New: a free AI proposal + interactive prototype for Malaysian SMEs this quarter.
AI & Automation

Hong Kong SMEs plan AI adoption, but only 23% have started: where to begin

AI is high on Hong Kong SME plans, yet a Dah Sing Bank 2026 survey found only 23% have adopted it. Here are three narrow first projects, a 30-day example plan and the personal data checks to make before customer data goes near an AI tool.

Most Hong Kong SME owners we speak to have already decided that AI belongs somewhere in their business. What they have not decided is where to start. That gap shows up in the numbers. Coverage of Hong Kong SME transformation plans by Human Resources Online puts AI near the top of owners' priorities, yet the Dah Sing Bank 2026 survey it cites found that only 23% of SMEs have adopted AI or generative AI so far.

So roughly three in four firms are still at the planning stage. This article skips the strategy talk. It sets out three first projects that a 20-person firm can scope in a month, the personal data points to check before customer information goes near an AI tool, and a simple way to choose between them.

A Hong Kong small business owner and a colleague planning together at a desk in a sunlit office
A Hong Kong office manager reviewing customer enquiries on a phone at a small trading company desk

Why so many firms are stuck at the planning stage

Most firms that have planned AI but not started are held back by scope, ownership or data concerns rather than budget alone. The project is too broad to price, nobody in a small team has time to own it, or the owner is unsure how customer data will be handled. Each has a practical fix. In more detail:

  • The project is too big. "Use AI across the business" is not a project. Nobody can price it, staff it or tell when it is finished.
  • Nobody owns it. In a 20-person firm the owner, the office manager and the one IT-minded staff member are all busy. AI becomes everyone's side task and no one's job.
  • Data worries. Owners are rightly cautious about putting customer names, phone numbers and invoices into tools they do not control.

Each of these has a practical answer. Pick one narrow task, give it a named owner and a 30-day window, and settle the data questions before any tool is switched on.

Three low-risk first projects

The best first project is one where the work is repetitive, the volume is steady, and a mistake is easy to spot and fix. These three meet that test for most service and trading SMEs in Hong Kong.

1. A WhatsApp enquiry agent

Many Hong Kong SMEs run sales and customer service through WhatsApp. Staff answer the same questions about opening hours, prices, stock and delivery many times a day, often after hours.

An AI agent connected to your WhatsApp Business account can answer those routine questions from an approved set of information, collect the details of a new enquiry, and hand anything unusual to a person. The key design choice is the hand-off: the agent should know what it must not answer, such as pricing exceptions, complaints or anything involving a refund.

What to measure: the share of enquiries answered without staff, the time to first reply, and how many conversations were handed over. Track these yourself for two weeks before launch so you have a baseline.

2. Invoice and document reading

Accounts staff in trading, distribution and professional services firms spend hours keying supplier invoices, delivery notes and receipts into accounting software. The documents arrive as PDFs, photos and email attachments, in English and Chinese.

Document AI reads those documents, pulls out the supplier, date, amounts and line items, and prepares them for checking. A person still reviews and approves each entry. The gain is in removing the typing, not the judgement.

What to measure: documents processed per week, time per document before and after, and the error rate found at review.

3. A weekly sales dashboard

Many owners still build their weekly picture from spreadsheets exported from several systems. A simple insights dashboard can pull sales, stock and receivables into one view that refreshes on its own, with a short plain-language summary of what changed.

This is the lowest-risk of the three because it reads data you already hold and does not talk to customers.

What to measure: hours spent preparing reports each week, and whether decisions such as reorders or chasing overdue invoices happen sooner.

A 30-day plan for a 20-person firm

The plan below is an illustrative example only, not a quote. Your own timeline will depend on your systems and data.

An operations manager pinning a weekly plan to a corkboard in a warehouse office
WeekExample activityOwner
1Choose one project, record the current baseline, list the data involvedOwner and office manager
2Agree data handling rules, prepare approved answers or sample documentsOffice manager
3Build and test with staff only, no customersSupplier and one staff tester
4Limited live use, daily review of errors and hand-offsOffice manager

At the end of the month you should have a clear answer to one question: did this save enough time, with few enough errors, to keep going? If yes, widen it. If no, you have learnt something useful at small cost and can try the next project.

