Latest / Hardware Startups with Fexingo: Devices, Robotics, and Manufacturing Tech Companies / How Hardware Startups Fix Yield Rate Disasters
Transcript
- Lucas: There is a moment in every hardware startup's life where the dream meets the factory floor. You have a beautiful prototype that works perfectly in the lab, but then you order ten thousand units, and suddenly half of them are broken. It is not just bad luck. It is usually a failure to understand yield rates early enough. Luna: Right, because the lab is a controlled environment, but the factory is chaos. How do you even measure that chaos? Lucas: You start by looking at the difference between functional yield and cosmetic yield. Functional means the device turns on and does its job. Cosmetic means it looks good. Most founders obsess over functional, but cosmetic rejects can kill your margins just as fast if you are paying for premium packaging or manual touch-ups. Luna: So you are saying the cost isn't just the broken part, but the labor required to fix or hide the flaw? Lucas: Exactly. Let us look at a concrete example from last quarter. A robotics startup in Austin was producing a small autonomous delivery bot. Their initial run had a sixty percent defect rate. That meant for every three bots they made, two were scrap or rework. At that price point, they were literally burning cash with every unit coming off the line. Luna: Sixty percent? That sounds catastrophic. Did they stop production immediately? Lucas: They did not have a choice. They halted the line and brought in a quality engineering firm. What they found was not a design flaw, but a process variance. The adhesive curing time was too sensitive to ambient humidity in the assembly bay. A tiny shift in weather changed the bond strength, causing failures later in drop testing. Luna: So the product design was fine, but the environment around the assembly was throwing it off? Lucas: Precisely. This is where Statistical Process Control comes in. Instead of just testing the final product, you monitor the variables that create the product. If you track humidity, temperature, and torque settings in real time, you can predict a failure before it happens. It shifts you from catching errors to preventing them. Luna: That sounds expensive. Do most startups have the budget for real-time monitoring sensors? Lucas: Not necessarily. But they can start with simpler data collection. Even basic logs from the screwdrivers or glue dispensers can show patterns. If every third unit fails, and you notice the glue nozzle was cleaned right after the second unit, you have found your root cause. It is about connecting the dots between machine behavior and defect outcomes. Luna: I love that. It turns guesswork into detective work. But what about visual defects? You mentioned cosmetic yields earlier. Lucas: Visual inspection is one of the hardest parts to scale manually. Humans get tired. After hour four, a human inspector misses scratches that were obvious at hour one. That is why leading companies move to Automated Optical Inspection, or AOI, very early. Cameras check every surface against a digital master model. Luna: Does AOI catch things humans would miss, or is it mostly for speed? Lucas: It catches consistency. A human might let a minor scratch slide on one unit and reject another for the same scratch. AOI applies the same standard to every single item. For a high-volume consumer electronics brand, that consistency is worth more than the speed. It builds trust with the customer because every box feels identical. Luna: So it is really about reducing variability. Whether it is humidity or human eyes, you are trying to remove the random factors. Lucas: Variability is the enemy of profit. When you reduce variability, your cost per unit drops. But there is another layer: the feedback loop. Most startups fail here because they wait weeks for test results. By the time they know a batch is bad, they have already built thousands of bad units. Luna: That delay must be deadly for cash flow. How do you close that gap? Lucas: You integrate testing directly into the assembly line. Inline testing means a robot arm checks a component before the next step is added. If the component fails, the robot rejects it instantly. No further labor is wasted on a dead part. It saves material and time simultaneously. Luna: Inline testing sounds like a heavy investment in automation. Is it viable for smaller teams? Lucas: For true small batches, maybe not full automation. But modular fixtures can help. A simple jig that holds the device in a fixed position allows a quick manual check to be repeatable. The key is standardization. If every operator uses the same tool in the same way, the variance drops significantly without needing robots. Luna: Standardization over automation. I think many founders skip straight to buying robots because it looks impressive, but the process isn't ready. Lucas: Spot on. Automating a bad process just gives you bad results faster. You need to stabilize the process first. Once the yield is stable and predictable, then you invest in automation to scale volume. Mixing those steps up is a common recipe for disaster. Luna: What about the supplier side? If your components vary, your assembly will vary regardless of your internal processes. Lucas: That is a huge factor. You cannot control a supplier's factory, but you can audit their metrics. Ask for their own yield reports. If a supplier has a low yield, they are likely hiding defects by reworking parts loosely. You want suppliers who have high yields and tight tolerances, even if they charge slightly more upfront. Luna: Higher quality inputs lead to smoother assembly. It makes sense. But how do you enforce that with new suppliers? Lucas: Through clear specifications and incoming inspection protocols. Define exactly what acceptable looks like. Use gauges and go no go tools. If a part does not fit the gauge, it gets rejected at the dock. Do not let questionable materials enter the production floor. It protects your brand and your bottom line. Luna: So it starts before the factory even opens its doors. It is all about setting expectations and verifying them. Lucas: Verification is key. And documentation. Every defect needs a record. Why did it fail? Was it the component, the assembly, or the design? This data becomes your roadmap for improvement. Over time, you build a knowledge base that prevents the same mistakes from recurring. Luna: That knowledge base is an asset. It separates mature companies from chaotic ones. Lucas: Absolutely. Now, before we wrap up this deep dive into manufacturing reality, I want to mention something practical. If these conversations about building and scaling businesses are useful for what you're creating or running, listener support helps keep this network independent and ad-free. Honestly, if today was worth a coffee to you, that's the link — buy me a coffee dot com slash fexingo. Luna: Thanks for keeping the lights on. Back to the floor — what about software bugs in hardware devices? Lucas: Good question. Software defects often masquerade as hardware failures. A sensor might read correctly, but the code misinterprets the signal. This is why firmware validation must happen alongside hardware testing. You need to stress-test the code under edge cases, not just happy paths. Luna: Edge cases are where the real bugs hide. So you are suggesting parallel testing tracks? Lucas: Parallel tracks are essential. Hardware engineers and software engineers need to share defect data daily. Silos create blind spots. When both teams see the same failure pattern, they can pinpoint whether it is a physical limit or a logical error much faster. Luna: Collaboration across disciplines seems to be the theme here. It is not just about better machines, but better communication. Lucas: It is. Hardware is never just hardware. It is a system of people, processes, and parts working together. Mastering yield rates is really about mastering that system. It requires patience, data, and a willingness to listen to the factory floor. Luna: Listening to the data, not just the design dreams. That is a hard lesson to learn, but a vital one. Lucas: Vital. As you move forward in your own projects, ask yourself: where is my variability coming from? Is it the environment, the supplier, or the process itself? Identifying that source is the first step toward fixing it. Luna: And once you find it, do you fix it locally or change the whole system? Lucas: Local fixes solve immediate fires. Systemic changes prevent future ones. Aim for systemic change, but do not ignore the fire. Balance is everything in operations. Luna: Balance. A good reminder for any founder staring at a pile of defective units. Lucas: Exactly. Next time you hear about a recall, remember that it probably started with a small variance that went unchecked. Catching those variances early is the difference between a successful launch and a costly nightmare. Luna: So the real product is not just the device, but the reliability behind it. Lucas: Reliability is the product. Everything else is just packaging. Thanks for tuning in to this look at the messy, vital work of making things.