What Should I Automate First in My Business?

The business owner sounded as if his office was on fire.

It was not literally on fire. No need to call SCDF.

His business was doing well. Orders were coming in, salespeople were closing deals and installation teams were moving around Singapore. This should have been good news.

Unfortunately, all the orders lived inside one large physical book.

One book.

Customer information, sales orders, deposit payments, final payments and installation dates were written inside it. When someone needed an answer, they had to search through the book like an archaeologist looking for an ancient civilisation.

He sent me a WhatsApp video showing how he was rummaging through it.

He had salespeople asking about commissions. Installers needed their schedules. Customers wanted updates. Payments had to be checked. Meanwhile, new orders continued arriving because business problems have a wonderful sense of timing.

“It is pure havoc,” he told me.

Today, many technology consultants would respond with two magical words:

AI agent.

Put AI here. Add AI there. Sprinkle a little AI on top. Soon even the office pantry may have an autonomous agent deciding who took the last packet of Milo.

But this business owner did not need an AI agent.

He needed an organised system.

That distinction can save a business a lot of time, money and unnecessary drama.

What should a business automate first?

The first process you should automate is usually a task that happens frequently, follows a reasonably clear process, consumes staff time and causes measurable problems when it is delayed or done incorrectly.

Do not begin by asking:

“Which AI tool should I buy?”

Begin with:

“Which repeated problem is wasting the most time or causing the most mistakes in my business?”

That is the short answer.

If you run a small or medium-sized enterprise in Singapore, your first automation project should ideally have a clear beginning and end, repeated steps, information that is already available, a result that can be checked and a manageable consequence if something goes wrong.

AI may eventually become part of the solution. It should not automatically become the solution just because everyone is talking about it.

Singapore businesses are adopting AI—but many still do not know what to do with it

AI adoption among Singapore SMEs rose from 4.2% to 14.5% within one year, according to IMDA.

That is significant growth. Business owners clearly understand that something important is happening.

But buying access to an AI tool and integrating AI into a business are two very different things.

It is like buying running shoes and registering for a marathon. Technically, you have started. Whether your legs agree is another matter.

The 2026 OECD survey on AI adoption among SMEs found that adoption is rising, while strategic and secure integration into business operations remains uneven.

A separate 2026 Goldman Sachs small-business survey found that business owners faced three major obstacles: data privacy and security concerns, a lack of technical expertise and difficulty choosing suitable AI tools.

This matches what I see when speaking to business owners.

They do not usually come to me saying:

“Please implement a multimodal autonomous agent architecture with dynamic tool orchestration.”

Thankfully.

They say things such as:

“My staff are doing this manually every day.”

“We keep missing things.”

“All our information is in WhatsApp.”

“I have one Excel file, but only one person understands it.”

“Can you make this easier?”

That is the language of a real business problem.

My job as a software developer is to translate that problem into the right system. Sometimes the answer involves AI. Sometimes it is a custom CRM. Sometimes it is straightforward automation. Sometimes I tell the business owner to use an existing product and save the development money.

Yes, a developer telling a client not to build software. Very scandalous.

Start with the pain, not the technology

When I first speak with a client, the client may not know exactly what is required.

That is normal.

The business owner understands the business. I understand software. We need to meet somewhere in the middle without either person pretending to be a wizard.

I ask leading questions. What happens when a customer places an order? Who receives the information? Who enters it again? Where do delays usually occur? What does the staff repeatedly copy and paste? What mistakes happen most frequently? If one employee goes on leave, what stops working?

The answers are usually more useful than a list of software features.

A business owner might ask for “an AI system”, but the conversation may reveal that the real problem is duplicate data entry. Another may ask for “a mobile app”, when a responsive internal web application would work better and cost less.

Someone may ask for a custom CRM, but an existing platform may already solve 90% of the problem.

The most expensive system is often the one built before anybody understood the problem.

The curtains-and-blinds business did not need an AI agent

Let us return to the business owner with the Great Book of Orders.

His problem involved customer records, sales orders, deposit and final payments, sales commissions, staff access, installation scheduling and management reporting.

An AI chatbot could certainly talk about these things in a very confident voice. It could probably produce a lovely poem about curtains too.

But it would not automatically create a reliable operational system.

The business required structured records, permissions, calculations and workflows. Therefore, I built a custom CRM and business management system for the curtains-and-blinds company.

The important improvement was not that the system looked futuristic.

It was that everyone could work from the same organised information.

