AI in Vietnamese businesses: where things stand and where they should go

We guess that in September 2026 you have a slightly contradictory feeling whenever you hear about AI adoption in Vietnamese businesses. The news says Vietnam aims to be in the top 3 in ASEAN for AI by 2030, every conference says “AI first,” and at your company the young staff use ChatGPT to write emails very fast, yet quotes are still made by hand over 2 days, inventory is still counted in Excel, and existing customers still wait for a free employee to reply. You feel behind, and you also doubt the pretty numbers in the news. That feeling isn't wrong, because the real measurements disagree widely: Microsoft measured in 2024 that 88% of knowledge workers in Vietnam had used generative AI at work, while VECOM measured in 2025 that only 11% of Vietnamese businesses use AI tools. Almost everyone uses it as a person, but only one company in ten does. AI adoption in Vietnamese businesses sits exactly in the gap between those two numbers.

In this article we answer 3 questions, using real measurements and what we see at the MONA consulting desk, where marketing brings in over 4,000 leads a month and where, since early 2025, clients ask about “doing AI” more than “building a website.” Question one: where do the country and your company stand? Question two: for each company size, how far is worth going in 12 months? Question three: if you do it right, what numbers can you achieve, and what can't be achieved no matter how hard you try? Three questions, one article.

Where MONA clients stand with AI: 8 parts trying it, 2 parts really using it

We have no survey, only the notes we write after each consultation, and those notes say something fairly consistent: of every 10 clients who come asking about AI, about 8 have already tried it themselves. ChatGPT, yes. Gemini, yes. Canva, yes. A fanpage chatbot at a few hundred thousand VND a month, yes. But only about 2 have integrated AI into an operating step with before-and-after numbers. Eight parts trying, two parts really using. One director of an electrical equipment distributor in Thu Duc sat counting on his phone for us and came up with exactly 6 AI apps, and he couldn't remember which were still valid or which had auto-renewed his card last month. That detail isn't special, except that it repeats in almost every meeting.

What we feel at the consulting desk matches the national numbers. A CPA Australia survey in August 2024 recorded nearly 8 in 10 Vietnamese businesses saying they had used AI in the past 12 months, above the survey average of 69%. At the same time, the Cisco AI Readiness Index, November 2024 edition, measured Vietnamese organizations with 500 or more staff and found only 22% in the group ready to deploy AI fully, down from 27% in 2023, and only 32% ranking AI as a top budget priority. Eight in ten say they use it, yet fewer than a quarter have the groundwork to use it through to the end. Those are two sides of the same story.

Over previous article we wrote that AI multiplies what a company already has, and this article goes one step further: if multiplication is right, what is the national number in front of the multiplication sign, and how big is it? To know that, you have to look at national numbers, not feelings.

What national numbers say: Vietnamese people use AI personally a lot, companies integrate it little

On the individual side, Vietnam's numbers are among the highest in the world, and we don't say that as praise. Microsoft Work Trend Index 2024 measured 88% of knowledge workers in Vietnam using generative AI at work, versus 75% globally, and 76% of Vietnamese leaders saying they won't hire people without AI skills. The 2026 edition of the same report, surveying 2,000 Vietnamese people, places 39% in the “Frontier Professionals” group who use AI most deeply, the highest in ASEAN and double the 16% global level. Google, Temasek and Bain's e-Conomy SEA 2025 also ranks Vietnam first in Southeast Asia for AI user levels, with 81% saying they use AI tools daily. All of these numbers measure one person, sitting at one machine, using AI for that person's own work.

On the company side, the numbers are very different. The VECOM E-commerce Index 2026, based on 2025 data, shows only 11% of surveyed businesses use AI tools, up from 7% the year before. In the same survey, 46% of businesses have a website, hovering around 43 to 46% for 5 years running; 24% use CRM software; 14% use ERP. That means more than half of businesses have no website, three quarters have no customer management software, and only one in ten uses AI. The VINASA Vietnam AI Report 2025, surveying nearly 500 businesses and organizations, adds a number that says more about depth than any percentage: 30% of units spend under 50 million VND a year on AI, less than 5 million VND a month, about the price of a personal software plan. The barriers they cite are a lack of skilled people at 45%, unclear legal frameworks at 30%, and data and compute limits at 23%.