Personal data checks before you start

Hong Kong's Personal Data (Privacy) Ordinance (PDPO) applies whenever an AI tool handles customer or staff personal data. The Office of the Privacy Commissioner for Personal Data has also published guidance on the use of AI. This is not legal advice, but these are the practical questions we work through with clients before any customer data goes near an AI tool.

Purpose and notice

  • Is using the data for this AI task consistent with the purpose you told customers when you collected it?
  • Does your privacy notice need updating to say an automated system may answer or process their enquiries?

Minimum data

  • Does the tool need full names, HKID numbers or addresses, or can it work with less? Collect and pass on only what the task needs.

Where the data goes

  • Which supplier processes the data, and where is it stored?
  • Does the supplier use your data to train its models? Get the answer in writing.
  • Is there a written contract covering security, retention and deletion?

Retention and access

  • How long are conversation logs and scanned documents kept, and who can see them?
  • Can you find and delete a customer's data if they ask?

Human oversight

  • Who reviews the AI's output, and how often?
  • What does the agent do when it is unsure? It should hand over to a person, not guess.

If any answer is unclear, sort it out before launch. It is much easier to design these controls in at the start than to add them later.

How to choose your first project

Choose the task that happens often, follows the same pattern each time, makes mistakes easy to catch and involves little personal data. Scoring each candidate from one to three on those four points gives a quick ranking, and the highest total is usually the best first project for a small firm. The four points:

A clinic receptionist at the front desk with an appointment book and paper forms
  1. Volume: how often does the task happen each week?
  2. Repetition: how similar is each instance?
  3. Risk: how easy is it to spot and fix a mistake? (Higher score for lower risk.)
  4. Data sensitivity: how little personal data is involved? (Higher score for less.)

The project with the highest total is usually your best start. For many firms that will be the sales dashboard, because it involves no customer contact. Firms with heavy WhatsApp traffic often find the enquiry agent gives the most visible result. Accounts-heavy firms tend to choose document reading.

You can see how other firms approached similar first projects in our case studies, and how we run a project from scoping to handover on how we work.

Next step

Codech Solutions builds AI agents, document automation and business insights for SMEs in Malaysia and Hong Kong. If you are one of the many Hong Kong SMEs with AI in the plan but not yet in use, the useful next step is a short, specific conversation about one task. Our AI audit looks at your current processes and data and recommends a first project, with the personal data points flagged. Book an AI audit or contact us to talk it through.

Frequently asked questions

How many Hong Kong SMEs have adopted AI?

The Dah Sing Bank 2026 survey, as reported by Human Resources Online, found that 23% of Hong Kong SMEs have adopted AI or generative AI.

What is a good first AI project for a small Hong Kong firm?

Choose one narrow, repetitive task where mistakes are easy to spot, such as a WhatsApp enquiry agent, invoice reading or a weekly sales dashboard. The dashboard is often the lowest-risk start because it does not talk to customers.

Does the PDPO apply when we use AI tools?

Yes, whenever the tool handles personal data of customers or staff. Check purpose and notice, use only the data the task needs, know where the supplier stores and processes it, set retention rules and keep a person reviewing the output. Take legal advice for your own case.

How long does a first AI project take?

It depends on your systems and data. As an illustrative example, a narrow project can be scoped, tested with staff and put into limited live use over about four weeks. Your own timeline may differ.

Will an AI agent replace our customer service staff?

A well-designed agent handles routine questions and passes anything unusual, such as complaints, refunds or pricing exceptions, to a person. Staff still own the customer relationship.

Free AI proposal + interactive prototype · no obligation

Want this working in your business?

Book a free 45-minute call. We send a free AI proposal and an interactive prototype within 1 to 2 weeks.

Free AI Proposal & Prototype

Get a free AI proposal and a prototype you can try

We reply within one business day.

By sending this form, you agree to our Privacy notice.