// Guide

AI for Singapore SMEs: Where to Start, Step by Step

Facts checked on 5 October 2026 against IMDA, EnterpriseSG and PDPC pages. This is general information, not legal or tax advice. Grant schemes change, so check the linked sources before relying on any figure here.

This guide sets out an order of work for AI implementation in a Singapore SME (a small or medium-sized business) that wants to use AI and does not know where to begin. It covers six steps, the mistakes that waste the most money, how grant funding works after the changes of 30 September 2026, what the PDPA (Singapore's personal data law) means for an AI project, and when it makes sense to bring in outside help.

We use one example throughout. It is illustrative, not a client: a 25-person aircon servicing company. Customers send job requests by WhatsApp and phone, a coordinator copies each one into a spreadsheet, finds a free technician, confirms the slot and later raises the invoice.

We build AI automation for Singapore SMEs, so we have a stake in what you decide. Most of the steps below cost nothing and need no consultant, and we say where that is the case. There is a short section on where we fit near the end.

Where should a Singapore SME start with AI?

Start with one task your staff repeat many times a week, measure what it costs you today, and try the AI features in software you already pay for before buying anything new. Automate only once you know the task, the volume and the baseline. Then measure again and decide whether to keep going, change course or stop.

Step	What you do	What you have at the end
1. Pick one task	Choose a single high-volume manual task	A task with a name, an owner and a weekly volume
2. Record a baseline	Count volume, time per item, errors and response time for two weeks	Numbers to compare against later
3. Try what you already pay for	Test the AI features in your current software on that task	A view of how far existing tools get you, at little or no extra cost
4. Automate	Connect the systems so the task runs with less copying and retyping	A working process with a named owner and a handover rule
5. Measure	Repeat the baseline measurement after a period you fixed in advance	A before-and-after comparison
6. Decide	Keep, extend to a second task, or stop	A decision you can explain to your team

If you are reading about AI for business or AI for companies and have no data team, these steps are written for you. They apply to a 5-person shop as much as to a 200-person firm.

How far has SME AI adoption in Singapore got?

About 1 in 7 Singapore SMEs had adopted AI by 2024. IMDA (the Infocomm Media Development Authority, the government agency for Singapore's digital economy) says in its Singapore Digital Economy Report 2025 that "Among SMEs, the AI adoption level tripled from 4.2% to 14.5% in 2024, driven mainly by uptake of off-the-shelf generative AI tools." Among non-SMEs the share rose from 44.0% to 62.5% over the same year.

The same report gives a picture of what AI-using firms do with it. In a short IMDA survey (a "pulse survey") of about 500 AI-using firms in May and June 2025:

84% used ready-made generative AI tools (AI that writes text or creates images, such as ChatGPT), 52% used AI features inside software built for a specific job (IMDA's examples include Xero Accounting and Info-Tech Cloud HRMS), and 44% had customised or in-house AI.
SMEs used AI in an average of 3 business functions, against 5 for non-SMEs.
Across all surveyed firms, AI was most widely used in IT (49%), customer service (43%) and finance and accounting (40%).

If you have not started, you are in the majority. Most of the firms that have started use ready-made AI tools, which is close to step 3 below.

Which task should you pick first?

Pick a task that happens many times a week, follows roughly the same pattern each time, and involves moving information from one place to another. Avoid tasks that need judgement on every item, happen once a month, or have no clear owner.

Good candidates usually look like one of these:

Answering the same customer questions on WhatsApp or email.
Copying details from messages or forms into a spreadsheet or system.
Writing routine documents from a template, such as quotes, job reports or follow-up emails.
Pulling figures from several places into a weekly report.

For the aircon company, the obvious choice is turning WhatsApp job requests into scheduled jobs. It happens dozens of times a week, every request needs the same details (address, unit type, problem, preferred time) and the coordinator retypes all of it. Invoicing is a second candidate, but it depends on the job record being right, so it comes later.

