If you are choosing an AI development company in Thailand, a team outside Thailand can be a practical option when it works on Bangkok's clock: Hanoi and Bangkok are both on UTC+7, so the whole working day overlaps. As of October 2026, our published price for a first AI Pilot is USD 3,000–15,000 for 3–4 weeks of work. The part that needs real planning is legal, not logistical. If a team abroad receives or can access your customers' personal data, the Thai Personal Data Protection Act's cross-border transfer rules are likely to apply. Assess the actual data flow, then put a written processing agreement and a lawful transfer basis in place before real personal data is used.
This guide is written by NKKTech Global, an AI engineering company based in Hanoi, Vietnam, that contracts through its Singapore company. We have no office in Thailand and no published Thai case study. So this is a buyer's checklist covering cost, PDPA, Thai-language quality and how to compare partners, not a list of past Bangkok projects.
Key takeaways
- Bangkok and Hanoi both use Indochina Time (UTC+7) with no daylight saving, so a Thai team and a Vietnamese team share the same office hours all year.
- As of October 2026, our published prices are USD 3,000–15,000 for an AI Pilot (3–4 weeks), from USD 10,000 for a fixed-scope build, and from USD 8,000 a month for a dedicated team with a 3-month minimum. Prices exclude taxes.
- Under the Thai PDPA, a company that hands personal data to a vendor must have an agreement controlling the vendor's processing (Section 40), and transfers abroad must meet Section 28 or 29.
- Thai generally has no spaces between words, and some general-purpose models split Thai text into more tokens than comparable English. Search quality, cost and response length therefore need testing on your own Thai documents with the model you choose.
- A good partner tells you where data is processed (including the model provider), who reviews Thai output, and what you own at the end of the pilot.
Why the same time zone matters more than the flight
Most offshore advice is about managing the gap between two time zones. Bangkok and Hanoi have no gap: both are on Indochina Time, UTC+7, and neither changes clocks during the year (timeanddate). That changes how a project actually runs:
- Feedback can be handled the same day. A product owner in Bangkok can review a demo in the morning and discuss changes in the afternoon, without waiting for another team's morning.
- Daily stand-ups fit normal hours for both sides, with no early or late calls.
- Your operations staff can follow the pilot live. When an AI assistant gives a wrong answer at 11:00, both teams are within their working hours to look at the logs, subject to the response times you agree.
Two things still need planning. Public holidays differ: Songkran in mid-April affects the Thai side, while Vietnam's Lunar New Year (Tết), in late January or February, affects the engineering side. Put both in the timeline. Language matters too. Agree the working language at the start (with a team outside Thailand it is often English), and name early who on your side reviews Thai-language output, especially if the engineers are not native Thai speakers.
What AI development costs for a Thai company
Prices from any vendor depend on scope, data and integrations. These are our published ranges as of October 2026, in US dollars, so you can see how the pieces relate:
| Engagement | Published price (USD) | Typical duration | When it fits |
|---|---|---|---|
| AI Pilot | 3,000–15,000 fixed fee | 3–4 weeks | Testing one use case on your real data before a larger budget |
| AI automation build | 10,000–80,000 fixed scope | 6–12 weeks | A workflow such as document processing or CRM automation |
| AI SaaS product | 25,000–300,000 fixed scope | 20–40 weeks | A full product with LLM integration, retrieval and billing |
| Dedicated team | 8,000–50,000 per month | 3-month minimum | Ongoing work where the scope keeps changing |
According to our Thailand page, invoices can be issued in THB, USD or SGD. The full price list and payment terms are on our pricing page. Our prices exclude taxes and duties. Ask your accountant how Thai VAT and withholding tax apply when you pay a foreign service provider, because that changes your real cost.
Running costs come on top of the build. They include model usage (billed per token, a small chunk of text), hosting, storage and maintenance. If your data needs to stay in Thailand, check whether your cloud provider has a Thai region. AWS, for example, lists an Asia Pacific (Thailand) region. Then check whether your chosen model is offered there.
A local Bangkok agency is sometimes the better choice. That is likely when you need people on site every week, Thai-native UX writing, or a contract that procurement says must be with a Thai-registered company.
If you are unsure, start small. Our AI Pilot is designed to take one use case to a working pilot over 3–4 weeks for a fixed fee, using your real data where the PDPA arrangements allow it, or prepared test data where they do not. You own the code, and nothing larger is committed until it proves its value.
Thai PDPA when your AI partner is outside Thailand
The Personal Data Protection Act B.E. 2562 (2019) was published in the Royal Gazette on 27 May 2019 and has applied in full since 1 June 2022. It is overseen by the Personal Data Protection Committee (PDPC). Do not confuse it with Singapore's PDPA, a different law covered in our Singapore PDPA guide.
