- 13
- September
"Thailand AI Readiness 2026: 53.7% of Organizations Use AI, but 86.4% Cannot Measure Results" — the short answer is that NECTEC (the National Electronics and Computer Technology Center, part of NSTDA) surveyed 574 Thai organizations across the public and private sectors and found the country has cleared the "we have started using AI" hurdle, but is stuck at the "prove it worked" hurdle. This article breaks down the scores across all five readiness dimensions, the split by industry and organization size, and gives executives a checklist for locating their own organization inside these numbers.
In one line: 53.7% of Thai organizations already use AI, but average readiness sits at the "Aware" level, governance scores lowest at 28.9%, and 86.4% cannot measure what their AI adoption actually delivered.
What AI Readiness 2026 Is, and Who Ran It
AI Readiness 2026 is Thailand's national survey of organizational readiness to adopt artificial intelligence, conducted by NECTEC (National Electronics and Computer Technology Center) under NSTDA and published on 3 September 2026.
What separates it from the usual AI trend report is that it measures Thai organizations against each other, not against a global average that then has to be reinterpreted locally. The sample is 574 organizations from both the public and private sectors across 10 target industries — including manufacturing, education, healthcare, public services, energy and environment, finance, and security — scored across five dimensions.
If you read our earlier piece, Over 70% of Thai SMEs Now Use AI — So Why Doesn't It Show in the Numbers?, and felt that adoption figures and business results were not lining up, this survey is the structural explanation.
The Headline: More Than Half Are Using It, but Stuck at "Aware"
The top-line number is that 53.7% of Thai organizations already use AI. That sounds healthy — until you look at the maturity rating, where the national average sits at the "Aware" level. In plain executive language: they know it matters, they have run experiments, but it has not been built into how the work actually gets done.
| Readiness dimension | Score | How to read it |
|---|---|---|
| Data & infrastructure | 51.0% | Highest of the five — most organizations already have systems and data |
| People | 45.9% | Some staff can use it, but capability is not spread across the organization |
| Strategy & organizational capability | 41.8% | AI has not yet become a corporate objective with a named owner |
| Technology | 32.2% | The gap between "we have data" and "AI is wired into the workflow" |
| Governance | 28.9% | Lowest — policy, accountability and auditability have not kept pace |
The ordering deserves a second read. The highest-scoring dimension is data and infrastructure; the lowest is governance. That means the bottleneck for AI in Thai organizations is no longer purely technical — it is control and proof.
The Biggest Gap Is "We Cannot Measure It"
The harshest number in the survey is not about adoption. It is about measurement.
The number executives should remember: 86.4% of organizations cannot measure the results of their AI adoption. Only 13.6% have documented positive outcomes. When the next budget cycle arrives, the first group has nothing to answer the question "what did last year buy us?"
| Finding | Share of organizations | Consequence |
|---|---|---|
| Cannot measure AI outcomes | 86.4% | Cannot renew funding or scale up — there is no evidence to point at |
| Have documented positive outcomes | 13.6% | The only group that can argue for expansion with weight behind it |
| Have not implemented AI ethics | 73.4% | No line drawn between what AI may decide and what needs a human |
| Lack a risk assessment procedure | 46.2% | Aware of the risk, but with no repeatable process for handling it |
| Have integrated AI fully into strategy | 8.4% | Most efforts are still departmental projects, not a corporate agenda |
| Systematically link AI to employee competencies | 2.9% | Training happens as one-off events, disconnected from career paths |
Notice how the numbers slide from 86.4% down to 2.9% in a consistent direction. The more a practice is about making AI last — measuring it, setting ethics, assessing risk, tying it to strategy, tying it to people — the fewer Thai organizations do it. That is the same pattern we described in How to Use AI Safely — AI Governance Policies.
Who Is Ready and Who Is Not: By Industry and Size
The national average hides very large differences between groups.
| Group | Readiness / adoption | Observation |
|---|---|---|
| Banking | 70.7% | The only group reaching the "Ready" level |
| Healthcare & wellbeing | 53.1% | Above the national average, still short of ready |
| Public administration & services | 50.8% | Close behind — data exists, governance is the open question |
| Non-bank financial institutions | 22.8% | Almost 48 points behind banks, inside the same sector |
| Energy & environment | Lowest of the industry groups | Identified by NECTEC as the least ready sector |
| Large / medium / small organizations | 67.2% / 48.5% / 43.8% | Share already using AI — a gap of roughly 23 points top to bottom |
The most instructive pair is banks at 70.7% against non-bank financial institutions at 22.8%. Both sit in the financial sector and both handle sensitive data, yet they differ by nearly threefold. The cleanest explanation is that banks were already required to maintain risk frameworks, internal controls and audit trails long before AI arrived — so when AI arrived, there was somewhere to put it.
26% Have Started with Agentic AI — While Governance Scores 28.9%
Among organizations already using AI, 26.0% have begun implementing agentic AI, most heavily in research and development (21.4%) and in strategy work (20.8%).
Agentic AI differs from a chatbot in that it acts rather than merely answers — it calls other systems, writes data back, and chains multiple steps on its own. We covered the mechanics in What Is Agentic AI? The New Era of AI Working Autonomously in ERP.
