Much of the global AI debate has centred on who can build the most powerful models. In Asia Pacific, the more immediate test is whether companies can make AI work inside the business.
Singapore shows both the scale of the opportunity and what it takes to capture it. Businesses in the city-state report spending an average of $18.9 million on AI last year and an average return of 16%, according to Oxford Economics research commissioned by SAP. Both figures are expected to rise sharply, with ROI projected to reach 29% and investment to grow by 38% within two years.
Yet, there is significant headroom to realise that potential. Seventy per cent of business leaders say they are unsure whether AI is delivering its full potential. Only 6% say they are fully prepared to deploy and scale AI agents. As investment accelerates, the opportunity now is to build out the data, workflows, skills, and governance that enable it to deliver.
Inside the workflow
Some of the biggest gains from enterprise AI are waiting in the inner workings of a company. Finance, HR, procurement, supply chains and customer operations are where spending is approved, people are managed, goods are moved, and customers are served. These functions run on processes and data, and they have significant untapped potential when AI is embedded rather than layered on top.
This is already happening across Asia Pacific. Darussalam Assets, for example, a company that owns and manages several of Brunei’s government-linked companies, used generative AI to speed up recruitment. By automating job descriptions, resume screening and candidate feedback, it cut hiring timelines from months to weeks.
In this case, AI is not being used as a separate tool that employees occasionally consult. It is being built into how decisions are made, how processes move and how work gets completed. This will be key as AI agents take on larger parts of enterprise workflows.
Much of today’s AI value comes from improving decisions, generating content or removing manual steps. AI agents go further. They can plan, act, call systems, trigger workflows and complete tasks across business functions. To help companies move from AI-assisted work to agentic operations, SAP introduced its Autonomous Enterprise vision at this year’s Sapphire conference. It introduced more than 50 domain-specific AI assistants across finance, supply chain, procurement, HR, and customer operations, designed to orchestrate specialised agents that execute processes end-to-end.
Data comes first
Understanding what unlocks the full value of AI in organisations requires looking beneath the AI model. The more important questions are: what data the model can use, how that data is structured, and whether the business trusts the system enough to act on it.
The same Oxford Economics research reveals that 58% of Singapore businesses lack confidence in their ability to integrate and share data across business functions — making data integration one of the clearest opportunities for companies looking to scale AI agents. For companies trying to scale AI agents, that is a fundamental constraint. ERP systems hold decades of process knowledge and transactional history, which is exactly the context an agent needs to act with accuracy. However, that data was built for human processes, not machine execution. It sits in functional silos, is inconsistently structured, and is governed by frameworks that predate autonomous systems.
PT Bank Danamon in Indonesia shows that laying the right foundations can unlock AI’s full business value. To bring AI into HR operations for more than 8,000 employees across 856 branches, the bank first had to unify employee data on a single cloud platform.
Manik Saha, managing director of SAP Labs East Asia, says, “A pilot can succeed even when the underlying business process is still fragmented. Once companies try to move from AI generating insights to AI taking action inside finance, HR or procurement workflows, they start discovering gaps that were previously hidden – inconsistent data definitions, unclear process ownership, missing auditability, and more. That is why scaling enterprise AI is ultimately less about the model itself and more about operational readiness.”
This is the part of AI transformation that companies often underestimate. A model can generate an answer in seconds and the businesses investing in clean, connected and well-governed data, which could take years depending on the size and legacy of the company, are the ones positioning themselves to act on those answers with confidence.
As AI agents become more autonomous, the stakes rise. A chatbot that gives a weak answer can be corrected. An agent acting on trusted data can approve, reconcile and escalate with speed and accuracy. That is why the next phase of enterprise AI will reward companies that invest in trusted data, clear accountability and systems that can explain what they did and why. For that reason, the next phase of enterprise AI will be less forgiving than the first. It will demand trusted data, clear accountability and systems that can explain what they did and why.
Why Asia Pacific cannot only import AI
If AI is to work across Asia Pacific, it has to understand the region’s business realities. This includes multilingual documents, cross-border operations, local compliance requirements, uneven digital maturity and industry-specific workflows that do not always match the assumptions of Western enterprise systems.
A model trained mainly on Western enterprise data, optimised for English-language documents and Western compliance frameworks, may perform well in a demo but struggle inside a regional business. To be useful, AI has to be engineered around how companies in Asia Pacific actually operate, not simply localised after the fact.
Regional AI capacity, therefore, matters. Asia Pacific cannot be treated only as a market for technology built elsewhere. If the region is where the complexity sits, it also needs to be where part of the product thinking, testing and engineering happens.
SAP Labs East Asia, spanning Singapore, Vietnam, Japan and South Korea, is one example of how that role is changing. The Singapore operation has nearly tripled in size to around 400 employees and filed close to 160 AI-related patents locally. SAP has also planned a €150 million investment in its Vietnam Labs location, signalling that its regional engineering footprint extends beyond a single hub.
The clearest global output from this base is SAP Document AI, a Singapore-led product built for the document complexity of Asian enterprises and now used by customers globally. “One of the things we learned early building SAP Document AI in Singapore, is that enterprise documents across the Asia Pacific are rarely standardised. A single workflow may involve multiple languages, different regulatory requirements, handwritten inputs, non-standard invoice formats, or documents that vary significantly between markets and industries. Building the product here forced our engineering teams to design for real-world complexity from the beginning, which ultimately made the product more resilient and globally relevant,” says Saha.
Building for that level of variation is different from training AI on clean datasets. It requires direct exposure to how companies operate. Over the past five years, SAP has co-developed solutions with more than 80 Singapore enterprises, creating a link between local deployment experience and global product development.
That feedback loop is what the regional engineering capacity is for. It gives product teams access to the messy reality of business before that reality becomes a customer problem at scale.
Execution test
IDC forecasts that by 2030, half of the new economic value from digital businesses in Asia Pacific will come from organisations that are scaling AI capabilities today. This raises the stakes for companies moving from pilots to deployment. The winners will not be those that simply spend more on AI, but those that can make it work inside the systems, workflows and decisions that run the business.
“Most companies do not need more AI pilots. What they need is the ability to operationalise AI consistently across the business. That requires changes that are often less visible than the technology itself, such as modernising data foundations, redesigning workflows, strengthening governance, and building trust in how AI decisions are made. The companies that capture the most value from AI over the next decade will be the ones that integrate it into the core operating model of the business, rather than treating it as a standalone innovation initiative,” says Saha.
A finance AI agent matters only if the numbers are trusted. A procurement AI agent matters only if it understands policy, risk and cost. Once AI moves from advice to action, precision is what turns capability into real business value.
Asia Pacific’s complexity may become its advantage if AI can work across fragmented systems, multilingual documents and local rules. The model race will continue, but for most companies, the more urgent race is inside the business.
