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

Does your chatbot make decisions about customers? What Malaysia's PDPA means for an SME

If your WhatsApp or website bot sorts, scores or decides about customers, the amended PDPA is a practical question for you. A four-question checklist: what the bot sees, who can opt out, who reviews, and where the data is stored.

Many Malaysian SMEs now answer customers through a WhatsApp bot or a chat window on their website. Most of these bots do simple things: quote opening hours, book a slot, check an order. Some do more. They decide who gets a callback first, which enquiry is marked as a likely sale, or whether a customer qualifies for a promotion. Once a bot starts making choices like these about people, the Personal Data Protection Act 2010 (PDPA) becomes a practical question for the owner, not just for the legal team.

This article is a plain checklist for owners and managers. It is not legal advice, and it draws on one piece of commentary, so please confirm each point against the official text from the Personal Data Protection Department (JPDP) or with your own adviser before you act on it.

What has changed, and what the commentary says

The PDPA was amended in 2024, and the changes came into force in stages. The amendments brought in duties that many SMEs had not dealt with before, including telling the regulator about personal data breaches, appointing a data protection officer, and placing direct obligations on the service providers that process data for you.

A commentary piece from a Malaysian training provider, PDPA AI compliance Malaysia 2026, describes guidance on automated decision-making and profiling, and makes a point worth taking seriously: if a chatbot shows one customer's details to another customer, that could count as a personal data breach. In other words, a bot error is not just an embarrassing screenshot. It may set off the same reporting duties as a lost laptop or a hacked email account.

Two cautions. First, this is vendor commentary, not the regulator's own wording. Second, guidance in this area is still developing. JPDP has also consulted on a draft framework on AI and personal data, which we covered in five checks for SMEs already using AI. Treat what follows as good practice that will hold up under most readings of the rules, rather than as a statement of exactly what the law requires.

Does your bot actually make decisions?

Start by being honest about what the bot does. There is a real difference between these three kinds of bot:

Bots that only answer questions

A bot that answers "what time do you close?" or "where is my order?" handles some personal data, such as a phone number or an order reference, but it does not decide anything about the person. Your main concerns are what it stores and who can see it.

Bots that sort or score people

A bot that tags an enquiry as "hot lead", ranks who gets a callback, or flags a patient as urgent is profiling, even if nobody calls it that. The person on the other end may never know they were sorted, and a wrong tag can mean they wait longer or never hear back.

Bots that decide outcomes

A bot that approves or declines a booking, offers a discount to some customers and not others, or screens job applicants is making decisions that affect people directly. This is where the commentary's points on automated decision-making matter most, and where a human check is hardest to justify leaving out.

If you are not sure which group your bot falls into, ask the person who set it up to walk you through one real conversation from start to finish and show you every place the bot chose between two paths.

The four-question checklist

For each bot you run, write a single page that answers four questions: what personal data the bot can see, how customers can opt out and reach a person, who reviews the bot's decisions, and where conversations are stored and for how long. Keep the page up to date whenever the bot's set-up or rules change.

Two colleagues at an office table sorting four blank cards laid out in a row

1. What does the bot see?

List every piece of personal data the bot can read or receive: names, phone numbers, IC numbers, addresses, health details, purchase history, photos customers send. Then ask whether it needs all of it. A booking bot rarely needs an IC number. A delivery-status bot does not need the customer's full purchase history.

Pay close attention to anything the bot can look up. The breach risk described in the commentary usually comes from this: a bot that can search your customer records may return the wrong record when two customers share a name, or when someone asks cleverly. Limit the bot to the records of the person it is talking to, matched on something more reliable than a name.

2. Who can opt out, and how?

Customers should know they are dealing with a bot, and they should have a simple way to reach a person instead. "Type STAFF to talk to our team" is enough for many businesses. If the bot sends marketing messages, people need a clear way to stop them.

Check that the opt-out actually works. Send the message yourself and see what happens. If a customer is told a person will reply and nobody does, the hand-over is not doing its job.

3. Who reviews the bot's decisions?

For bots that sort or decide, name a person who checks the results, and decide how often. That might be a manager who reads a sample of bot conversations every Friday, or a rule that any declined booking is looked at by staff before the customer is told.

