesses, 90% of which are SMEs, Chan has a front-row seat to how smaller companies approach technology.
But the company is now putting more weight behind AI. At its GROW AI Summit in August, Exabytes said it was allocating more than 50% of its budget to AI, targeting one million businesses adopting AI by 2030. It also opened an HPC/AI colocation facility in Penang and signed an MoU with Aurora Mobile as it expands its AI infrastructure and solutions.
Chan says the shift is already showing up in Exabytes’ business, with AI and cloud now accounting for around 30% of its revenue.
“The AI space is still relatively new, and changes are happening almost every week,” Chan says. “Most employees have started using AI in a personal capacity to assist with basic workloads. That helps, but it doesn’t replace daily tasks. Today, agentic AI can actually do that. But moving to that next level requires two things. One is re-engineering existing workflows and two, strong support from top management.”
According to Chan, the core shift happening in technology right now is the move from simple conversational assistants to autonomous agentic AI. In his view, traditional chatbots are rigid and rule-based, often limiting users to predefined interactions, like, “press one for sales, or press two for support”.
“Agentic AI, by contrast, operates with natural language and autonomy. It can analyse context, access necessary internal applications, filter emails, update calendars, and complete tasks on behalf of employees,” Chan says.
A phased approach to agentic AI adoption
So why are so many Malaysian SMEs hesitating to make this leap? According to Chan, the roadblock is rarely the technology itself. Instead, it often comes down to leadership awareness and a lack of clear operational playbooks.
“First, the CEO must start using and learning AI,” says Chan. “If the driver of an organisation has no clue what AI can do, they will either hold misconceptions or block initiatives because they don’t see the power of the technology. Second, the organisation must have a clear objective. Whether you want to drive a 20% productivity improvement or completely transform a workflow, you need a defined goal.”
Chan points out that fear and uncertainty are also playing a role. However, hesitation during major technological shifts is nothing new.
“When the internet was first introduced, people worried about harmful information. When e-commerce came out, people worried about credit card security. When the cloud arrived, they worried about data privacy. Now, with AI, we see those same concerns again,” he says.
To help SMEs navigate this shift without feeling overwhelmed, Chan recommends a phased, step-by-step approach. Rather than trying to overhaul an entire enterprise overnight, businesses should think in terms of adoption levels.
“Level 1 implementation involves setting up standalone AI agents for isolated tasks,” Chan says. “These agents use basic, self-contained data like product catalogs or basic spreadsheets without needing deep integration into core systems.”
At deeper levels of adoption, AI agents can be integrated directly into enterprise resource planning (ERP), customer relationship management (CRM), or point-of-sale systems.
“This allows the AI to trigger real-time actions like issuing invoices or updating stock levels,” Chan says.
For business owners wondering where to start, Chan suggests beginning with a small workload or project where they can quickly see the benefits.
“Usually, we’ll recommend our clients to do a mapping. Let’s say, list out the top five tasks or workloads that they have to deal with on a day-to-day basis,” Chan says. “And among these top five workloads, do a rating or ranking. How much time or hours or resources does it consume every day? Then we’ll check each of them to see which ones potentially can be replaced by using AI or an AI agent.”
Consider customer inquiries over messaging platforms like WhatsApp. Responding to repetitive questions manually consumes massive amounts of staff time.
“We recently worked with a local durian merchant to address this bottleneck,” Chan says. “During peak durian season, the seller was inundated with constant WhatsApp messages asking about daily stock, price variations, and reservation slots. Handling these manually was tedious and exhausting for staff.”
The merchant received at least 50 inquiries a day, with that number rising to 200 or more during peak periods. Exabytes deployed a 24/7 AI agent capable of holding natural conversations, answering specific product queries, and managing customer bookings automatically. With the AI agent, customers received responses within seconds.
In January 2025, Exabytes introduced internal AI agents to handle its own customer service workload.
“Today, those agents autonomously resolve over 50% of basic inquiries regarding product details and billing,” Chan says. “This frees up our staff to build long-term client relationships and profile customer needs.”
Measuring ROI and managing security risks
When it comes to tracking return on investment, Chan suggests adapting metrics as the business matures.
“In the first three to six months, companies should track employee token consumption,” Chan says, referring to measuring whether team members are actively using the AI tools available to them.
“Once usage becomes routine, businesses can measure direct cost-versus-output ratios. In software engineering, for example, developers working alongside AI coding assistants can produce code two to ten times faster than those working manually.”
The productivity gains from AI can vary widely. A May 2026 METR survey of 349 technical workers found a median self-reported speed increase of 3x, but cautioned that such estimates may overstate actual gains and do not necessarily translate directly into staffing or cost savings.
Addressing data readiness, Chan cautions against waiting for perfect data before getting started.
“Start with Level 1 implementations that require minimal complex integration,” Chan says. “As you run those initial projects, you will naturally begin gathering and organising clean data for more advanced integrations six months or a year down the road.”
On security, Chan advises granting AI agents system access gradually. “Working with a proper vendor allows you to establish proper organisational guardrails and access controls. You don’t hand over sensitive financial accounts or master credentials on day one. You grant system access step-by-step as trust is established.”
Ultimately, Chan emphasises that local businesses must look past temporary hesitations and focus on where the technology is heading.
“The big wave that is coming is agentic AI, where agents actually perform tasks on our behalf rather than just answer questions. As with every past tech shift, the critical step right now is to stop fearing it, start trying, and build that understanding.”
Full article by Digital News Asia.