Looking more broadly, the Statistics Office announced in January 2026 that the digital economy made up 14.02% of GDP in 2025, up from 12.87% in 2021, while Resolution 57 sets a target of at least 30% by 2030. As of December 31, 2024, the country had 940,078 active businesses, nearly 98% of them small and medium, and by October 2025 it had passed 1 million. The 11% using AI sits on top of that base of nearly 1 million businesses. Putting both sides together gives one short sentence: Vietnamese people already use AI, Vietnamese companies don't yet. The gap between those two numbers is where we see the money sitting.

We see that gap in our own clients, in a very concrete form: employees use AI through personal accounts, the boss doesn't know, and customer data leaves the company through those accounts every day. A small MIT NANDA survey in mid-2025 recorded employees at over 90% of companies regularly using personal AI, even though only 40% of companies bought an official plan, and we believe the number in Vietnam is even higher. So where exactly does AI sit inside a Vietnamese company?

AI sits in 3 places in Vietnamese companies, and is absent where the money is made

Visiting many companies, we see AI in exactly 3 places. The first is personal machines: writing emails, translating documents, summarizing contracts, making post images. This is genuinely useful, but it doesn't touch how the company operates, and when one person leaves, it leaves with them. The second is a bot on a fanpage or website, a cheap plan that answers template questions, can't read price lists or customer history, so when a customer asks something deeper the bot freezes and staff have to jump in. The third is the “islands” at large groups, with their own teams and budgets: Masan has WinCommerce running the WiNARE system that automates over 66% of ordering and saves about 300 billion VND a year, per Thuong hieu va Cong luan in September 2025; Techcombank with akaBot automated over 100 processes, freeing over 586,000 work hours. Impressive numbers, but those are 2 among nearly 1 million businesses.

The empty spot is the ground floor, where money moves every day: quotes, caring for existing customers, stock in and out, reconciling receivables, screening recruitment applications. Why hasn't AI reached there? Because of three reasons we hear again and again: data is scattered across Zalo, notebooks and Excel files, one copy per person; processes are drawn on paper by no one; and nobody is responsible when AI answers wrongly. Cisco measured in Vietnam in 2024: 75% of organizations handle data inconsistently, only 21% have fully centralized data, and 66% say a lack of skilled people is the biggest data barrier. AI can't plug into a place with no outlet.

A case we met last year says exactly this. A construction materials distributor has 25 salespeople, each using their own ChatGPT to write quotes, producing 25 different quote styles. For the same item, two customers got two different discount levels, and when they compared, they saw the gap and suspected the company of playing with prices. That company had AI in all 25 places, and not one of them made money. We had to pull it back into one quoting step with software, one price list and one discount rule, and only then integrate AI to draft quotes according to that list. It took 4 months, not 4 days. But there is one industry in Vietnam that went through this stage much earlier, and it's ours.

Software is ahead: Vietnamese developers are the 2nd heaviest AI users in the world

The Anthropic Economic Index, February 2026 edition, measuring data from November 2025, ranks Vietnam 2nd in the world for the share of AI use going to software development, at 42.1%, behind only India at 45.2% and ahead of Egypt at 39.2%. That means that of everything Vietnamese people hand to AI, nearly half is writing code. Google's DORA 2025 report, surveying 4,867 software professionals, records that 90% have used AI at work. This industry is ahead because it already has the 3 things AI needs: repetitive structured work, data that is code and documentation already clean, and results that can be checked immediately by running them. Other industries don't lack AI; they lack those 3 things.

The most real number we know about AI writing code comes from FPT. Mr. Do Cao Bao told CafeF in July 2024 about FPT's CodeVista tool: AI raised code-writing productivity by 30%, but whole-project productivity rose only about 10%, because the bottleneck of a software project is clarifying requirements with the client, not typing code. We read that number and nodded, because it's the same at MONA. The way MONA builds software is now a senior writes the spec clearly enough, AI generates the code, the senior reviews and catches errors, and the most time-consuming part is still sitting with the client to learn what they truly need. AI speeds up the middle. The two ends are still people.

