When Automation Makes a Mistake, Can You Explain It?
If you run a Malaysian SME, you may already use automation for customer messages, delivery scheduling, payment reminders, stock updates, staff attendance or document processing. These tools help your team work faster, but they also create a responsibility that many owners overlook: you must be able to explain what the system did when something goes wrong.
A recent investigation into a Waymo robotaxi collision shows why this matters. A vehicle struck a nine-year-old child near a school in Santa Monica while travelling at low speed. The United States National Highway Traffic Safety Administration, or NHTSA, asked Waymo for information about speed limits, school-zone behaviour, complaints and software. Waymo later submitted documents, with some material withheld as confidential business information. You can read the source report from TechCrunch.
Your business may not operate autonomous vehicles, but the underlying lesson is directly relevant: automated decisions need clear rules, reliable records and a human review process. If a customer is charged incorrectly, a delivery is sent to the wrong address or an employee receives the wrong instruction, you should not need to guess what happened.
TL;DR
Automation is only dependable when you can trace its decisions, identify exceptions and show who approved important actions. Build records, controls and escalation steps before you automate more sensitive work.
For Malaysian SMEs, start with three basics: define the rule, keep an activity log and require human approval for high-risk actions.
What This Means
The Waymo case is about more than autonomous driving. It highlights the importance of accountability in automated systems. When a system makes or assists with a decision, a business should be able to answer five straightforward questions:
- What information did the system receive?
- What rule or instruction did it follow?
- What action did it take?
- Was a person involved in approving or checking the action?
- What happened afterwards?
For a robotaxi, those questions may involve speed, road conditions, pedestrian movement and braking behaviour. For your business, they may involve a customer order, a stock alert, a credit application, a staff request or a complaint response.
This does not mean you need a large technology department. It means your automation should be designed with a visible trail. An activity log can record the time, user, input, decision and result. A workflow can pause an action when information is incomplete. An approval step can prevent an automated message or transaction from reaching a customer before someone checks it.
The safer question is not “Can we automate this?” but “Can we explain and correct it when the automation is wrong?”
The source article reports that NHTSA asked Waymo 12 questions, including sub-questions, and granted additional time for 10 of them. TechCrunch reported these investigation details. The practical business lesson is that documentation is not an afterthought. When scrutiny arrives, organised records make it easier to respond accurately and quickly.
How This Applies to Malaysian SMEs
Customer service is a good starting point. Suppose you use automation to reply to WhatsApp enquiries, Instagram messages or website questions. The system may provide an incorrect delivery date or misunderstand a customer’s request. Keep a record of the customer’s original message, the automated reply, the information source used and the staff member who corrected it. Create a rule that sends complaints, refund requests and unclear messages to a human instead of allowing the system to continue on its own.
Inventory and purchasing need the same discipline. An automated workflow may alert you when stock falls below a chosen level or prepare a supplier order. Before it sends anything, record the stock count, the minimum threshold, the recommended quantity and the person approving the order. This is particularly useful for cafés, workshops, retailers, wholesalers and online sellers where stock movement may come from several channels. If the figures appear wrong, you can see whether the problem came from a delayed update, duplicate entry or incorrect product setup.
Staff administration also benefits from clear controls. You might automate attendance summaries, leave requests, sales commission calculations or rostering. These processes affect your employees directly, so do not rely on a hidden formula that nobody understands. Show the underlying attendance entries, the calculation period and any manual adjustment. Require a manager to review unusual results, such as an unexpected absence record or a large change in commission. A clear review trail helps you resolve disputes fairly and avoid repeated manual checking.
Delivery and field-service businesses should record exceptions. If your team schedules installations, repairs or deliveries automatically, the system should capture failed visits, address changes, customer cancellations and safety concerns. A driver or technician should have a simple way to flag an issue. The workflow should then notify the right person rather than repeatedly sending the same appointment instruction. This is the business equivalent of handling an unusual road situation: normal rules are useful, but exceptions need attention.
Financial administration requires extra caution. You may automate invoice reminders, payment matching, supplier bill capture or approval routing. Keep the original document, the extracted details, the matching rule and the final approval. Any unusual transaction should be placed into a review queue. Do not allow an automation to silently change important records without showing what changed and who confirmed it.
Numbers to Track in Your Automation
You can begin with a small set of operational measures. These are not industry benchmarks; they are practical internal controls that help you see whether a workflow is dependable.
| Measure | What to record | Useful action |
|---|---|---|
| 100% of high-risk actions | Approval, timestamp and responsible staff member | Do not allow the action to proceed without review |
| Every exception | Reason, owner and resolution | Review repeated exceptions each month |
| At least 3 test scenarios | Normal case, missing information and unusual case | Test before changing a live workflow |
| One named owner | Person responsible for each automation | Give that person authority to pause it |
| One review schedule | Date for checking rules and records | Review monthly or after a serious error |
These control targets are recommendations for your internal process, not external statistics. The reported investigation itself involved one crash, 12 questions from NHTSA and additional responses submitted over several months, according to TechCrunch. The broader point is that detailed questions can arise well after an automated event, so records should remain available and understandable.
Practical Takeaways
- Map the workflow first: Write down the trigger, decision rule, action and expected result.
- Separate low-risk and high-risk tasks: Automate routine status updates, but keep approvals for sensitive customer, employee and financial actions.
- Keep an activity log: Record inputs, outputs, changes, approvals and errors.
- Create an exception queue: Send unclear, incomplete or unusual cases to a person.
- Assign an owner: Someone should review the workflow and be able to pause it.
- Test three situations: Check a normal case, a missing-data case and an unexpected case.
- Make corrections visible: Preserve the original result and record why it was changed.
- Train staff on escalation: Your team should know when not to trust an automated result.
- Review supplier systems: Ask what logs, approvals and export options are available before adopting a tool.
The Bigger Picture
As automation becomes more involved in daily operations, customers and employees will expect businesses to provide clear answers when a system fails. Regulators and industry bodies may also ask for evidence that important processes are controlled. The Waymo investigation demonstrates how a company can be required to explain software behaviour, operating rules and past complaints after an incident. The details come from a specific United States transport investigation, as reported by TechCrunch, but the management principle travels well.
For an SME, good automation governance is not about creating complicated paperwork. It is about making your business easier to manage. When you know which rule produced an outcome, you can fix the actual cause instead of blaming a staff member or restarting the entire process. When your team knows who owns a workflow, issues reach the right person sooner. When customers receive a clear explanation, trust is easier to maintain.
Start with one workflow that causes repeated errors or consumes too much staff time. Document its rules, add an activity log, define exceptions and introduce one approval checkpoint. Then observe the results for a month before expanding to another process. Small, controlled improvements will give you a stronger foundation than automating everything at once.
Ready to Streamline Your Operations?
Your business should run itself. AutoRunBiz deploys AI agents to automate your daily operations — WhatsApp orders, invoicing, customer follow-ups, and accounting. Book a free 15-min ops audit to see where automation fits your business →