Security warning: When 26% are deploying AI that takes actions on its own, while the national governance score is 28.9% and 46.2% still have no risk assessment process, a meaningful number of organizations are granting write access to a system whose boundaries and audit trail nobody defined first.
Why Governance Scores Lowest
In practice, AI governance is not one policy document. It is four questions that must be answerable every time someone asks them.
- Who approved this use? Which tasks may use AI, which may not, and who decides.
- What data can it see? The AI's data scope should equal the permissions of the person invoking it — not the permissions of the system administrator.
- What has it already done? Logs that show who asked, when, and what resulted.
- Who is accountable when it is wrong? The process owner is still a person, not a tool.
The first three are not policy questions — they are systems questions, and the answers usually live in the back-office system that holds the real data, not in the AI tool. They are the same questions organizations must already answer under Thailand's data protection law, as covered in PDPA and ERP — Managing Personal Data Legally.
Checklist: Where Does Your Organization Sit?
Use these questions for a quick self-assessment. Two or more "no" answers, and your organization is likely sitting with the national average.
| Dimension | Question you must be able to answer | Yes? |
|---|---|---|
| Strategy | Is AI in this year's corporate plan, with a named owner? | ☐ |
| Data | Does the data you would feed AI already live in a system with clear ownership and permissions? | ☐ |
| People | Can more than one person continue the work if the person who started it resigns? | ☐ |
| Technology | Does AI connect to the real workflow, or are files still copied by hand? | ☐ |
| Governance | Do you have logs showing who used AI on which data set? | ☐ |
| Measurement | Can you state how long the task took, and how many errors it produced, before AI was introduced? | ☐ |
That last row is what produces the 86.4% figure. Most organizations cannot measure AI outcomes not because they lack measurement skills, but because they never recorded the baseline. Once AI improves things, there is no way to prove by how much.
NECTEC's Three Policy Recommendations, Translated
The survey closes with three policy proposals. Read through an organizational lens, they mean:
- National AI Governance Pack — a baseline standards kit for organizations. Meaning: you do not have to write an AI policy from a blank page; adapting a central standard will be faster.
- Sectoral AI Readiness Clinics — industry-specific AI clinics. Meaning: hospitals, factories and government agencies face different problems and should not be handed one generic set of advice.
- AI Value Enabler — support for SMEs and mid-sized organizations, which matches the finding that small organizations use AI at 43.8% against 67.2% for large ones.
The Saeree ERP View: You Cannot Measure AI Without a Baseline
To be direct: this survey does not say "organizations need to buy an ERP." But the two lowest-scoring dimensions — governance at 28.9% and technology at 32.2% — are areas where a back-office system of record genuinely helps more than any single AI tool can.
Three things overlap in practice:
- Baselines for measurement. Closing cycle time, the number of purchase orders that had to be corrected after the fact, stockout rates — these already sit in the system. Record them before an AI project starts and you will not join the 86.4%. See Late Financial Closing — Root Causes and Solutions and Why Executives Never See the Reports They Need.
- Permissions and audit trails. An ERP has had per-user permissions and audit trails from day one. When AI is connected, what it can see and do should inherit those same permissions rather than open a separate, separately-audited channel.
- Connections with enforceable boundaries. Saeree ERP takes on work connecting AI assistants to business systems via MCP, where you define which tools the AI may call, what level of data it can see, and every call is logged. We are not citing any customer case in this article — we ran it on our own operations first, and only then opened it to outside work.
Our position is plain: the ERP is the source of truth; AI is the assistant. If the source of truth is unsettled, adding AI makes answers faster without making them more correct.
Conclusion
NECTEC's AI Readiness 2026 survey confirms with data what many organizations already sensed — Thailand is past the "not using AI yet" phase. This year's problem has moved to governance and proof.
| Start here | Not urgent yet |
|---|---|
| Record the baseline for the process AI will take over | Buying more AI tools before knowing how you will measure them |
| Define which decisions AI may make and which need human approval | Writing a complete AI policy before the central standards land |
| Tie AI's data access to existing user permissions | Letting AI write back into systems before an audit trail exists |
| Pick one process with a measurable outcome and finish it | Announcing an organization-wide programme all at once |
The 13.6% who documented positive outcomes are the ones who can ask for a bigger budget next year. The other 86.4% will have to restart the same conversation. The difference between those two groups begins with recording a number before you start — not with the technology.
Organizations that cannot measure their AI are not losing because the technology underperformed. They are losing because nobody wrote down how bad it was beforehand.
- The Saeree ERP team
References
- NECTEC, NSTDA — AI Readiness 2026 survey of Thai organizations (3 September 2026)
- National Electronics and Computer Technology Center (NECTEC), NSTDA
Verified 13 September 2026. Every figure in this article comes from the NECTEC survey; no estimates have been added by the author.
Can your organization actually measure its AI?
Talk to the Grand Linux Solution team about baselines, access permissions and audit trails — before the AI project starts, not after.
Request a Free DemoTel 02-347-7730 | sale@grandlinux.com
