DEV Community: Pang Jun Han

How much does AI automation cost for a Malaysian SME? Real numbers.

Pang Jun Han — Tue, 27 Oct 2026 07:11:21 +0000

"How much will it cost?" is the first question every business owner asks me about AI. Most answers online are either "it depends" or a big agency package. So here are my actual numbers, from setting up AI workflows for Malaysian contractors and SMEs.

First: there's no fixed package

AI automation works best when you start with the one task that costs your team the most time. So I charge for time used, not a bundle:

Free task check (30 minutes): you show me the paperwork that eats your week. I send a written scope.
AI introduction and set-up: RM600 per 4-hour session, usually once.
Workflow build: RM400 per 4-hour session, on-site or remote.
Monthly support: RM300 a month for fixes, small changes and a weekly check-in. Month to month, no minimum term.
Managed service: RM1,000 a month, once automatic jobs are running every day.

What a process actually costs to build

Simple (staff claims, timesheets, letters and summaries): 1–2 sessions, RM400–800
Medium (WhatsApp site returns, stock lists, progress reports): 2–4 sessions, RM800–1,600
Large (delivery orders vs purchase orders, invoices vs DOs, linked to your accounting or project system): 4–10 sessions, RM1,600–4,000

A typical first phase

Here's an example (an example, not a package):

Free task check: RM0
AI introduction and set-up: RM600
One main process, like matching DOs to POs (6 sessions): RM2,400
One simple process, like staff claims (1 session): RM400
First month of support: RM300
Total: about RM3,700

What it saves

Two anonymous examples from real projects:

An electrical and ELV contractor in Penang: about 1,160 hours a year of staff time saved, worth about RM26,000.
A piling contractor: 1,300+ delivery orders read and 18,000+ purchase order lines matched by AI.

Costs people forget

AI tool subscriptions, if you need them, are billed by the AI provider directly: about US$20–25 per user per month.
Your team's time to check what the AI flags. A good workflow keeps that list short, but it isn't zero.

How to keep the cost down

Start with one boring, repetitive document, not a company-wide "AI strategy".
Pick the task that is done most often, not the most impressive one.
Keep a person on the exceptions. AI that says "I'm not sure" is cheaper than AI that's quietly wrong.

If you run a Malaysian SME or contractor and want to know what your first process would cost, the 30-minute task check is free.

Pang Jun Han: AI automation for Malaysian construction companies and SMEs. Full pricing

We taught AI to read delivery orders. The AI was the easy part.

Pang Jun Han — Mon, 26 Oct 2026 12:37:07 +0000

Every construction site runs on paper. Concrete, steel, diesel, rebar: each delivery comes with a signed delivery order (DO), and someone in the office has to key every one into the system, match it to the purchase order, and file the scan. At a busy contractor that's hours of work every week, and it's the kind of work nobody notices until it falls behind.

Over the past few months I set up an AI workflow at a Malaysian construction company to take that job over. Here's what it does and what I learned.

How it works

Site and store staff scan the signed DOs into one folder. That's the only manual step.
Overnight, the AI reads each DO (supplier, items, quantities, dates) and matches it to the right purchase order or site transfer.
It enters the record in the company's system with the signed scan attached.
Anything it isn't sure about goes on a short list. In the morning, staff only check that list.

What I learned

The hard part isn't the AI, it's the paper. Handwriting, stamps over text, carbon copies, photos taken at an angle. Most of the work was teaching the workflow what "good enough to trust" looks like.

Letting the AI say "I'm not sure" is the most important feature. People trust the system because it doesn't pretend. A short list of flagged items is easy to check; a system that is quietly wrong is worse than doing it by hand.

Missing paperwork is as valuable as processed paperwork. Once every DO is tracked, you get a daily list of what hasn't been scanned yet, so the office knows exactly what to chase from site.

Start with one boring document. Not a company-wide AI strategy. One document type, one folder, one list. Once that works, the next one (store transfers, diesel, claims) is much faster.