Orders no longer depended on somebody locating the correct page in a physical book. Installation work could be scheduled more clearly. Customer and sales records were connected. Commission information became easier to review.

This was normal software solving an abnormal level of chaos.

And that is perfectly fine. Your business does not receive bonus points for using AI where a database would do a better job.

Automation and AI are not the same thing

These terms are frequently thrown together, so let us separate them.

Workflow automation

Workflow automation follows defined rules.

For example: when a new enquiry arrives, save it in the CRM, assign it to a staff member and send an acknowledgement.

The steps are predictable. The system does not need to sit under a tree and think deeply about the meaning of life.

Rules-based software

Rules-based software calculates or processes information using known formulas and conditions.

The custom nutrition macro calculator built for Take-Out Salad is a useful example. A user enters information, the application applies defined nutritional rules and the results are presented clearly.

This is software doing what software has always done well: following instructions accurately and repeatedly.

Calling every calculation “AI” does not make the calculator more valuable. It only makes the marketing department more excited.

Custom business software

Custom software becomes useful when your workflow is specific enough that off-the-shelf products do not fit properly.

This may include a custom CRM, booking platform, internal dashboard, customer portal, mobile application or management system.

It gives you greater control, but it also requires proper planning, testing, hosting and maintenance.

AI agents

An AI agent becomes helpful when the task involves variable information, interpretation or decisions that cannot easily be represented by fixed rules.

An agent might read enquiries written in different ways, extract information from documents, categorise incoming requests, summarise customer histories, search an internal knowledge base or prepare a suggested response.

Even then, the agent needs boundaries.

Giving an AI unrestricted access to customer records, email, payments and business systems is not “innovation”. It may simply be a very advanced method of asking for trouble.

A real automation example: Singapore accident alerts

I also developed a Singapore accident-alert Telegram automation system.

Its job is focused. Relevant accident information is processed and delivered to a Telegram group in a useful format. Duplicate alerts are controlled, failed deliveries can be retried and map information helps recipients understand the location.

This system solves a time-sensitive communication problem.

Could AI be added to analyse or classify less-structured information? Yes.

But the core value comes from reliable automation: receive the information, check it, prevent unnecessary duplicates, send the alert and give the user a useful action.

Sometimes the most valuable business automation is not the one with the most intelligence. It is the one that does a small but important job consistently.

Like a dependable staff member who arrives on time and does not organise a three-hour meeting to discuss the meaning of the “Send” button.

How to choose your first business process to automate

Write down five tasks that your staff perform repeatedly. For each task, give it a score from one to five in these areas:

• How frequently does it happen?
• How much staff time does it consume?
• How often do mistakes happen?
• How costly are the delays?
• How clear are the steps?
• Can the outcome be checked?

The strongest first automation candidate usually has high frequency, high time consumption, frequent mistakes, clear steps and a verifiable result.

However, begin carefully if failure could cause serious financial, legal, safety or privacy consequences.

Automating a weekly internal report is relatively safe.

Allowing an unsupervised AI agent to approve refunds, send legal replies and move company money around is more adventurous. Perhaps too adventurous unless you enjoy explaining things to your accountant.

Use the simplest suitable solution

1. Improve the existing process

Some workflows do not need software yet. They need fewer unnecessary steps, clearer responsibilities or a standard operating procedure.

Automating a terrible process produces a terrible process that moves faster.

Very efficient. Still terrible.

2. Check for an existing product

An established accounting, CRM, booking or project-management product may already solve the problem.

I have recommended existing platforms to clients since the WordPress era. If WordPress or another established product makes better financial sense, I say so.

Custom development should provide a genuine advantage. It should not be used merely because somebody knows how to build it.

3. Connect the products you already use

Sometimes the information is trapped between your website, email, spreadsheet, CRM and messaging applications. An integration may solve the problem without replacing everything.

4. Build custom software

Custom development becomes sensible when your workflow is important, specialised and poorly served by available products.

This is especially true when staff must work across several disconnected systems or when your business has developed a process that gives it a competitive advantage.

5. Add AI where interpretation is required

Use AI when the system must understand variable language, documents, images or less-predictable situations.

Even then, combine AI with normal software.

The AI may interpret an enquiry. The normal application should still validate required fields, enforce access permissions, record the action and ask for human approval when necessary.

AI should not be the entire building. It is one very useful room inside the building.

What should you not automate first?

Do not choose a process simply because it sounds impressive.