Two roles matter. The data controller decides why and how personal data is used, which is usually your company. The data processor handles the data on the controller's instructions, which is usually the development partner. The role of each cloud or model provider depends on its contract and how it uses the data, so assess each one separately.
| What the Act requires | Where (official Thai text) | What to agree with an AI partner abroad |
|---|---|---|
| The processor acts only on instructions, keeps security measures and records; the controller must arrange an agreement controlling the processor's work | Section 40 | A data processing agreement covering those duties, plus terms you negotiate, such as return or deletion of data at the end |
| The processor tells the controller about breaches | Section 40(2) | A deadline and contact for breach notice to you |
| The controller notifies the PDPC Office without delay and within 72 hours of becoming aware of a breach where feasible, unless it poses no risk to individuals; if the risk is high, data subjects must also be told | Section 37(4) | A tested incident runbook covering both teams |
| Transfers abroad need an adequate destination standard or one of the listed exceptions | Section 28 | Which ground applies to each destination, including the model provider's |
| Data protection policies within the same business group, reviewed and certified by the PDPC Office; or, failing that, appropriate safeguards with enforceable rights and effective legal remedies | Section 29 | For an outside vendor, usually safeguards such as contract clauses that follow the PDPC's published criteria |
| Sensitive data, such as health, religion or biometric data, generally needs explicit consent unless an exception applies | Section 26 | Whether the pilot can exclude this data entirely |
The Section 28 exceptions include complying with law, the data subject's consent after being told the destination's protection may be inadequate, and performing a contract with the data subject. The PDPC has since issued notifications with detailed criteria for transfers under Sections 28 and 29 (published in late 2023). This article has not been reviewed by a Thai lawyer, so read the current versions on the PDPC website with your DPO or counsel before relying on any transfer ground.
AI adds one more data flow that is easy to miss. If the system calls a hosted large language model, your prompts and documents go to that provider's servers as well as to the development team. Map both. For a pilot, a lower-risk route is often the simplest: start with documents that contain no personal data, or synthetic test data, while the agreements and transfer basis are settled. Pseudonymised data (names and IDs replaced) can still be personal data, so assess it rather than assume it is out of scope.
This is general information, not legal advice.
Thai-language AI – what to test before you commit
Thai brings a few engineering issues that are worth knowing about before you compare quotes:
- No spaces between words. Thai generally uses spaces between phrases or sentences rather than between words, so search and retrieval systems need proper Thai word segmentation. RAG (retrieval-augmented generation, where the model answers from your own documents) suffers when documents are split in the wrong places.
- Token cost. Some general-purpose model tokenisers break Thai into more tokens than comparable English content. Where that happens, cost per request rises and the model's context window fills faster. Measure token counts and charges with your chosen model on your real documents during the pilot.
- Register and politeness. Customers notice when a chatbot uses the wrong level of formality or polite particles. Write a short style guide and test against it.
- Mixed Thai and English. Product names, English terms and transliterated words appear in the middle of Thai sentences. Your test set should include them.
- Dates and scanned documents. Thai documents often use Buddhist Era years (2026 is B.E. 2569). OCR on scanned Thai text with stacked vowels and tone marks also needs its own accuracy check.
Thai-focused open models, such as Typhoon, are an option alongside the major commercial models, and the right choice depends on your data, hosting rules and budget. In either case, build a test set of real Thai questions or documents (sized to the task; a few dozen to a few hundred is a common starting range), have a native Thai reviewer on your side score the answers, and decide your pass mark before the build starts.
How to evaluate an AI development company in Thailand or abroad
Use the same questions for a Bangkok agency and an overseas team:
| Question | A strong answer looks like |
|---|---|
| Who signs the contract, and under which law? | A named legal entity and governing law. Ours is a Singapore-law MSA with NKKTech Global Pte. Ltd. (UEN 202445701K) |
| Where will our data be stored and processed? | Every location listed, including the model provider and any subcontractors |
| How will Thai-language quality be measured? | A test set, a reviewer, and a pass mark agreed in writing |
| What do we own at the end? | Ownership of the code stated in the contract, and the other deliverables (prompts, test sets, documentation) listed in the written scope |
| How are scope changes handled? | Quoted and approved in writing before work on them starts |
| What evidence can you show? | Published case studies with named results, and independent reviews |
On evidence: none of our published case studies is Thai, so judge the pattern rather than the geography. Within ASEAN, we built a RAG property search for a Singapore PropTech startup over 14 weeks. The published case study reports, for that project, 65% faster property discovery, 40% higher user engagement and zero hallucinated property details. For independent signals, we hold ISO 9001:2015 and ISO 22301:2019 certification and a 5.0 rating on Clutch from 10 reviews. Ask any vendor for the equivalent and check it yourself. Thai-language readers can find more on starting AI on a small budget.
Frequently asked questions
Is it legal for a Thai company to send customer data to an AI developer in Vietnam? It can be, if the transfer meets Section 28 or 29 of the Thai PDPA. That means an adequate destination standard, a listed exception such as informed consent, or appropriate safeguards that follow the PDPC's criteria. You also need a processing agreement under Section 40. Confirm the details with your DPO or legal counsel, because this is general information, not legal advice.
How much does it cost to hire an AI development company for a Thai business? It depends on scope. As of October 2026, our published AI Pilot range is USD 3,000–15,000 for 3–4 weeks; fixed-scope builds start from USD 10,000, and dedicated teams from USD 8,000 a month with a 3-month minimum. Add running costs such as model usage and hosting, plus any Thai taxes on paying a foreign provider.
Can AI chatbots understand Thai well enough for customer service? Often, yes, for well-defined questions, but quality varies by model and topic. Thai word segmentation, polite register and mixed Thai-English text are the usual weak points. Test any model on a set of your real customer questions, scored by a native Thai reviewer, before you decide.
Next step
Working on Bangkok time and want to test one AI use case before committing a larger budget? Book a discovery call in your own office hours. We will talk through one use case and your data, and send a fixed-scope proposal within 3 business days.
Senior engineers in Vietnam, a clearly defined scope, and a written proposal within 3 business days.
10 years of experience working with Japanese companies — two of those years based in Japan. Has run the business since 2018 (as NKKTech since 2025) with a senior-only engineering model.