Results so far

1,300+ delivery orders read
18,000+ purchase order lines matched
41,000+ records checked

If you run a contractor or SME

You don't need a big IT project to start. If your team spends hours keying paper into a system, that's where AI pays for itself first. I'm happy to share what worked.

Pang Jun Han: operations & AI automation, founder of MyCreatorPulse. pangjunhan.vercel.app

I built a daily analytics app for Malaysian creators (Next.js + Neon + Vercel)

Pang Jun Han — Mon, 28 Sep 2026 05:25:40 +0000

A lot of creators in Malaysia post the same video on TikTok, Instagram and YouTube, some on Douyin and Xiaohongshu too, and every app shows its numbers in its own place and its own way.

So I built MyCreatorPulse: one dashboard for all of them. Every night at 2am Malaysia time it reads a creator's public numbers on TikTok, Instagram, YouTube, Facebook, Douyin and Xiaohongshu, saves a snapshot, and in the morning shows what each post gained overnight. No platform login, you just give it your username.

It's still in its pilot with a handful of creators. Here's what I've learned building it.

The stack

Next.js 15 (App Router), React 19, TypeScript
Tailwind v4
Postgres on Neon, with Drizzle
Better Auth for sign-in
Vercel (Singapore region) with two cron jobs
Resend for email, Cloudflare for DNS

Two crons do most of the work

"crons": [
  { "path": "/api/cron/snapshot", "schedule": "0 18 * * *" },
  { "path": "/api/cron/digest",   "schedule": "0 0 * * *" }
]

18:00 UTC is 2am in Malaysia, when everything gets read. 00:00 UTC is 8am, when the morning email goes out. Each run claims the day before doing anything, so if a cron ever fires twice, the second one does nothing.

Time zones will get you

Every date in the app is pinned to Malaysia time explicitly. My PC is on UTC+8, so a bug that leaned on the machine's clock would never show up locally. The tests run with TZ=UTC, same as Vercel, to catch those.

Tests that block the deploy

The numbers are the whole product, so the checks run inside the build:

"build": "npm run check && next build"

They replay real bugs on an in-memory Postgres (PGlite) through the same code the dashboard uses. If one fails, Vercel refuses the deploy and the site stays on the last good version.

Lesson from this week: a test must never depend on the build machine's environment. One of mine created fake PayPal subscriptions on a made-up plan ID. Locally that env var was empty, so it passed. On Vercel the real one was set, the check failed, and three deploys quietly never went out. Now the test sets what it needs and puts it back after.

Engagement rate is easy to get wrong

Sounds simple: likes + comments ÷ views. But photos don't have views. Add up likes over every post and views only over videos, and a photo-heavy account ends up with a crazy rate. Only count posts that actually have a view count:

const withViews = posts.filter((p) => p.views && p.views > 0);
const views = withViews.reduce((s, p) => s + p.views!, 0);
const reactions = withViews.reduce((s, p) => s + (p.likes ?? 0) + (p.comments ?? 0), 0);
const rate = views > 0 ? (reactions / views) * 100 : null;

I turned this into a free calculator anyone can use, no sign-up: https://mycreatorpulse.com/engagement-rate-calculator

Platforms don't all tell you the same thing

YouTube rounds public subscriber counts to three figures (12.3K), and the API does too, so the app shows what YouTube shows instead of pretending it's exact.
Instagram only has view counts on reels and videos, and a hidden like count is unknown, not zero.
Xiaohongshu doesn't show views publicly at all.

The rule I ended up with: if a platform doesn't publish a number, the app doesn't guess it.

HTML email is 2005 again

The weekly summary email has charts in it. Gmail strips <style> tags, SVG and plenty more, so the "charts" are table cells with a height and a background colour:

<td valign="bottom">
  <div style="height:36px;background:#1f9d6b;border-radius:3px 3px 0 0"></div>
</td>

Tables and inline styles only. Ugly to write, works everywhere.

What's next

Instagram and YouTube pages just went up, Douyin and Xiaohongshu are there for creators who post in Chinese, and there are plans for talent agencies now too. If you're a creator, or know one in Malaysia, the live demo works without signing up: https://mycreatorpulse.com

Happy to answer anything about the build in the comments.