When Your Business Cannot See What Is Going Wrong

You may already have cameras around your workshop, warehouse, or production area. They record hours of footage, but when something goes wrong, someone still has to search through the video manually. A missed safety step, misplaced carton, damaged item, or bottleneck may only become obvious after a customer complains or a delivery is delayed.

That is the practical problem behind a new wave of visual artificial intelligence. Instead of using cameras only for security, businesses can use computer vision to understand what is happening in a physical environment and support decisions. A new startup called Perceptron is one example, developing software intended to help robots perceive, reason, and act in industrial settings. Source: TechCrunch

TL;DR: Visual AI can help you monitor processes, identify exceptions, guide workers or robots, and turn recorded video into useful operational information. For a Malaysian SME, the sensible starting point is not a fully automated factory; it is one clearly defined workflow where better visibility can reduce delays, errors, or safety risks.

What This Means

Visual AI is software that interprets images or video. Traditional camera systems mainly show you what happened. Visual AI attempts to understand what is happening: whether a package is in the correct area, whether a worker is following a process, whether an item is missing, or whether a robot has enough space to move safely.

Perceptron’s Isaac 0.5 model is designed for industrial environments such as warehouses and factory floors. The company says it can support vision-guided robots by helping them navigate complex spaces and extract information from video recorded by those robots. The model is being released as an open-weight system, meaning its parameters and training materials can be inspected by others. Source: TechCrunch

The important idea is flexibility. Sorting a box is not one action. A system may need to read a label, locate the package, understand the surrounding space, choose a box, plan the order of several movements, and place each item correctly. Visual AI aims to connect these steps instead of treating every movement as a separate, rigid instruction.

Perceptron says its model was trained using one million hours of general video, along with ego video and UMI video that capture human actions from a first-person perspective or record repeated movements. Source: TechCrunch You do not need that scale of data to begin. Your business may start with a much smaller set of approved examples from one process, provided the footage is clear and the business purpose is specific.

Key insight: The value of visual AI is not the camera itself. It is the ability to turn routine physical activity into timely information that helps you act.

How This Applies to Malaysian SMEs

For a Malaysian SME in food manufacturing, packaging, electronics assembly, or general production, visual AI could support quality checks. A camera positioned above a workstation may identify whether a component is present, whether packaging is sealed, or whether a label appears in the correct position. This does not mean removing human judgement. It means giving your team an additional check before goods move to the next stage. It can also create a record of recurring defects, helping you identify whether the issue is related to materials, equipment, training, or a particular production step.

In wholesale, distribution, and e-commerce fulfilment, visual AI may help with picking and dispatch. It could compare a parcel with an order record, confirm that a carton is placed in the correct dispatch zone, or flag an item that appears damaged. A small warehouse may not need robots at all. Fixed cameras and software alerts could already help your supervisor spot repeated misplacements, long queues, or areas where staff must walk back and forth unnecessarily.

For workshops and service businesses, the use case may be process monitoring rather than robotics. A vehicle service centre could use visual records to confirm that inspection steps were completed. A printing company could check alignment and surface quality. A furniture or fabrication business could monitor whether protective equipment is being used in designated work areas. You should be careful with privacy and avoid turning the system into constant employee surveillance; the strongest business case is usually a clearly defined process standard, not watching people for its own sake.

Malaysian SMEs also face varied environments. A warehouse may be hot, crowded, reflective, dim, or frequently rearranged. Bahasa Malaysia, English, Chinese, and handwritten notes may appear on labels and documents. This means a demonstration in a clean laboratory is not enough. You need to test the system in your actual premises, with your real lighting, packaging, uniforms, shelves, and operating pace.

If you use external vendors, ask where video is stored, who can access it, how long it is retained, and whether your footage is used to train another system. Personal data obligations may apply when identifiable employees, visitors, or customers appear in recordings. You should seek appropriate professional advice on compliance and document a simple internal policy before deploying cameras beyond ordinary security use.

A Simple Way to Evaluate the Opportunity

Area to assess	Questions for you	Useful first measure
Process	Which repeated task creates the most errors or delays?	Count exceptions per week
Visibility	Can a camera see the task clearly and consistently?	Review sample footage from different shifts
Decision	What action should follow an alert?	Record who responds and how quickly
Accuracy	How often does the system miss or wrongly flag an event?	Compare alerts with supervisor checks
Privacy	Are people or sensitive documents visible?	Define access and retention rules

Practical Takeaways

• Choose one workflow first. Start with carton verification, defect detection, stock movement, or a safety checkpoint rather than attempting to analyse the whole site.
• Define the decision before buying technology. An alert is useful only if someone knows what to do next.
• Collect representative footage. Include different shifts, lighting conditions, product variations, and normal exceptions.
• Keep a human review step. Treat early alerts as assistance, not unquestionable truth.
• Measure operational outcomes. Track missed items, rework, dispatch errors, response time, and process interruptions before and after the pilot.
• Ask vendors about integration. Check whether the system can connect with your inventory, order, production, or reporting software.
• Protect employee trust. Explain the purpose, limit access, and avoid collecting footage that is not needed for the approved workflow.
• Test locally. A system that performs well in another country may struggle with your site layout, labels, lighting, and work practices.

The Bigger Picture

The longer-term direction is a closer connection between software and physical operations. Business systems already manage orders, stock, schedules, and customer records. Visual AI adds another layer by interpreting what is actually happening on the floor. This could narrow the gap between the status shown in your system and the condition of your real inventory or production line.

Robotics will become more useful when machines can cope with variation instead of repeating one perfectly controlled movement. Perceptron’s founders describe their approach as general-purpose, covering perception and action across different environments rather than focusing on one narrow task. Source: TechCrunch Whether any particular model delivers that promise remains something businesses must validate through testing.

For you, the strategic lesson is straightforward: do not wait for a fully autonomous factory before learning how visual data can improve your operation. Begin by identifying where your supervisors spend time checking, searching, counting, or correcting. Those activities often reveal the best pilot. Build a clear process, protect personal information, involve your staff, and judge the technology by measurable improvements in daily work.

Visual AI will not replace good layout planning, reliable procedures, or trained people. However, it may help you notice problems earlier and give your team better information at the moment it matters. That is a practical foundation for smarter automation on the Malaysian SME factory floor.

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 →