Write the task down in one sentence, name the person who owns it, and note roughly how often it happens. If you cannot do those three things, the task is not ready for automation yet.

How do you record a baseline before using AI?

Count the task for two normal weeks before changing anything. Record how many times it happens, how long each one takes, how often something goes wrong, and how long the customer waits. Without these numbers, any later improvement is an impression rather than a measurement.

A simple tally sheet is enough. For the aircon company, the coordinator would note:

Measure	How to record it	Illustrative figure
Volume	Job requests received per week	150
Time per item	Minutes from reading the request to confirming the slot	6 minutes
Errors	Wrong address, wrong unit, double bookings	4 a week
Response time	Minutes between the customer's message and the confirmation	45 minutes on average

These figures are invented for the example. At 150 requests and 6 minutes each, the task takes 15 hours of coordinator time a week, which is the number any automation has to beat.

Pick a quiet period and a busy period if you can. A baseline taken only in a slow fortnight will make any change look smaller than it is.

Should you try the AI tools you already pay for?

Yes, before buying anything else. IMDA's 2025 survey found that more than half of AI-using firms (52%) were using AI built into job-specific software such as accounting and HR systems. Check the plan page or admin settings of the software you already use, because the feature you need may already be paid for.

Give the existing tools a fair trial on the one task you picked. For the aircon company that might mean asking a business AI assistant to pull the address, unit type and problem out of a pasted WhatsApp message and lay them out in the spreadsheet's column order. It will save some typing. It will not move the details into the spreadsheet, check technician availability or send the confirmation by itself.

That limit is common. MIT Project NANDA's report, The GenAI Divide: State of AI in Business, 2025, found that tools like ChatGPT and Copilot "primarily enhance individual productivity, not P&L performance" (P&L means profit and loss), and that most of the enterprise-grade AI systems organisations evaluated "fail due to brittle workflows, lack of contextual learning, and misalignment with day-to-day operations." The same report says "95% of organizations are getting zero return" from generative AI. It is not a study of Singapore SMEs, so treat it as a warning about method rather than a forecast for your business.

Before staff paste customer details into any AI tool, check whether your plan uses that data for training. Our PDPA guide to ChatGPT and AI chatbots covers which plans do and how to switch it off.

When should you automate rather than prompt by hand?

Automate when most of the time left in the task goes on moving information between systems. If staff are copying an AI tool's output from one window to another dozens of times a week, the next saving comes from connecting the systems so the copying stops.

For the aircon company, that means the WhatsApp request is read automatically, the details land in the job list, the coordinator sees which qualified technician is free, and the customer gets a confirmation once the coordinator approves. The coordinator still makes the assignment. The retyping goes away.

Approach	For the aircon company	Cost pattern	Fits when	Main risk
AI features in tools you already pay for	Assistant drafts the job record from a pasted message	Often included in the current subscription	Volume is modest and one person does the task	Saves minutes per item but leaves the copying in place
Ready-made AI tool you subscribe to	A WhatsApp chatbot product that answers common questions	Monthly subscription, plus message fees on WhatsApp	The task matches what the product already does	Changing your process to fit the product
Connecting your existing tools	WhatsApp, the job list and the calendar linked with AI in between	Project fee, then running costs	The task crosses several systems you already use	Someone has to own it when one of those systems changes
Custom build	A purpose-built dispatch system	Largest project fee and upkeep	Nothing on the market fits and the volume justifies it	Cost and dependence on whoever built it

The third row is what we call AI integration. Our building maintenance dispatch project shows what it looks like in practice.

How do you measure whether AI worked?

Repeat the baseline measurement, in the same way, after a period you fixed before you started. Compare volume, time per item, errors and response time. If the numbers have not moved, the project has not worked yet, whatever it feels like.

For the aircon company, the coordinator would tally the same four measures for two weeks after go-live. Check one more thing: the new work the automation created. Someone has to review the job records it produces, and that time belongs in the comparison.

