New: a free AI proposal + interactive prototype for Malaysian SMEs this quarter.
Professional services

No AI expert on staff? A first AI project for Hong Kong SMEs that does not need one

A Dah Sing Bank survey found that a lack of knowledge or skills is the main reason Hong Kong SMEs hold back on AI. You do not need to hire an expert to start. Choose one narrow process, set a baseline in hours, and run a 30-day pilot with an outside partner.

Many Hong Kong business owners would like to use AI but do not know where to start, and few have anyone on staff who could tell them. A Dah Sing Bank survey of local SMEs found that, among firms not yet using AI, a lack of relevant knowledge or skills was the most commonly cited barrier (Marketing-Interactive). Financial uncertainty, including doubt about whether the spending would pay back, also held firms back.

Public support is on the way. Financial Secretary Paul Chan has said an enhanced version of the Digital Transformation Support Pilot Programme, with HK$300 million set aside, will launch this year to help SMEs adopt ready-made AI tools (The Standard). Check the final eligibility rules when they are published before you plan around the funding.

Neither the skills gap nor the budget question has to stop you. You do not need an in-house AI expert to run a sensible first project. You need one well-chosen problem, a number to measure it by, and a partner who does the technical work for a fixed period. This article sets out how.

Why the worry is reasonable

Owners are right to be cautious. Without someone in-house who understands the technology, it is hard to judge a vendor's claims, hard to know what a fair price looks like, and hard to tell whether a tool is working once it is in. The common result is either doing nothing or buying a broad platform that nobody has time to set up.

The answer is not to hire an AI specialist before you know what you need one for. It is to make the first project small enough that an owner or operations manager can judge it on business terms alone: hours, errors and Hong Kong dollars.

Choose the problem yourself

You already know your business better than any consultant. The first project should come from a task your staff complain about, not from a list of AI features.

What a good first project looks like

Look for a process with four traits:

  • Repetitive. The same steps, many times a week.
  • High enough volume. Dozens or hundreds of cases a month, so the time adds up.
  • Mostly rule-based. A trained junior could do it from a checklist.
  • Easy to measure. You can count the cases and time them.

Avoid processes that depend on judgement, rare exceptions or relationships with key clients. Those can come later.

Two common starting points

Enquiry handling. Many SMEs answer the same WhatsApp and web questions every day: opening hours, prices, availability, booking slots. An AI agent can answer routine questions, take bookings and pass anything unusual to a person with the conversation attached.

Invoice and document entry. If someone in your office retypes supplier invoices, receipts or forms into the accounting system, Document AI can read them and post the data, leaving people to check exceptions.

Both are narrow, measurable and low risk if something goes wrong, because a person still sees the exceptions.

Set a baseline before anything is built

This is the step most first projects skip, and it is the one that answers the financial question.

For two weeks, count how many cases the process handles and time a sample of them by hand. Note the error rate if you can: wrong bookings, missed enquiries, invoices keyed twice.

Working out your baseline

Take a trading company that keys supplier invoices into its accounting system by hand. The calculation is simple:

  • Invoices per month, multiplied by the average minutes per invoice from your timed sample, divided by 60, gives hours per month.
  • Hours per month, multiplied by your own hourly staff cost, gives the monthly cost of the process in Hong Kong dollars.

Use your own payroll figures, not an industry average. The point is to have a number that belongs to your business.

With that number written down, you can judge any proposal plainly. If a tool and its setup cost more than the time it can realistically save over a year, it is not the right first project, however impressive the demonstration.

Run a 30-day pilot with an outside partner

A pilot lets you test the idea without hiring, without a long contract and without changing the whole business.

A Hong Kong business owner and a consultant pinning planning cards to a board in a small office

Week 1: set up

The partner connects the tool to the one process you chose, using a small set of real cases. You agree in writing what counts as success, for example hours saved per month and an acceptable error rate.

Weeks 2 and 3: run in parallel

The new process runs alongside the old one. Staff check every output. This is where you find the edge cases the demonstration did not show.

Week 4: measure and decide

Compare the pilot against your baseline. Count the hours, count the errors, and ask the staff who used it whether they would keep it. Then make one of three decisions: stop, adjust and extend, or roll out.

Our how we work page sets out how we structure pilots like this.

Who should own the pilot inside your business

The partner brings the technical skills. Someone inside your company still has to own the result. Pick a manager who runs the process day to day, not the most technical person in the office. Their job is simple: give the partner real examples, check the outputs during the parallel run, and report the numbers at the end of the month.

Set aside a few hours a week of their time for the 30 days, and tell the team why the pilot is happening. Staff who understand that the aim is to remove dull retyping, not to replace them, will point out problems early instead of working around them quietly.

Questions to ask any partner

You do not need technical knowledge to ask good questions. Before you sign, ask:

A Hong Kong businesswoman asking questions across a meeting table, notebook open, harbour view behind
  1. What exactly will be live after 30 days? Get the scope in writing.
  2. Where is our data stored, and who can see it? Customer data is covered by the Personal Data (Privacy) Ordinance (PDPO), whoever runs the software.
  3. What happens to our data if we stop? You should be able to take it with you.
  4. What will this cost after the pilot? Ask for monthly running costs, not only the setup fee.
  5. Who on your side handles problems, and how quickly? Name a person, not a support address.

A partner who cannot answer these clearly is not ready to work with you.

When to stop, and when to scale

Stop if the pilot does not beat your baseline, or if staff spend more time fixing the tool's output than they saved. A clean "no" after 30 days is a good outcome. It cost little and you learnt something specific.

Scale if the numbers hold and your staff want to keep it. The second project is easier: you now know how to set a baseline, run a pilot and judge a vendor. That is the in-house skill most SMEs were missing at the start.

Next step

Codech Solutions builds AI agents, document automation and business insights for SMEs in Malaysia and Hong Kong. If you would like help choosing a first project, book an AI audit. We will look at your processes with you, suggest one or two candidates, and help you set the baseline before anything is built.

Frequently asked questions

Do we need to hire an AI specialist before starting?

No. For a narrow first project, an outside partner can do the technical work. What you need in-house is a manager who knows the process, can supply real examples and can judge the result in hours and errors.

What is a good first AI project for a small business?

A repetitive, rule-based task with enough volume to measure, such as answering routine WhatsApp enquiries or entering supplier invoices. Avoid processes that depend mainly on judgement or key client relationships.

How do we know whether the project paid off?

Time the process by hand for two weeks before anything is built, then compare the pilot against that baseline at the end of 30 days. If it does not beat the baseline, stop.

Can we use the government's enhanced AI support programme?

Possibly. The Financial Secretary has said an enhanced Digital Transformation Support Pilot Programme will launch this year. Check the published eligibility rules and application process before planning around the funding.

Does the PDPO apply if a vendor runs the AI tool?

Yes. Your business remains responsible for the personal data it collects, even when a vendor's software processes it. Ask any partner where data is stored, who can access it and what happens to it if you stop.

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.