- Start with repetitive tasks that use data that’s already digital: quote requests, invoices, customer questions, monthly figures.
- Plug automation into your current tools (Zalo, POS, MISA, Excel, bank) instead of replacing everything.
- A first batch of 1 to 3 actions, measured before and after. We expand only if the gain is proven.
Where do SMEs in Vietnam stand?
Your employees already use AI. In 2024, a Microsoft survey found that 88% of office workers in Vietnam used generative AI at work, versus 75% worldwide, and that 89% of Vietnamese business leaders saw AI as necessary to stay competitive (Viet Nam News).
Businesses are moving more slowly. According to research by Strand Partners for AWS, 18% of Vietnamese businesses had adopted AI in 2025 (AWS). That figure is 26% in 2026, or about 245,000 businesses (TechRepublic).
The same 2026 study shows where things get stuck: 61% of businesses using AI are still experimenting, only 13% regularly measure what it brings them, and 54% cite a lack of skills as a barrier.
So the problem isn’t access to tools. It’s method: knowing what to automate, in what order, and how to prove the gain.
What to automate first
The best first projects aren’t the most spectacular. They’re repetitive tasks that come up every day, follow known rules and use information that’s already digital. In an SME in Vietnam, it’s almost always the same ones:
| Department | Today | Once automated |
|---|---|---|
| Sales | Quote requests come in through Zalo, Messenger, email and the website. Some go unanswered. | All requests in a single tracker, quotes sent within 24 hours, automatic follow-ups. |
| Accounting | E-invoices, bank statements, Excel files and payment confirmations on Zalo, all re-entered by hand. | No invoice entered by hand, payments reconciled, automatic reminders for unpaid invoices. |
| Customer service | The same questions come back every day, on every channel, including in the evening. | An assistant answers common questions from your documents, and passes complex cases to the team. |
| Management | The month’s figures arrive around the 15th, after exports and copy-pasting. | A dashboard that brings together sales, accounting, Excel and bank data, up to date from the first day of the month, with alerts. |
These four cases have one thing in common: they don’t require changing your trade or your software. They connect information that already exists but lives in separate tools.
How to spot the right tasks: 4 questions
Before talking tools, ask these four questions about every task that feels heavy:
- How many times a week? A daily task pays off more than a monthly one, even if it’s shorter.
- How many people spend time on it? Add up the hours of everyone who touches it, not just the person who finishes it.
- Is the information already digital? An XML e-invoice, a Zalo message, a POS export or an Excel file can be processed with no manual entry. A paper notebook takes one more step.
- What does a mistake cost? A forgotten quote, an unpaid invoice nobody chased or a supplier price increase nobody noticed often costs more than the time spent.
Then rank the tasks on two criteria: impact (time, errors, money) and ease (available data, number of tools to connect). Pick the first batch from the top right: high impact, simple implementation.
Connect before you replace
The temptation is to look for software that does everything. For an SME of 15 or 50 people, that’s rarely the right answer: switching is expensive, takes months and discourages teams.
It’s best to start from what you already use: Zalo, your point-of-sale software, MISA or another accounting tool, Excel, your bank and your e-invoices. Automation plugs into these tools, reads their data and connects them to each other.
Vietnam has an advantage here: since 2022, every business issues e-invoices as machine-readable XML files. That’s a valuable data source, and usually an untapped one (see our article on e-invoicing).
Two rules to keep in mind:
- people stay in control: every automation includes an approval step and a quality check, and no important decision is made by AI alone;
- your accounts and your data stay yours, exportable at any time, with no lock-in to the provider.
The 5-step approach
Most AI projects fail because they start with the technology. Here is the method we apply, for our clients and for ourselves:
- Understand. Observe the real work, role by role, and measure the “before” situation: time spent, errors, turnaround times.
- Prioritize. Rank the possible measures by impact and ease, and pick 1 to 3 for the first phase.
- Design. Map out the target process and have it approved by the people who will use it.
- Implement. In 2-week cycles, with a demo on your real data at each cycle, and a test period running in parallel with the old way of working.
