Why Singapore AI Pilots Stall Before Scale
Release date:
Short answer
New research from Omdia and HCLTech, reported by iTnews Asia, finds that Singapore enterprises are putting money into AI but struggling to turn it into consistent, enterprise-level results, with outcomes often staying localised. For an SME, the lesson is about scaling AI in Singapore: the blocker is usually not the AI model, it is fragmented data and systems that were never connected in the first place. You do not need a bigger AI budget before you fix that.
What happened
The iTnews Asia piece reports on a joint study by Omdia and HCLTech. Its finding: Singapore enterprises are investing heavily in AI, but legacy technology, fragmented data and weak governance are stalling results. Most teams are not short of good ideas. They cannot bridge the gap between a successful test run and a full rollout, so AI activity stays localised rather than becoming a consistent, enterprise-level outcome. The report frames the fix around modernising legacy applications and data pipelines, building governance controls, and choosing a tech stack without getting lost in vendor noise. It is partner content, so read it as a vendor's view of a research finding rather than independent journalism. The finding itself is the part worth your attention.
Why it matters for an SME in Singapore or Southeast Asia
You may read "enterprise AI" and assume it does not apply to a 30 or 80-person firm. The pattern does. The scale is just smaller.
A mid-size Singapore business rarely has a legacy mainframe. It has something more common: a CRM that sales half-uses, an ERP or accounting system that finance trusts, a stock or job-tracking spreadsheet that operations maintains by hand, and a WhatsApp group where the real decisions get made. Each tool holds part of the truth. None of them holds all of it.
That is the same fragmentation the research describes, and it produces the same result. An AI assistant built on top of it reads partial data, gives partial answers, and the team stops trusting it. The pilot looks like a success in a demo and dies in week three. We have written before about why AI on scattered data mostly automates the mess, and this study is a regional data point for that view.
The second thing the research names is governance. For an SME that is not a compliance department. It is knowing who can see customer data, which AI outputs a person must check before they go out, and what happens when an assistant is wrong. Skip that and the first bad answer can cost you a client's trust.
What to do about it
Before you spend on another AI pilot, work through this in order.
- Map where the truth lives. For one process, say quote-to-cash or order-to-delivery, list every system and spreadsheet involved and who updates each one. You will usually find two or three places where the same fact is keyed twice.
- Connect before you add intelligence. Get the CRM, the ERP and the operational spreadsheet talking, so one record is the record. That is the work behind an AI-ready CRM, and it is unglamorous.
- Replace the manual report with a live view. If your monthly numbers are assembled by hand, the underlying data is not connected enough for AI to reason over. Live dashboards are the cheapest test of whether your data is ready.
- Automate the re-keying first. The person copying order details from email into the ERP is the clearest sign of a gap. Close it with workflow automation before you automate anything that needs judgement.
- Then pick one AI use case with a person in the loop. A support assistant that drafts replies for a human to send, or a lead-scoring aid a salesperson reviews, is a safer first step than an autonomous agent.
- Write the governance down. One page: what data the AI can see, who checks its output, who owns a wrong answer. That is enough for most SMEs.
If you want a second opinion on where your own gap is, book a discovery call and we will look at your current operations before scoping anything.
Where it does not apply, or what to be careful of
A few honest caveats.
- This is partner content from a vendor that sells modernisation services. The finding lines up with what we see, but treat the framing with the usual caution. The study is not a reason to buy anything in particular.
- Not every business needs AI yet. If your operations run on one well-used system and a clean process, an off-the-shelf tool may be the whole answer. Custom work is not always justified.
- Replacing a working ERP is slow and expensive. Sometimes an integration layer or a cleaned-up data pipeline is enough. We look at what can be connected before recommending a replacement, and a custom ERP build is the exception, not the default.
- AI will not remove the need for people. Results depend on how clean and connected your data is, and someone still has to check the output. Do not plan around headcount savings you have not tested.
- "Governance" can become bureaucracy. For an SME, a short written policy beats a committee. If a framework is slowing you down more than it protects you, it is too heavy.
Takeaway
The Omdia and HCLTech research says what we keep seeing: AI stalls when the systems underneath it are not connected. The fix is not a bigger model or a longer pilot. It is getting your CRM, ERP, spreadsheets and reporting onto one source of truth, then layering AI on top of data you trust. If you are weighing up where to start, our piece on digital transformation versus AI transformation sets out the order, and our case studies show what connected operations look like in practice, including an AI support and lead system built on a connected base.
Frequently asked questions
Does this research mean my company should pause AI projects?
No. It means the order matters. If your CRM, ERP, finance and operations data sit in separate tools, an AI assistant has nothing reliable to read. Connect the systems first, then add AI on top of data you trust.
We are a 40-person firm. Is this enterprise problem relevant to us?
Yes, in a simpler form. A 40-person firm rarely has a legacy mainframe, but it often has five spreadsheets, a WhatsApp group and two systems that do not talk. That is the same fragmentation at smaller scale, and it stalls AI in the same way.
How do we know if we are ready for AI?
A practical test: can you answer a basic question about a customer or an order from one screen, without asking someone to check another file? If not, the data is not connected enough for AI to be useful.
Do we need to replace our legacy systems to scale AI?
Not always. Sometimes an integration layer or a cleaner data pipeline is enough. Replacing a working ERP is expensive and slow, so we look at what can be connected before recommending a replacement.
What does governance mean for an SME, practically?
It means knowing who can see what data, which AI outputs a person must check, and what happens when an assistant gives a wrong answer. For most SMEs that is a short written policy, not a department.