The lesson for other industries is right there in FPT's number. AI enters fast and produces numbers in steps that have 3 things: lots of repetition, clean input data, and results that can be checked immediately. In a distribution company, the quoting step has all 3; negotiating a big contract does not. In a spa chain, appointment reminders have them; handling complaints does not yet. The owner's job is to find the step in their own company that has all 3, and the next question is how far to go once you've found it.

How far to go: one level per company size, measured by steps, not slogans

Our view is simple: the level a company “should reach” is measured by the number of steps integrated with AI and measured before and after, not by whether staff use ChatGPT or whether the company declares “AI first.” An 8-person company that integrates exactly 1 step with numbers has gone further than a group that buys 10 AI plans and can measure none. After two years sitting with clients, we put together the table below, and we use it to pull clients down a level when they ask for too much, or push them up a level when they're too timid.

Two numbers stand behind that table. One is from McKinsey State of AI 2026, a survey of 1,719 businesses in 97 countries: only 6% are truly high performers, and about 75% of that group redrew their workflows before integrating AI, versus about 25% of the rest. The “prerequisites” column in the table is exactly that workflow redrawing, written in the language of Vietnamese businesses. The other is the MIT NANDA 2025 survey, a small sample so we cite it cautiously: projects bought from outside partners reach production about 67% of the time, versus only about 33% for internal builds. So for companies under 300 people, we almost always advise against building an AI team yourself. Buy the step from people who have it built, and keep your own people for review.

We once had to pull down a level the owner of a 15-branch electronics retail chain: he came demanding “AI across the whole company” in 6 months, and when we looked at the data, the 15 branches ran 3 different sales software systems that hadn't even been merged, so we advised him to do exactly 3 steps first. Conversely, the owner of a 12-person print shop placed herself in the “not our turn yet” group, while her website already had over 2,000 price inquiries a month, meaning the chat step already had enough foundation to integrate right away. Same table, and each person reads a different level. Once the level is clear, it's time to pick the jobs. Not every job can be handed off.

3 jobs to hand to AI this year, and 3 not yet

The three jobs we advise handing over now share traits: lots of repetition, ready data, and errors that show up immediately and are cheap to fix. The first is chat and repeated Q&A on the website, fanpage and Zalo. An AI agent for chat, hay chatbot for customer care, reading the right price list and customer history, can answer most questions about opening hours, prices, stock and delivery time, and pass hard questions to a person. The level we see with clients is response time from a few hours down to under 1 minute, 24 hours a day. The second is quoting and drafting documents from templates using ready data, such as Gấu Báo Giá reading price lists and norms, then producing a draft, with a person reviewing and sending. The third is content, SEO, and existing-customer emails with a reviewer, such as Gấu Thức Khuya drafting emails from each customer's purchase history. Stanford AI Index 2026 compiles productivity gains by type of work at 14 to 15% in customer support, 26% in software development and 50% in marketing, in line with the three jobs above. If you want to see more common steps, we've listed them in the article 9 common AI applications in business.

Three jobs we don't advise handing fully to AI this year, and we say so with reasons, not fear. One is big-money decisions: approving receivables, closing big contract prices, granting credit limits, because the result can't be checked immediately and one mistake costs more than a year of savings. Two is legal and final contract text: AI drafts very well, Gấu Pháp Chế from us does exactly that drafting job, but the signer still has to read every clause. Three is decisions about people: AI can screen 159 applications down to 1 person , but who to hire, how to evaluate and who to let go is for a person to decide. The Law on Artificial Intelligence No. 134/2025/QH15, effective March 1, 2026, also places decisions like these in a higher-risk group, so keeping people there is both right for the job and right under the law.

