AI Readiness Checklist for Thai Enterprises

A practical checklist to test your data, systems, people and PDPA governance before you start an AI project, with a simple way to score the results.

An AI readiness assessment in Thailand checks six things before you commit to a project: your data, your systems, your people, your governance under the PDPA, your use-case shortlist and your delivery model. If most of these are in place, you can pilot. If several are missing, fix those first, because that is where most AI projects fail.

This checklist is for executives, IT leads and operations heads at Thai enterprises. Use it in a one-hour workshop. Score each item as yes, partly or no, and treat every "no" as a task, not a verdict.

What an AI readiness assessment actually tests

Readiness is not a measure of how modern your company is. It is a measure of whether one specific AI use case can work in your environment. A bank with strong data but no clear owner for a chatbot is not ready. A small factory with clean sensor data and a motivated plant manager may be.

That is why the checklist below mixes technical and organizational questions. The technical ones tell you what is possible. The organizational ones tell you what will survive past the pilot.

Area The question it answers Typical gap
Data Do we have the data this use case needs, in usable form? Data exists but sits in scattered files
Systems Can AI reach our systems safely? No API, legacy ERP, no test environment
People Who will use, review and own the result? No named owner, no reviewers
Governance Are we allowed to do this with this data? PDPA basis unclear
Use cases Is there a process worth automating? A vague wish to "use AI"
Delivery Can we run it after launch? No one to monitor quality

Data readiness checklist

Data is the most common blocker, and it is rarely about volume. It is about whether the right records exist, can be found and can be trusted.

  • Can you name the data source for your top use case? "Customer emails" or "maintenance logs" is a good answer. "All our data" is not.
  • Is the data digital and accessible? Paper files and scanned PDFs can work with Thai-capable OCR, but quality varies, so test a sample.
  • Is it in the language you need? Check how much is Thai, English or mixed, and whether informal chat language appears.
  • Is there a single version of the truth? If three departments hold three customer lists, an AI will inherit the confusion.
  • Can someone tell you who owns each dataset? Without an owner, nobody can approve its use or fix errors.
  • Is quality checked at all? Even a simple monthly sample review counts as a yes.

If you answer "no" to the first two, stop and fix them before anything else. Our data and analytics practice often starts exactly here.

Systems and infrastructure checklist

AI that cannot connect to your systems ends up as a demo. The question is not whether you have modern software but whether there is a safe way in.

  1. Does the system the AI must read or update have an API, database access or an export? If the answer is a person copying values by hand, plan an integration step.
  2. Is there a test environment? Never point a pilot at live ERP or CRM data without one.
  3. Where can data be processed? Some organizations, especially in banking and government, require in-country hosting or on-premise deployment. Decide this early because it changes the architecture.
  4. Do you have identity and access control that the AI can respect? An internal assistant should only show a document to people who may already open it.
  5. Is there a way to log what the AI did? You will need it for audits and for debugging.

Government and regulated teams should look at our notes on sovereign AI infrastructure and the government and public sector page.

People and ownership checklist

Technology is usually the easier half. These questions decide whether people adopt the result.

  • Is there a business owner who feels the pain of the current process and can make decisions? An IT-only sponsor is a warning sign.
  • Are there two to five staff who can review AI output during the pilot and say what is wrong?
  • Does leadership agree on what success looks like, in words, before the pilot starts?
  • Have affected staff been told what changes and what does not? Fear of replacement quietly kills projects.
  • Is there someone who can keep the system running after the consultants leave? This can be an internal person or a support agreement, but it must be named.

Governance and PDPA checklist

Thailand's Personal Data Protection Act (PDPA) applies when AI touches personal data such as names, phone numbers, ID cards or health records. The regulator is the Personal Data Protection Committee (PDPC). Check its current guidance for your sector, and confirm your own position with counsel.

