The phrase "ChatGPT integration" covers a lot of ground. It can mean anything from a chatbot on your website to a production workflow that classifies thousands of documents a day. This post is the opposite of the usual vendor case study — no screenshots of flashy dashboards, no "transformation journey" language. Just five things we actually built for Malaysian SMEs, what they changed, and whether they were worth the effort.
Details are lightly anonymised to respect client privacy, but the numbers, timelines and technical choices are real.
1. WhatsApp assistant for a property agency (Klang Valley)
The problem: Twelve agents were collectively spending 30+ hours a week answering the same prospect questions on WhatsApp — "is this unit still available," "what's the maintenance fee," "how many car parks."
What we built: A WhatsApp Business integration that routes incoming messages to an AI assistant grounded in the agency's live listing database. The assistant answers factual questions immediately, captures contact details for genuine leads, and hands off to a human agent for anything it's unsure about. Agents still get every lead — they just don't have to field "is this unit still available" for the hundredth time.
- Build time: 5 weeks
- Stack: OpenAI API + WhatsApp Business API + a small admin dashboard
- Outcome after 3 months: 68% of incoming messages handled end-to-end by AI. Agents now average ~6 hours/week on comms instead of 30+. Lead response time dropped from ~2 hours to under 30 seconds.
2. Invoice extraction for a logistics firm (Shah Alam)
The problem: A logistics client was receiving 400–600 supplier invoices per month as PDFs and scanned images, and one finance staff member was spending roughly three days a month retyping the data into their accounting system.
What we built: A small web app where finance drops invoices into a queue. A GPT-powered pipeline extracts supplier name, invoice number, line items, tax, and total, then presents the result for a quick human review before pushing to the accounting API. Confidence scores flag anything the model is unsure about.
- Build time: 7 weeks (including accounting-system integration)
- Stack: GPT for extraction + Bukku API for posting invoices + human-review queue
- Outcome: Monthly invoice processing time dropped from 3 days to roughly 4 hours. Freed-up finance capacity was redirected to cash-flow work the owner had been wanting to do for two years.
3. Internal knowledge assistant for a mid-sized manufacturer (Penang)
The problem: A manufacturing SME with ~80 staff had accumulated ten years of SOPs, compliance documents, HR policies and technical manuals across SharePoint, Google Drive and a legacy file server. New hires took months to know where to look for anything.
What we built: An internal chat tool that searches across all their document repositories and answers questions with citations ("according to Quality SOP v3.2, section 4…"). Staff ask in plain English or Bahasa Malaysia; the tool returns the answer plus a link to the source document so they can verify.
- Build time: 6 weeks
- Stack: Retrieval over document indexes + LLM for answer generation with inline citations
- Outcome: New-hire time-to-productivity dropped noticeably. More importantly, the HR manager stopped being the human search engine for the entire company.
4. Product-description generator for an e-commerce store (KL)
The problem: A homeware e-commerce SME was adding 40–60 SKUs per week. Writing listings (title, bullets, long description, SEO meta) in English + Bahasa Malaysia was taking their copywriter two full days a week.
What we built: A small admin tool where staff upload product photos and fill in a few key facts (material, dimensions, care instructions). The tool generates bilingual listings, the copywriter edits anything that doesn't sound right, and one click pushes to Shopify.
- Build time: 3 weeks
- Stack: LLM with a tuned brand-voice prompt + direct Shopify admin API integration
- Outcome: Copywriter time per SKU dropped from ~35 minutes to ~8 minutes. Just as importantly, listing tone stayed consistent across the catalogue — something humans struggle to do at volume.
5. After-visit summary assistant for a clinic group (Selangor)
The problem: Doctors at a three-branch clinic were dictating notes that clinic staff then had to type up for patient records and WhatsApp summaries.
What we built: A voice-to-text pipeline that transcribes dictations, then uses a carefully-prompted LLM to produce a structured clinical note plus a plain-language WhatsApp summary for the patient. Doctors review before anything is sent; nothing goes out automatically.
- Build time: 9 weeks (most of which was compliance review, not code)
- Stack: Speech-to-text + clinical-note-structured LLM prompt + patient-facing WhatsApp summary generator, all behind doctor approval
- Outcome: Admin time per visit dropped by ~60%. Patients consistently comment that the follow-up summaries are clearer than the old handwritten notes.
Got a repetitive task that's eating your team's week?
Tell us what it is. If AI is the right tool, we'll scope it honestly; if it isn't, we'll say that too.
Describe your workflowPatterns we've noticed
After twenty-plus of these integrations, a few things are always true:
- The winning projects target one narrow, painful workflow. "Automate customer service" always fails. "Handle the 'is this unit still available' WhatsApp message" works.
- The human stays in the loop for the first three months, minimum. Everyone who removed the human too early had an expensive story to tell about it.
- The integration bit — not the AI bit — is usually where the time goes. Talking to an accounting system or a WhatsApp API cleanly is the unglamorous work that makes the project actually ship.
- Running overhead is almost always lower than clients expect. For typical SME volumes, ongoing usage is modest relative to the time savings — the maths usually works out comfortably within the first year.
None of these are flashy. None of them will feature in a Ministry of Digital press release. But they're the kind of work that quietly gives Malaysian SMEs back a day a week — and that compounds into real competitive advantage year over year.
More reading: AI for Malaysian small businesses: a practical guide for 2026 · Custom AI vs. off-the-shelf tools: what Malaysian businesses need to know