One more note on how to assemble it, and we say it for maintenance reasons. The three jobs above are all high-traffic steps running day and night. If you assemble them yourself with drag-and-drop tools, the flow breaks when a third party changes its interface, one wrong setting stops the whole flow and nobody notices, and the more flows you have, the more people it takes to watch them. Drag-and-drop suits light personal work. A mission-critical step needs custom-built software load handling, data that stays in your own system, and a team responsible when something fails. Once the right jobs are handed off, the remaining question is how much you can achieve. We answer in numbers.

What's achievable: real numbers come from individual steps

Global numbers first, so you can set your expectations correctly. McKinsey 2026: 89% of businesses use AI in at least one step, 37% say AI contributes somewhat to pre-tax profit but mostly under 5%, and only 6% are the group where AI brings in 5% or more of profit. IBM surveyed about 2,000 CEOs in 2025: only 25% of AI projects reached their expected return on investment, only 16% scaled across the enterprise, and the average return on investment is about 5.9%. PwC asked 4,454 CEOs in January 2026: only 12% saw AI both cut costs and raise revenue in the past 12 months, and 56% have yet to see significant financial benefit. BCG in September 2025 split companies into 5% creating outsized value, 35% scaling up, and 60% gaining almost nothing despite spending. Four surveys, one answer: making company-wide money from AI within 1 year is for the minority. Setting expectations at company level uses the wrong unit, and the wrong unit gets everything wrong.

The right unit is the step, and at the step level the numbers look pretty good. The director of a pharmaceutical distributor with 30 salespeople integrated AI into the step of making price lists for pharmacies and measured for exactly 30 days: time to produce a price list dropped from 2 days to 2 hours, the number of price lists sent tripled, and nobody new was hired. The owner of a 40-worker mechanical workshop in Binh Duong spent 8 months laying groundwork, building a website and internal quoting software first, then integrating AI to read drawings and propose prices, and now price lists come out in 2 hours instead of 2 days. The owner of a 12-branch spa chain with 40,000 customer records integrated AI into appointment reminders and treatment suggestions, and saw the numbers after 3 weeks because 6 receptionists no longer made calls by hand. Three cases, three industries, one kind of number: hours, days, number of price lists, number of customers replying. None was measured by company-wide profit, because 12 months is not enough to measure that.

We'll also say what can't be achieved in 12 months, so you don't buy the wrong expectations. You can't cut half your staff, because the work AI takes on is the repetitive part, while reviewing, meeting customers and handling exceptions still need people. There's no such thing as “AI running the company by itself,” since even at FPT, a software company, whole-project productivity rose only 10% even though code was 30% faster. Company-wide profit doesn't show in year one, because McKinsey's 6% are places that have worked at it for 3 to 4 years. What you can achieve is 1 to 3 steps with numbers. That's enough to keep going next year, and how far next year goes we've drawn as three milestones.

12 months, 3 years, 2030: three milestones MONA sees as feasible for Vietnamese businesses

The 12-month milestone is groundwork plus 1 to 3 steps with numbers, as in the table above: a website that brings in customers, data in one place, clear processes, then integrating AI into the most repetitive step and measuring for 30 days, just as the 5 mistakes we've listed advise avoiding doing it backwards. The Binh Duong workshop went through this milestone in 8 months of groundwork and 4 months of integration, exactly 12 months. The 3-year milestone is 5 to 8 steps, with AI agents connected directly to a CRM or ERP, one person in charge of data, and the repetitive office workload dropping enough that the company grows orders without growing headcount in proportion. For companies of 50 to 300 people, we see this 3-year milestone as a real one, not a conference one, because over the past two years some clients have already made it halfway.