At readiness stage, you are not expected to have everything solved. You should be able to answer these:

  • Which personal data would this use case touch?
  • What is the lawful basis for using it, and is it documented?
  • Can you limit or mask the data the AI sees?
  • Where is the data stored and processed, and who are the processors?
  • Who reviews decisions that affect a person's money, health or rights?
  • Is there a way to explain what the system did and why?

Sector rules may add more. Banks, hospitals and listed companies often have their own regulators and internal policies, so list those too. Our healthcare and banking pages describe typical constraints in each sector.

Use-case readiness checklist

A use case is ready when it passes these tests:

  1. It is repetitive. It happens daily or weekly.
  2. Volume matters. Enough work flows through it to justify a project.
  3. "Correct" is defined. Two experts would agree on a right answer most of the time.
  4. You have a baseline. You can measure today's time per task, error rate or response time.
  5. A fallback exists. If the AI is unsure, a person takes over.

Use cases that fail test three or four are not bad ideas. They are ideas that need a definition or a measurement first. For examples of where to look, read our guide to AI automation in Thailand.

How to score your results

Count your answers across the six areas.

Result What it suggests What to do
Mostly yes in every area Ready for a pilot Pick one use case and scope a proof of concept
Strong in use cases and people, weak in data Ready to prepare Spend four to eight weeks on data cleanup and access first
Strong in data, weak in ownership Not ready, but close Name an owner and reviewers before starting any build
Many "no" answers Start with strategy Run an assessment and roadmap before any pilot

Do not average the scores. One red area can stop a project even when five are green. A missing lawful basis or a missing owner is a blocker no matter how good the data looks.

What can go wrong

Scoring yourself generously. Teams tend to answer "yes" when they mean "we think so." Ask for evidence: a data sample, a named person, a written policy.

Treating readiness as a one-time exercise. Systems, rules and staff change. Repeat the checklist before each new use case.

Letting the checklist replace a pilot. A checklist cannot tell you whether the model reads your Thai invoices correctly. Only a test on your own documents can.

Over-preparing. Some teams spend a year on data lakes and never start. If one process is ready, test it now while you fix the rest in parallel.

Skipping the people questions. A technically sound pilot with no reviewers and no owner usually fades after the demo.

How to start

Run the checklist in a single workshop with IT, operations, legal and one business owner in the room. Write down the evidence behind every "yes". Rank the "no" items by how much they block your top use case.

Then decide: pilot now, prepare for a few weeks, or build a roadmap first. Our AI strategy and consulting practice offers a structured assessment covering data readiness, a ranked use-case portfolio, PDPA and AI governance, and a pilot plan. If you would like a second opinion on your results, talk to our team.

Frequently asked questions

What is an AI readiness assessment?

It is a structured review of whether your data, systems, people, governance and use cases can support a specific AI project. It produces a list of gaps to close and a decision: pilot now, prepare first, or build a roadmap.

How long does an AI readiness assessment take?

A first self-check can be done in a one-hour workshop. A full assessment with interviews, data review and a roadmap typically takes several weeks, depending on how many departments and systems are involved.

Do we need perfect data before starting AI?

No. You need data that is good enough for one specific use case, with a named owner. Many teams pilot on one clean dataset while they improve the rest in parallel.

Does the PDPA stop us from using AI?

No, but it applies when AI processes personal data. You need a lawful basis, limited and controlled access, and records of processing. Check current guidance from Thailand's PDPC and confirm your position with counsel.

Who should take part in the assessment?

Include IT, operations, legal or compliance, and the business owner of the process. Add the staff who would review the AI's output, because they know where the process really breaks.

UPSPIRE Team

AI consulting and infrastructure, Bangkok. UPSPIRE's consultants and engineers build AI agents, Thai-language knowledge systems and sovereign AI infrastructure for enterprises in Thailand and Southeast Asia.

Ready to put this to work?

Turn this playbook into a working system

From a first agent to a national platform, the same team plans it, builds it and runs it with you.

Book a consultation