- Measure. Compare against the “before” measurement in a gains report. We move to the next phase only if the gain is proven.
On the team side, name a point person in each department, run a short training for each role, and share the first gains with everyone. An automation nobody uses earns nothing.
What budget should you plan for?
Here are our public prices, to give you a sense of scale. The price of the audit or diagnostic is deducted from implementation if you sign within 30 days.
| Step | For whom | Price |
|---|---|---|
| Express AI Audit | Fewer than 20 employees · 2 days, including one day on site · 1 to 3 costed measures | 12M VND, fixed price |
| Full AI Diagnostic | 20 employees or more · about 2 weeks · 4 to 8 measures and a 6- to 12-month roadmap | 35M VND (20 to 40 employees), 55M VND (41 to 99), 85M VND (100 or more) |
| Implementation | In 4- to 8-week phases, quoted based on the audit · fixes included for 30 days after go-live | from 50M VND per phase |
| Monitoring and improvement | Maintenance, monitoring and a monthly gains report · 6 months minimum | 2.5M VND / month (Essential), 6M VND / month (Growth) |
| Team training | Role-specific workshops, using your own documents | 300,000 VND per person per hour |
Depending on the industry, the return-on-investment targets we set range from 1 to 4 months. These are typical targets, confirmed or adjusted during the audit, not promises.
Three real examples
A restaurant in Ho Chi Minh City
At Baieta, supplier invoices, payments and sales were tracked separately. The manager knew neither the real margin per dish nor the losses. The tool automatically reads the XML e-invoices, matches them against bank statements and POS sales, and calculates the cost of each recipe.
An international rankings group
Eduniversal, a client since 2010, tracks more than 100,000 programs and 10,000 schools worldwide, across thousands of websites that change constantly. A tool collects the information from the web, AI classifies each program by level and specialty, and everything is updated every month.
Our own company
We apply the method to ourselves first, and we measure it every month: preparing a project now takes 1 day instead of a week, accounting and HR save a week each month, and each developer saves 5 days a month thanks to AI that helps write, test and review code.
Pitfalls to avoid
- Starting with the tool. “We want a chatbot” is not a need. “We lose quote requests on weekends” is one.
- Automating everything at once. A first phase of 1 to 3 measures can be managed, measured and corrected. A big 12-month project, much less so.
- Not measuring “before.” Without a starting point, you can’t prove a gain, and projects stall at the trial stage.
- Forgetting the teams. The people doing the work today know where the real problems are. They should be involved from the observation stage.
- Letting everyone use AI with no rules. A customer file pasted into a free AI with a personal account has left the company. Business versions of the same models don’t use your data for training, and you choose where sensitive data lives. Personal Data Protection Law No. 91/2025/QH15 has applied since January 1, 2026.
Frequently asked questions
Is my business too small to automate?
No. The Express AI Audit is designed for businesses with fewer than 20 employees: 2 days, 12M VND, and 1 to 3 costed measures. A restaurant or a shop often sees the fastest gains, because the same tasks come back every day.
Do we need to change our software?
No. We start from what you already use: Zalo, point-of-sale software, MISA or another accounting tool, Excel, e-invoices. We connect before we replace.
How long before we see results?
An implementation batch takes 4 to 8 weeks, with a demo on your real data every 2 weeks. Depending on the industry, typical return on investment comes within 1 to 4 months.
Is our data safe?
It stays yours. We use the business versions of AI models, which don’t train on your data, and sensitive data can stay on servers in Vietnam, or on your own premises.
Sources
- Viet Nam News – Rate of using AI to work in Việt Nam is higher than world’s average: Microsoft (May 25, 2024)
- AWS – New AWS research shows strong AI adoption momentum in Vietnam (September 2025)
- TechRepublic – Vietnam’s AI Adoption Jumps to 26%, but Most Adopters Are Still Experimenting (October 2, 2026)
Founder of Silkwires, based in Ho Chi Minh City since 2010. A digital specialist since 2000, he leads the AI diagnostics and visibility projects.
