The Pilot Phase Just Ended for Half of Southeast Asia
To scale AI beyond a pilot in APAC, business leaders need to move through three layers: from ad hoc testing, to a defined workflow with a named owner and a single measurable outcome, to a compounding system where AI outputs improve downstream decisions. The gap between these layers is almost always a people and process problem, not a technology problem.
A recent Stanford report confirmed what many of us in the region are seeing on the ground: nearly half of companies in Southeast Asia have moved beyond AI pilots. Indonesia, Thailand, and the Philippines are showing some of the strongest momentum globally in AI adoption.
That's a significant shift. Eighteen months ago, the conversation was about whether to start. Now the conversation is about what you actually have to show for it.
If you're in the half that's still running pilots — or you've deployed something but aren't sure it's working — this article is for you.
Why Most AI Deployments Stall After the Pilot
Here's what I see repeatedly: a business owner runs a ChatGPT experiment, someone on the team gets excited, they buy a tool, productivity jumps for two weeks, then it quietly fades back to old habits.
The pilot wasn't the problem. The absence of a handoff system was.
A pilot answers the question: Can AI do this task? But that's the wrong question to build a business around. The right question is: Where does this task sit in a workflow, who owns it, and what changes when AI handles it consistently?
Without answers to those three questions, you get a tool, not a capability. Building genuine AI capability across your organisation requires a more structured approach than most pilots provide.
The Three-Layer Test: Pilot, Process, or Compound?
I use a simple diagnostic with business owners to figure out where they actually are:
Layer 1 — Pilot: You've tested AI on a task. Someone on the team uses it sometimes. There's no formal process, no measurement, no owner.
Layer 2 — Process: AI is embedded in a defined workflow. There's a responsible person, a quality check, and it runs without you pushing it. You can measure time or cost saved.
Layer 3 — Compound: The AI system generates data or outputs that improve a downstream decision. Each cycle makes the next one more accurate or faster. This is where ROI starts to multiply.
Most APAC SMBs I work with are sitting at Layer 1 and calling it Layer 2. The gap between those two layers is almost always a people and process problem, not a technology problem.
What a Layer 2 Deployment Actually Looks Like
Take a mid-sized agency in Singapore with 10 staff. They had been using AI to draft customer update emails — useful, but ad hoc.
Moving to Layer 2 meant three things:
- Defined trigger: Every client status change above a certain threshold automatically queues a draft email in the ops team's dashboard.
- Defined owner: One ops executive reviews and sends — five minutes of work versus thirty.
- Defined measurement: They track average response lag before and after. Within six weeks, customer complaint tickets dropped by roughly 30%.
Nothing exotic. No custom model. Just a clear workflow with AI sitting in the right seat.
That's the difference between a pilot and a process. For businesses ready to go further, custom AI systems built around your specific workflows can accelerate the move to Layer 3 significantly.
The Cost Curve Is Working in Your Favour — But Only If You Move
Here's a number worth anchoring to: according to Stanford's AI Index, the cost of running GPT-3.5-level inference fell 280-fold between late 2022 and late 2024. That trajectory hasn't stopped.
What this means practically: the tools available to a 20-person company in Singapore today are roughly equivalent in capability to what only large enterprises could afford two years ago. The barrier is no longer cost or access.
The barrier is operational readiness — whether your team, your workflows, and your data are set up to actually use what's available.
Companies that are building Layer 2 and Layer 3 systems now are locking in a compounding advantage. Every month of delay isn't neutral — it's a month your competitors spend refining a system that gets better with use.
The Four Questions to Ask Before Your Next AI Investment
Before you buy another tool or expand an existing deployment, run through these:
1. What workflow does this sit in? Name the specific process — not a department, a process. "Customer onboarding" is too broad. "Drafting the welcome email after a contract is signed" is specific enough to build around.
2. Who owns the output? If the answer is "whoever has time," the system will collapse within a month. One named person, one accountable role.
3. How will you measure whether it's working? Pick one number. Time saved per week, error rate, response speed, cost per output. One number you can track without a data analyst.
4. What breaks if the AI gets it wrong? This is the risk filter. High-stakes outputs — legal documents, financial advice, client-facing commitments — need human review baked in. Low-stakes outputs — internal summaries, draft copy, research briefings — can run with lighter oversight. Know which category you're in before you remove the human from the loop.
The Honest Assessment Most Leaders Won't Do
I'll say what most AI consultants won't: the majority of AI spend in APAC SMBs right now is producing activity, not results.
Subscriptions are running. Tools are open in browser tabs. Someone on the team is a self-appointed AI champion. But when you ask the CEO what AI is actually contributing to revenue or cost reduction this quarter, the answer is vague.
That's not a criticism — it's where the market is. The pilot phase was about learning what's possible. The next phase is about building what's repeatable.
The companies that will dominate their categories in this region over the next three years aren't the ones that explored the most tools. They're the ones that went deep on two or three workflows, built real processes, and kept iterating. A good starting point is understanding the full range of AI consulting services that can accelerate that transition.
Start Here This Week
Pick one AI tool your team is already using. Map the workflow it touches. Find the gap between how it's used today and what a Layer 2 version would look like. Assign an owner. Set one measurement.
That's not a pilot. That's a foundation.
The rest compounds from there.