Set the review date in advance, for example four or eight weeks after go-live, so the decision does not drift.

How do you decide what to do next?

There are three honest outcomes: keep it and move to a second task, fix what is not working, or stop. Stopping is a result. A project that saved nothing after a fair trial has told you something about the task, the tool or the timing, and it cost less than continuing would have.

For the aircon company, a good result on job requests makes invoicing the natural second task, because the job record it depends on is now reliable. A poor result might mean the requests vary more than expected, and the better fix is a short standard form sent to every customer before any AI is involved.

None of these steps begin with replacing anyone. In the example, the coordinator's hours move from retyping to the jobs that need a decision.

What should SMEs avoid when starting with AI?

Starting with the tool. Buying a product and then looking for a use for it reverses the order above.
Automating a process nobody has written down. If two staff do the task differently, the automation copies one of them, or neither.
Skipping the baseline. Without it, nobody can tell whether the project worked, including the vendor.
Many pilots at once. One task done properly teaches more than five half-finished trials.
Building because a grant pays half. If the project only makes sense at half price, it is probably not worth building.
No named owner. Someone in the business has to notice when it breaks.

How does AI funding work for Singapore SMEs?

Since 30 September 2026, the main business grant is the EDGE Grant. EnterpriseSG states that "EDG, MRA and PSG ceased on 29 September 2026. From 30 September onwards, apply for business grant support under the EDGE Grant." EDG and MRA were the Enterprise Development Grant and the Market Readiness Assistance Grant. Guides written before that date may still describe the Productivity Solutions Grant (PSG), which no longer takes new applications.

The basics, from EnterpriseSG (Enterprise Singapore, the government agency that runs the grant) and its EDGE FAQ:

Support of "up to 70% for SMEs and up to 50% for non-SMEs", varying by activity. The FAQ gives up to 50% for SMEs on most activities. The 70% applies to internationalisation (overseas expansion) and sustainability activities.
Up to S$100,000 in total grant support per company per year, across all activities. Automation and digitalisation projects together count towards a lower cap of S$30,000 a year within that total.
The business must be registered in Singapore with at least 30% Singaporean or Singapore PR ownership.
"Businesses must not have started work or made any payment or deposit to a supplier, vendor, or third party prior to application submission."

IMDA's SMEs Go Digital programme lists "over 300 essential business solutions" that have been pre-approved, and includes a GenAI Navigator for finding AI tools. Some EDGE activities require a vendor from that list. EDGE's GenAI Customer Engagement Chatbot activity, for example, sends you to that list to choose a vendor, and pays up to 50% for SMEs on a cost capped at S$15,000. Our guide to EDGE funding for an AI chatbot covers that route.

Apply only after steps 1 and 2. The grant does not change whether the project works, and the baseline gives you the numbers to scope it. Our guide to AI grants for SMEs after PSG covers which EDGE routes fit an AI project, the application order, and how the AI tax deduction under the Enterprise Innovation Scheme interacts with a grant.

What does the PDPA mean for an AI project?

The PDPA applies whenever your AI project handles personal data, such as customer names, phone numbers or addresses. The PDPC (Personal Data Protection Commission, which enforces the law) describes it as "a baseline standard of protection for personal data in Singapore" with requirements "governing the collection, use, disclosure and care of personal data". Using AI does not change those duties.

In practice, for a first project:

Know where customer data goes when it enters an AI tool, and which company processes it.
Use business plans whose terms say how your data is handled, not staff members' personal accounts.
Tell customers what you collect and why, and collect only what the task needs.
Decide who can see the records the automation creates.

For the aircon company, customer addresses and phone numbers pass through WhatsApp, the AI step and the job list, so all three need checking. The PDPA checklist for AI chatbots goes through each point in detail.

What happens when the AI gets something wrong?

It will, occasionally. Plan for it before go-live rather than after the first complaint.