2030 is a national milestone, and we will say plainly what we think. Resolution 57 sets targets for 2030: a digital economy of at least 30% of GDP, R&D spending of 2% of GDP, Vietnam among the top 3 ASEAN countries in AI research and development, and support for at least 500,000 small and medium businesses to go digital. The national AI strategy from 2021 aims for 4th in ASEAN and 50th in the world, 10 reputable regional AI brands and 3 large national data centers. In August 2026, Prime Minister's Decision 1671 went further: at least 500,000 small and medium businesses, household businesses and cooperatives to be supported in adopting AI. The infrastructure is really running: FPT and NVIDIA have been building a 200 million USD AI factory since 2024, Viettel opened a 140MW data center in April 2025 and plans 24 data centers by 2030, and investment in Vietnamese AI startups rose from 10 million USD in 2023 to 130 million USD in 2025. Oxford Insights ranked Vietnam 45th out of 195 countries for government AI readiness in 2025, up from 51st.

What we see missing from the 2030 picture is the bottom layer. A digital economy at 14.02% of GDP in 2025 cannot more than double to 30% in 5 years on the back of a few conglomerates and a few AI factories, because nearly 98% of businesses are small and medium, and only 11% of them use AI tools while 46% have a website. Whether the top 3 ASEAN AI target is reached depends on whether that 98% moves from personal AI use to building AI into the workflow, and we do not believe any milestone holds without a foundation. If an AI program for small and medium businesses reaches 500,000 companies and each one manages to build AI into just 1 step with measurable results within 12 months, that is a bigger change than all the corporate islands combined. That is also where MONA stands.

How MONA brings clients up to that level

We do this by the 4-level table above, and we start with a 30-minute foundation review, usually at the client's office. In those 30 minutes we look at 3 things: how many customers the website brings in each month, how many places the customer data sits in, and which step of the process breaks. Many clients are surprised that we open no AI presentation at all, we just ask and take notes. After that session, the client knows which level they are at and which step is the first worth building AI into. Sometimes the answer is "not yet, build the website and consolidate data first", and we say so even if it costs us an AI contract right now. Foundation first, AI after.

When the foundation is ready, we build in custom-built AI agents for exactly one step: Gau Cuoi handles chat and leads 24 hours a day on the website and fanpage, Gau Bao Gia reads price lists and norms, Gau Thuc Khuya drafts follow-up emails to past customers based on purchase history, Gau Tuyen Dung screens resumes, and Gau Phap Che drafts contracts. Each one stands on a step that has data, has a person who reviews, and has one number measured over 30 days, and only then does the next step open. We dare to work this way because MONA took the same path itself: founded in 2016, more than 14,000 projects, 85% of clients keep using our services, SEO has been the channel bringing in the most customers since 2019, and AI has been built into each of our own steps since a content-writing machine with human review all the way to how we build software, rather than standing alone in one corner. In August 2026 MONA was accepted into the OpenAI Select Partner program, and we think that is the result of building AI into real work rather than of buying many plans. AI in Vietnamese businesses, the way we do it, is a matter of one step, one number, one quarter at a time.

If you read the 4-level table and are not sure which level you are at, or you know your level but not what is missing to reach the next one, that 30-minute foundation review is for you. We do not sell a package. We answer exactly 3 questions: which level the company is at, what is missing to reach the next level, and which step is the first worth building AI into. Call 1900 636 648 and say "foundation review for AI" and we will schedule it. And if you would rather go a quarter on your own first and call afterwards, the way to do it is in the next section.

This quarter, start with one step and measure it with one number

In the first month, call a few department heads into a 1-hour meeting, list the 5 most repetitive steps, and pick the step with the most data rather than the one that sounds most impressive. Then do one very manual thing: take the 100 most recent customer records and count how many have a correct phone number and how many record purchase history. If fewer than 70% are clean, this month is only for cleaning data and nothing gets built in. In the second month, build AI into exactly that step. For a company under 300 people, buy from a partner with a ready-made product instead of building your own, and name exactly one person responsible for reviewing the AI's output every day. In the third month, measure for 30 days with a single number, such as hours to produce a price list, the share of past customers who reply, or the number of repeat orders, then decide whether to open the next step or stop. One step, one number, one quarter.

