The person most likely to notice a hazard is standing next to it. A blind corner where a forklift comes through too fast. A packing station set at the wrong height. A pallet that keeps splitting when it's stacked three high. Most of those observations never reach anyone who could act on them. The reason is usually not indifference. It's that the path from noticing something to reporting it takes longer than the moment lasts. ## The problem Someone spots a problem at 10:40 in the morning. To report it, they need to find the shift lead, who is somewhere else. Or wait for the end-of-shift huddle, four hours away. Or walk to the terminal in the break room, remember a login they use twice a year, and fill in a form built for a desk. By the time any of that is possible, the urgency is gone. The forklift didn't hit anyone. People move on. This is a problem with the reporting path, not with motivation. The company usually has a system for exactly this kind of report. The people who see the problem can't reach it. ## What Amazon does Amazon built an internal tool called Dragonfly for this. Employees submit safety suggestions in real time, either from their own device or from workstations on site. Submissions go to the employee's manager and escalate to site leadership when needed. The interesting part is how people get in. Amazon puts green panels on the floor that read "Have a safety idea?" with a QR code and one line of instruction underneath. Scan, submit, back to work. No hunting for a terminal, no URL to remember. According to Amazon's 2025 safety update, more than 200,000 employees used Dragonfly to share ideas that year. The suggestions range from more lighting along walkways to sensors and vibration alerts that warn people when a forklift is nearby. Amazon reports its global recordable incident rate has improved 43% since 2019. Two things are worth taking from that. The volume is one. When reporting is easy enough, a lot of people do it. Two hundred thousand people is not a safety committee. It's the workforce. The loop is the other. Amazon added a feature that lets employees rate how their manager responded to a submission. That detail matters more than the form does. A channel that takes input and produces nothing visible stops being used quickly. ## The part most companies get wrong Companies without Amazon's budget usually conclude they need to build a Dragonfly. They don't. They need the last thirty seconds of it. Almost every company already has somewhere for a safety report to land. A form in ServiceNow. A Jira project. A Freshservice queue. An SAP EHS module nobody logs into. A Microsoft Form somebody in HSE built two years ago and still maintains. The form exists. What's missing is a route to it that works while someone is standing in a warehouse aisle wearing gloves. ## How a QR code closes the gap A QR code turns a physical location into an entry point. Someone scans it with the phone camera, the form opens, they type two sentences and get back to work. No app, no login, no walk to the break room. Amazon's panels route into A to Z, the app their employees already have. If your workforce doesn't have an app, the code can point straight at a web form instead. The mechanism is the same either way, and it isn't sophisticated. That's why it works. A few things decide whether it actually gets used. The form has to be short. If the QR code opens fourteen fields with a mandatory cost center, you've moved the friction rather than removed it. Three fields is a good target: what did you see, where, and optionally who you are. Allow anonymous submissions. Some of the most useful reports are the ones people won't put their name on. The target has to be reachable without a company account. Frontline workers often don't have one. If the link asks for a Microsoft login, the report doesn't happen. The URL behind the code has to be stable. This is where most of these projects quietly die. The form moves, the Jira project gets renamed, HSE switches tools, and two hundred printed stickers now point at a 404. Nobody reprints them. The channel is dead and nobody notices for six months. The fix is a redirect between the code and the form. The QR code points at a short address that never changes, and you update the destination behind it when the target moves. With Lora, that's a shortcut like `short/safety`, which comes with its own QR code and keeps working after the target URL changes. ## Where to put the codes Placement gets treated as an afterthought and then decides how much of this works. Sensible spots: at eye level in break rooms and changing areas, next to time clocks, on safety noticeboards, at the entrance to a production hall, on the inside of locker doors, on specific machines, positioned so the code is read when the machine is off rather than while it's running. Less sensible: anywhere that asks someone to stop and look down at a phone in a lane where forklifts, pallet trucks, or moving equipment operate. A reporting channel that creates a new hazard is a bad trade. If something needs reporting from inside an active area, the person should step out of it first. The same logic applies to lighting and height. A code at ankle level behind a rack in a badly lit corner will not be scanned, whatever's printed under it. Print size and contrast matter more than people expect once codes go on painted metal or shrink-wrap. Label every code with what it does. Amazon's panels don't show a bare code, they ask a question: "Have a safety idea?" An unlabeled QR code gets ignored, or gets mistaken for a material tag. ## What happens after the scan The reporting channel is the easy half. The harder half is what you do with what comes in. Decide before you print anything: who reads the queue, how fast the person who submitted gets a response, and what happens to reports that get rejected. "No" is an acceptable answer. Silence isn't. If you can't staff the triage, don't open the channel. A safety inbox nobody empties teaches people their input goes nowhere, and they'll apply that lesson to the next thing you launch. ## Measuring adoption Submission count is the obvious metric and the least useful one. It rises at launch and drifts down afterwards regardless of quality. Scans per code location tell you more. If the code in the break room gets scanned 60 times and the one in Hall 2 gets scanned twice, the question is whether Hall 2 is safe or whether the code there is badly placed. You can only see that if the redirect service counts scans per code, which is easier to set up before the codes go on the wall than after. The second number worth tracking is time to response. If it stretches past a week, submissions tend to fall off no matter how good the placement is. ## Limits QR codes don't replace a safety culture. If people don't believe they can stop work when something feels wrong, a sticker won't change that. Amazon pairs Dragonfly with a stated philosophy called Safe to Go, which lets employees speak up, slow down, or stop work if something doesn't feel right. The tool works because that policy exists. They also don't replace formal incident reporting. A recordable injury goes through the process it has to go through, and in the US that means the OSHA 300 log and the recordkeeping rules around it. What a QR code covers is the layer underneath: near misses, hazards, and the small suggestions that never make it into any system today. ## Effort An afternoon. Take the form you already have, put a stable short link on it, generate the QR code, print twenty to thirty durable stickers, and walk the site deciding where they go. The walk takes longest. It's also where you'll find half the hazards before a single code is scanned.