Also decide who a customer can complain to if they think the bot treated them unfairly, and make sure that person can see the conversation and override the outcome.

4. Where is the data stored, and for how long?

Many chatbot tools store conversations on servers outside Malaysia, and many keep them indefinitely by default. Find out which country the data sits in, which company holds it, and how long it is kept. Set a retention period that matches your business need, for example a few months for routine enquiries, and delete the rest.

Under the amended PDPA, the provider that runs your bot is likely to have its own duties as a data processor, but you remain responsible for choosing it and for what you tell your customers. Ask the provider for its security and breach-reporting arrangements in writing.

What a sensible set-up looks like

Here is an example of how a small professional services firm might put this into practice. It is an illustration, not a client case.

The firm's WhatsApp agent answers questions about services and fees, and books consultations. It can see only the name and number of the person messaging, plus the firm's public calendar. It cannot search past client files. Every conversation starts with a line saying the customer is talking to an automated assistant and can type STAFF at any time. Enquiries the bot cannot answer go to a shared inbox that a named staff member checks twice a day. Conversations are deleted after a fixed period, and the firm has a one-page note of what the bot does, which it reviews when the bot changes.

None of this needs a large budget. Most of it is a matter of deciding, writing it down, and testing that the bot behaves as the note says. For more on how we approach this in practice, see our page on AI agents and the TGN AI chatbot case study.

Common gaps we see

Most gaps fall into five groups: bots with access to more records than they need, bots nobody owns, rule changes nobody recorded, conversations kept indefinitely, and no plan for a breach. Each one comes from convenience at set-up time rather than bad intent, and each can be put right by deciding and writing down how the bot should behave.

An office administrator in a warehouse office reviewing paper records in a binder, cartons stacked behind
  • The bot can search too much. It was connected to the full customer database for convenience, and nobody narrowed it down later.
  • Nobody owns it. The bot was set up by a freelancer or a staff member who has since left, and no one knows what it was told to do.
  • No record of changes. Prompts and rules were edited over time, but there is no note of what changed or why, which makes it hard to explain a decision later.
  • Default retention. Conversations are kept forever because that was the tool's default setting.
  • No breach plan. If a customer reported seeing someone else's details, staff would not know who to tell or what to do in the first hours.

Each of these is fixable in days, not months. If you work in a regulated field, our notes for professional services firms cover the extra points that usually apply.

Where to start this week

Start small. If you run one or more bots today, three tasks will tell you most of what you need to know about your exposure: write the one-page note for each bot, test the hand-over to a person yourself, and ask your provider where conversations are kept. In detail:

  1. Write the one-page note for each bot, answering the four questions above.
  2. Test the opt-out and the hand-over to a person yourself.
  3. Ask your bot provider where conversations are stored and how long they are kept.

Then check the official JPDP material for any guidance that applies to your sector, and update the note when it changes.

Codech Solutions builds AI agents, document automation and business insights for SMEs in Malaysia and Hong Kong. If you would like a second pair of eyes, our AI audit reviews the bots and automations you already have and gives you a short list of what to fix first. You can also contact us to talk it through.

Frequently asked questions

Does the PDPA apply to a chatbot run by a small business?

If the bot collects or uses personal data from customers in the course of commercial transactions, the PDPA is likely to apply, whatever the size of the business. Confirm the details for your situation with JPDP's official material or an adviser.

Is a chatbot showing one customer's details to another a data breach?

Commentary on the 2026 guidance suggests it could count as a personal data breach, which may bring reporting duties. Treat it as a breach risk and plan for it, and confirm the exact duties against the official JPDP text.

What counts as automated decision-making for an SME bot?

Any time the bot chooses an outcome for a person without a human check, such as declining a booking, ranking leads or offering a discount to some customers only. Bots that only answer general questions usually do not make decisions about people.

Do I need to tell customers they are talking to a bot?

It is good practice to say so at the start of the conversation and to offer a simple way to reach a person. Check current JPDP guidance for any specific requirement in your sector.

Who is responsible if my chatbot provider has a breach?

Under the amended PDPA, processors have their own duties, but you remain responsible for choosing the provider and for your obligations to customers. Ask the provider for its security and breach-reporting arrangements in writing.

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.