In the first quarter, AI in Vietnamese businesses tends to fail in two places, and we see it again and again. The first is letting staff keep using AI through personal accounts with customer data, which both leaks data and gathers nothing for the company. The AI law that took effect in March 2026 also begins to put responsibility on AI users inside organizations, so this is no longer a small matter. The second is buying AI before having data, which repeats exactly what the dental clinic owner in the previous article did: installed a chatbot for 2 weeks and then switched it off. Recall the number at the start of this article: 8 trials, 2 in real use, and what separates those 2 from the other 8 is clean data plus one person who is accountable.

If you want to know which step in your company already has enough data to build AI in right away, send a few lines describing the step that is drowning in work to info@themona.global, or call 1900 636 648. Within 2 working days, a MONA engineer answers exactly 3 questions: whether this step can take AI yet, what data to gather first, and if you go ahead, how long until you see a number, such as 2 days down to 2 hours like the pharmaceutical distributor, or 3 weeks to see a number like the spa chain owner. If it cannot be done yet, we say not yet, along with what to do first. We do not take work blindly.

Frequently asked questions about AI adoption in Vietnamese businesses

How far have Vietnamese businesses adopted AI as of 2026?
Individuals use it a lot, companies integrate it very little. Microsoft measured 88% of knowledge workers in Vietnam using generative AI (2024), while VECOM measured only 11% of businesses using AI tools (2025 data), 46% with a website and 24% using a CRM. Cisco ranks 22% of large organizations in Vietnam as ready to deploy AI fully (2024).

How far should a company with under 10 people go with AI?
One step is enough. Usually that means chat and Q&A on the website, fanpage and Zalo using a ready-made package from a partner, provided you already have a channel bringing in customers. Measure it by response time to customers and the number of replies per day, run it for 30 days, then consider a second step.

How much does using AI in a business cut costs?
Per step it's measurable: Stanford AI Index 2026 compiles productivity gains of 14 to 15% in customer support, 26% in software development and 50% in marketing. Company-wide it's rare: in McKinsey 2026, only 6% of businesses get 5% or more of pre-tax profit from AI, and IBM measures an average return on investment of about 5.9%.

Will Vietnam reach the goal of top 3 in ASEAN for AI by 2030?
Infrastructure and policy are heading the right way: the AI law in effect from March 2026, the FPT and NVIDIA AI factory, Viettel's 140MW data center, and AI startup funding up to 130 million USD in 2025. The deciding part sits with the nearly 98% of small and medium businesses, where only 11% use AI. When that layer moves, the milestone follows.

Should you build your own AI team or hire a partner?
Under 300 people, buy from a partner who has it built and keep your own people for review: the MIT NANDA 2025 survey records partner-bought projects reaching production about 67% of the time, versus about 33% for internal builds. Only above 300 people, with years of data and a long-term budget, should you consider your own team.

How far Vietnam gets depends on whether the 98% of small businesses lay the groundwork

Pulling the three answers together. How far we have come: Vietnamese people are among the heaviest personal AI users in the world, yet only 11% of Vietnamese companies build AI into their work, and AI sits on personal machines, in fanpage bots and on a few corporate islands, not yet on the ground floor where money flows. How far we should go: one level per company size, from 1 step for a shop under 10 people to 4 to 6 steps connected to a CRM for a company of 50 to 300, measured by the number of steps with real numbers rather than by slogans. How far we can get: within 12 months it is the numbers of individual steps, 2 days down to 2 hours, 3 weeks to see a number, 3 times the price lists without hiring more; company-wide profit is a 3-year matter, and globally only 6% have managed it.

"How far we have come" is a question for the whole country, while "how far we should go" and "how far we can get" are questions for each company, and the two meet at exactly one point: nearly 98% of Vietnamese businesses are small and medium, so Vietnam's ranking in 2030 will be decided by that layer, not by a few islands. That layer rises on a foundation: a website that brings in customers, data in one place, a clear process, and only then AI. AI transformation in Vietnam, if it happens, will take place on the ground floor of hundreds of thousands of small companies, one step each, and MONA chooses to stand on that floor. One step with a real number beats ten plans nobody uses. As for why this journey is mandatory rather than optional like the 2020 digital transformation, we wrote about it in the article you could miss digital transformation, but AI transformation is mandatory.