Your first project should probably not be a general AI agent that “runs the whole company”, a complicated system without a clear owner, a process nobody can explain or a low-frequency task that saves ten minutes per month.

Avoid automating personal customer conversations too aggressively.

Customers can tell when they are trapped inside a machine.

A carefully designed assistant can help staff answer faster. A badly designed one sends the customer around in circles until the customer becomes an amateur cybersecurity expert and starts looking for ways to escape.

How to calculate whether automation is worth paying for

You do not need a degree in finance.

Start with four numbers:

1. How many times does the task happen each month?
2. How many minutes does each occurrence consume?
3. What is the approximate hourly cost of the people involved?
4. What do mistakes, delays or missed opportunities cost?

Suppose three employees collectively spend 40 hours every month preparing and checking a report. If an automation reduces that to five hours, it saves 35 hours every month.

Now add less-obvious benefits: faster customer responses, fewer errors, easier staff handovers, better management visibility, less dependence on one employee and more time for revenue-producing work.

Compare those benefits with the initial development cost, subscriptions, hosting and ongoing maintenance.

Do not calculate ROI using imaginary savings such as:

“This AI agent will replace twelve employees.”

If you currently have four employees, replacing twelve may be somewhat challenging.

Use real numbers from your own operation.

Keep a human involved where consequences matter

The best automation does not always remove people. It helps people act faster and more accurately.

A human should generally remain involved when a decision affects large payments, contracts, employee discipline, sensitive customer data, safety, medical information, public statements or unusual refunds and disputes.

AI agents are powerful because they can act. That is also why they require limits.

Singapore’s AI Agents Sandbox findings highlighted their potential while also identifying risks involving cybersecurity, privacy, oversight and governance.

“Human in the loop” may sound like technical jargon.

In plain English, it means:

Before the computer does something serious, let a responsible person check first.

Revolutionary concept, I know.

Your first automation project should be slightly boring

This may be the least glamorous technology advice you receive this year.

Choose something slightly boring.

Choose a workflow your staff already understand. Make the result measurable. Keep the risk controlled. Fix it properly. Let the team gain confidence.

Once that project works, move to the next one.

A good first automation might capture website enquiries in one place, prepare quotations from approved information, send appointment reminders, assign new jobs to staff, generate a daily operational summary or notify the correct person when something requires attention.

These projects may not make you look like the emperor of artificial intelligence on LinkedIn.

They may, however, save real working hours and reduce daily frustration.

Your staff will probably appreciate that more.

My advice after more than two decades in software

I have worked as an employee, business owner and software developer across web applications, mobile applications and custom business systems.

Technology changes quickly. Business fundamentals change more slowly.

A business still needs to know who is responsible, what information is required, what happens next, what can go wrong and how the team knows the work is complete.

AI does not remove the need to answer these questions.

It makes the questions more important.

When a client cannot explain what is needed, I do not expect the client to suddenly become a systems analyst. I ask questions, study the workflow and help uncover the actual problem.

That human conversation is often where the most valuable work happens.

The final software may contain AI, automation, custom code or an existing platform. The business owner should not need to decide the architecture before explaining the problem.

That part is my job.

The simple answer

What should you automate first in your business?

Automate the smallest important process that repeatedly consumes time, causes mistakes and has a result you can measure.

Use ordinary automation when the steps are predictable.

Use established software when a suitable product already exists.

Build custom software when your workflow is specialised and commercially important.

Use AI when interpretation is genuinely required.

And if you are still unsure, start by recording a short video of the problem as it happens.

That is how one business owner showed me his Great Book of Orders.

The video did not contain a technical specification. It did not need to.

It showed the problem clearly.

Sometimes that is enough to begin.

Before you buy another AI subscription

AI can be extremely useful. I use AI tools in my own development work, and I believe they will become a normal part of many business systems.

But a tool is useful only when it solves the correct problem.

Before buying another subscription or appointing an autonomous digital employee with a very impressive name, look at what your staff are doing every day.

Find the repeated task.

Find the missing information.

Find the bottleneck.

Find the part that makes somebody say, “Alamak, this again.”

That is probably where your first automation project is waiting.

If you want a practical second opinion, see my custom software and business automation services in Singapore, browse my real CRM and automation case studies, or show me the workflow that is giving your team trouble.

What is the most repetitive or troublesome process inside your business today? Share it in the comments when I post this article, or send it to me. I would be interested to know whether it needs AI, normal automation, custom software—or simply a better way of working.