Key Takeaways
- DBS Hong Kong's November 2026 survey: 38% of SMEs already use AI or digital tools, another 38% are exploring.
- The top barrier is a lack of skills or expertise (41%), followed by implementation and running costs (38%).
- Data security, system integration and unclear returns tie at 35% each.
- Every barrier has a cheap first move: one task, one month, one number to measure.
It is 6 p.m. in a 12-person trading office in Kwun Tong. Thirty quotation requests sit in the inbox, and the clerk who handles them will spend about six hours turning them around tomorrow.
The owner knows AI could help. The question is why he still has not started. A new survey of 168 Hong Kong SMEs gives a clear answer: the same five roadblocks stop almost everyone, and none of them needs a big budget to clear.
What stops Hong Kong SMEs from using AI?
According to DBS Hong Kong's survey released on 12 November 2026, the five biggest barriers are a lack of skilled talent or expertise (41%), high implementation and maintenance costs (38%), and, tied at 35%, data security worries, system integration and uncertain return on investment.
The DBS Hong Kong survey polled 168 SME owners and senior decision-makers between 6 August and 7 September 2026. It found that 38% have fully or partly built AI and digital tools into their operations, while another 38% are exploring or planning to.
So most owners are not saying no. They are stuck at the same five points. Save this table:
| Rank | Barrier | Share of SMEs | First move |
|---|---|---|---|
| 1 | No skills or expertise | 41% | Start with one task, not one tool |
| 2 | Implementation and running cost | 38% | Price it per task, check subsidies |
| 3= | Data security | 35% | Write a one-page data rule |
| 3= | System integration | 35% | Start where data already sits |
| 3= | Uncertain return | 35% | Pick one number before you start |
Why is a lack of skills the number one barrier?
Small teams have nobody whose job is to set up AI, so the work lands on the owner, who has the least time. The fix is not hiring an AI specialist. It is narrowing the job to one repetitive task that a supplier or one staff member can set up and own.
Most failed starts look the same: someone signs up for a general chatbot, tries it on ten different jobs, and gives up after two weeks. Nobody learned anything, because nothing was measured.
A better start is one task with a clear volume, such as answering repeat WhatsApp questions about delivery times, or drafting the first version of a quotation. One person owns it, and checks the output every day for the first month.
How much does AI really cost a small business?
For a small firm, the useful comparison is not the software price but the staff hours it replaces. Count the hours a repetitive task takes each month, multiply by the hourly cost, and compare that with the monthly fee. If the fee is lower and quality holds, it pays.
Here is a floor calculation. Hong Kong's statutory minimum wage is HK$43.1 per hour from 1 May 2026. If a clerk spends 2 hours a day on repeat enquiries, that is 2 × 22 working days × HK$43.1 = at least HK$1,896 a month, before MPF, leave and the fact that most clerks earn well above the minimum.
Subsidies can also cut the bill. The government's Digital Transformation Support Pilot Programme matches spending 1:1 up to HK$50,000 per business for eligible food and beverage, retail, tourism and personal services firms with fewer than 50 full-time staff, but only for solutions on its approved list.
Is it safe to put customer data into AI tools?
It can be, if you control which data goes in and where it is stored. The Privacy Commissioner's guidance on staff use of generative AI asks firms to set clear rules on what may be entered, which tools are allowed and who reviews the output.
The PCPD's checklist for employees' use of generative AI is short enough to read in one sitting. For a small firm, it comes down to three rules:
- No ID numbers, bank details or medical information in free consumer AI tools
- Use only tools the company has approved, on company accounts
- Know where the data is stored and whether it can be deleted
What if AI cannot connect to my systems, or the return is unclear?
Start with work that already happens in tools AI connects to easily, such as WhatsApp, email and shared spreadsheets, rather than an old accounting system. Then choose one number before you start, such as reply time or hours saved, so the return is measured, not guessed.
Integration and unclear returns tie with data security at 35% in the DBS survey, and they are linked. When a project tries to connect to everything at once, it takes months and nobody can say what it achieved.
A 30-day test fixes both. Week one: set up one task. Weeks two to four: track one number against last month. If the number has not moved, stop. This kind of task-level test is something every owner can explain to staff and to the bank.
Which barrier should you tackle first?
Tackle the barrier that blocks your first task, not the biggest one in the survey. For most small firms that is skills: once one task is chosen and owned, the cost, data and return questions become concrete and much easier to answer.
Being honest about limits matters too. If your team handles fewer than a dozen enquiries a day, a good template and a shared inbox may be enough for now. AI earns its keep where the same work repeats every day at volume.
Five roadblocks, one starting point
The DBS survey shows that most Hong Kong SMEs are not against AI. They lack a clear first step. Pick one repetitive task, set one data rule, measure one number for 30 days, and the five roadblocks shrink to a single decision you can make this month.
Empowering AI. Centered on you.
Start with one task, without the commitment
UD's AI Staff Solution sets up an AI employee for one job, such as customer service, admin or accounting, with HK$0 setup fee and data kept on Hong Kong servers. Sign up for free and a UD consultant will follow up for a free consultation, with no obligation to buy.
Frequently Asked Questions
If the AI makes a mistake, who is responsible?
Your business is still responsible to the customer, the same as with any staff member or supplier. That is why the first month should include a daily human check of the AI's output, and why your supplier agreement should state who fixes errors and how fast.
How much of my own time will the first month take?
Plan for a few hours in week one to explain the task and share examples, then around 15 minutes a day to review output. If a supplier needs much more of your time than that, the task is probably too broad.
How will my staff feel about an AI employee?
Staff usually worry about being replaced. Tell them early which repetitive task the AI takes over and what they will do instead, and let the person who did that task review the AI's work. They become its supervisor, not its competitor.
Can I stop at any time and take my data with me?
Ask this before you sign. A good agreement lets you cancel on short notice, export your data in a common format and have the supplier delete its copy. If a supplier cannot say where your data is stored, choose another.
Is the DBS survey representative of all Hong Kong SMEs?
It is a useful signal, not a census. DBS surveyed 168 SME owners and decision-makers from 6 August to 7 September 2026, so treat the percentages as a direction rather than exact figures for every industry.