Wrong output. The automation misreads an address or a unit type. Keep a person reviewing anything that reaches a customer until the error rate is known and acceptable to you.
It does not know. A customer asks something outside the task, such as a refund or a complaint. Write down which messages go straight to a person, and who that person is.
Something upstream changes. A supplier changes a form or an app updates, and the automation degrades quietly. Without a named owner checking it, you find out from a customer.
The person who set it up leaves. Keep a one-page note of what the automation does, which accounts it uses and how to switch it off.
Costs drift. Usage-based fees, such as AI usage and WhatsApp message fees, rise with volume. Check the bill monthly for the first quarter.

What should you ask any AI vendor?

Which single task will this handle, and what volume is it designed for?
What baseline will we record before starting, and which number should change?
What does it cost to run each month after setup, and what makes that cost rise?
Where does our customer data go, which companies process it, and is it used to train any model?
What happens when it does not know the answer, and how does a person take over?
Who fixes it when one of our other systems changes, and how quickly?
Do we own what you build, and can we take it to another provider?
If this is grant-funded, which EDGE activity does it fall under, and will you wait until we have applied before starting work or taking a deposit?

Does your business need AI at all?

Not always. AI is worth trying when a task is frequent, repetitive and costs real hours. It is a poor first project when the task is rare, when the process changes every week, or when nobody can own it.

Hold off if:

No single task takes more than an hour or two a week. The setup will cost more than it saves.
Your records are mostly on paper. Digitising them is the first project.
The process is different every time someone does it. Standardise it first.
The main reason is that a competitor has done it, or that a grant would pay half.

Some SMEs ask about AI transformation programmes. For a business of 10 to 50 people, a sequence of small, measured projects usually makes more sense than a company-wide programme, and you can stop after any one of them.

When should an SME bring in outside help?

Bring in help when you have done steps 1 to 3 yourself and the next saving needs systems connected, or when nobody in the business has the time to own the work. Steps 1 and 2 need no outside help at all.

Outside help is also worth it when the task crosses several systems, when customer data makes the PDPA questions harder, or when your team needs training to use the tools you already pay for. See AI consulting for how we approach that work. For teams that need to get more from existing tools, see AI training.

Where Cortex Lab AI fits

We are a Singapore consultancy that builds AI automation inside the tools our clients already pay for, and we work to the PDPA. We are not a pre-approved EDGE vendor. If you have a task in mind and want a second opinion on whether it is worth automating, book a free 30-minute consultation.

Frequently asked questions

Where should a small business in Singapore start with AI?

With one frequent, repetitive task. Record how long it takes for two weeks, try the AI features in software you already pay for, and only then consider automating it. Measure again afterwards and decide whether to continue.

How many SMEs in Singapore use AI?

14.5% of SMEs had adopted AI in 2024, up from 4.2% in 2023, according to IMDA's Singapore Digital Economy Report 2025. Among non-SMEs the figure was 62.5%.

Is PSG still available for AI tools?

No. PSG, EDG and MRA closed on 29 September 2026. New projects apply under the EDGE Grant. For SMEs it pays up to 50% on most activities and up to 70% on overseas expansion and sustainability. Automation and digitalisation projects share a S$30,000 annual cap, within S$100,000 per company per year.

Do I need to hire a data scientist to use AI?

No. The first steps in this guide need someone who knows the task well, not technical staff. Connecting several systems may need outside help for one project. It does not need a new hire.

Will AI let us cut staff?

That is not the right goal for a first project. In most SMEs, the hours saved on retyping and routine replies go to work that needs a person, such as handling exceptions, complaints and sales.

Sources, all checked 5 October 2026: IMDA, Singapore Digital Economy Report 2025 (pages 4 and 16); IMDA SMEs Go Digital; EnterpriseSG EDGE Grant; EDGE Grant FAQ; EDGE GenAI Customer Engagement Chatbot activity; PDPC, Personal Data Protection Act overview; MIT Project NANDA, The GenAI Divide: State of AI in Business, 2025 (PDF copy hosted by MLQ.ai, page 3).