By Luis Lopez, AI transportation consultant, CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast
Scrap metal and recycling freight does not look like the freight most AI tools were built for. There are no neat pallets, no consistent dimensions and often no purchase order with a clean line item. Loads are measured by weight, graded by eye and moved in roll-off containers, dump trailers, open-top trailers, flatbeds and export containers. The paperwork that matters most is the scale ticket.
That messiness is exactly why AI is starting to show up in this corner of the industry. When the data is scattered across scale houses, yard notes, driver photos and mill receiving reports, software that can read, compare and predict has real work to do. Here are ten places where it is making a difference, and where it still falls short.
1. Computer Vision Grading With Human Graders in the Loop
Scrap value depends on grade, and grade has traditionally been decided by an experienced inspector looking into the container or trailer. Camera systems at scale houses and unloading areas can now capture images of each load and suggest a likely grade or mix of materials. Used well, these tools give the grader a second opinion and a photo record. They do not replace the grader. Lighting, rain, dust and loads that hide material underneath all limit what a camera can see, so the final call stays with a person who can climb up and look. For more on how cameras are being used in freight yards, see computer vision in warehouse and yard freight.
2. Contamination Detection Before the Load Ships
Contamination is one of the biggest sources of disputes in recycling freight. Non-metallic material in a metal load, wrong materials in a bale, or prohibited items in a container can lead to downgrades or rejections at the receiving mill or port. Image models can flag suspect items, such as sealed tanks or objects that should not be in the load, so the yard can pull them before the truck leaves. The value is catching problems at origin, where fixing them is cheaper than at destination.
3. Pickup Scheduling From Fill-Level Data
Roll-off containers sitting at generator sites, such as manufacturers, demolition jobs and fabrication shops, need to be swapped when they are full. Some operations now use fill-level sensors, driver photos or simple customer check-ins to estimate how full each container is. Forecasting models then predict when a container will need a swap and group pickups so trucks run fewer partial trips. The result is fewer emergency calls and fewer wasted dry runs to containers that were not ready.
4. Routing Between Yards, Mills and Ports
Recycling freight has a network shape: feeder yards collect material, processing yards prepare it, and finished material moves to mills, smelters or export terminals. Route and dispatch tools can sequence roll-off swaps, plan inbound and outbound loads together and consider facility hours, scale house queues and port cutoffs. The data that matters most here is local: which gate opens when, which scale backs up in the afternoon, and which mill takes loads only by appointment.
5. Overweight Risk Prediction
Scrap is dense and loaded by grapple, magnet or loader, which makes it easy to overload a trailer without realizing it. On the Interstate system, federal limits are 80,000 pounds gross, 20,000 pounds on a single axle and 34,000 pounds on a tandem axle under 23 CFR 658.17, and states have their own rules and permit programs off the Interstate. AI tools can use past scale tickets for the same material, container type and loader to estimate the likely weight before the truck leaves the yard, giving the loader a warning to stop short. Onboard scales and a certified scale remain the real check. If a ticket does happen, read who pays for overweight tickets.
6. Scale Ticket Reconciliation
A single load can generate an origin scale ticket, a destination scale ticket, a grade report and an invoice. When the weights or grades do not match, someone has to figure out why. Document AI can read scale tickets from photos or PDFs, match them to the correct load and flag differences above a set tolerance. That turns a manual end-of-month reconciliation into a daily exception list. The rules for what counts as an acceptable difference still need to come from the people who negotiate the contracts.
7. Handling Market-Driven Volume Swings
Scrap volume moves with commodity prices, construction activity and manufacturing output. When prices rise, yards ship more; when prices fall, material often sits. Demand forecasting models that combine a company’s own shipment history with outside signals can help carriers and yards plan capacity a few weeks out. These forecasts are useful for direction, not precision. Nobody predicts commodity markets reliably, and a forecast should be treated as a planning input that gets revisited often.
8. Export Container Planning
A lot of recycled material moves overseas in ocean containers. Loading scrap into containers involves weight limits for the box, the chassis and the road, plus the need to balance material inside the container. Planning tools can suggest how much of a given material fits within the allowed weight, track booking cutoffs and match container pickups at the port with loading slots at the yard. Paperwork tools can draft export documents from yard data for a person to check before filing.
9. Safety Monitoring
Scrap yards combine heavy equipment, moving trucks, loose material and people on foot. Camera systems can detect when a person enters a zone where a loader is working, when a driver is outside the cab in a restricted area, or when a trailer door or tarp is not secured before departure. Securement for open loads still follows FMCSA cargo securement rules in 49 CFR 393.100 through 393.136, and a vision system flagging an unsecured tarp is a reminder, not a substitute for the driver’s inspection.
10. Knowing the Limits
AI in recycling freight has clear limits. Models trained on one yard’s material may not work at another. Cameras cannot see what is buried in a load. Weight predictions are estimates, not certified readings. Grade suggestions can be wrong in ways that cost real money if nobody checks them. As an AI transportation consultant, the first thing I look at is whether a tool keeps a human decision point where errors are expensive: grading, rejections, overweight calls and billing. Good tools make those people faster. Bad tools hide the decision inside a score nobody can explain.
Where to Start
- Digitize scale tickets first. Almost every other use depends on clean weight data.
- Capture a photo of every load at origin and destination, tied to the ticket number.
- Pick one pain point, such as reconciliation or overweight warnings, and measure it before and after.
- Keep graders, loaders and dispatchers involved in setting the rules the software uses.
Recycling freight will always be physical, heavy and a little unpredictable. The opportunity is to use AI where the data already exists and keep people in charge of the calls that matter.
For more conversations on freight and AI, subscribe to the Freight Guru Podcast.
About the author: Luis Lopez is a Miami-based AI transportation consultant and logistics entrepreneur, the CEO of Go Hub.io Holdings Corp and subsidiaries, and host of the Freight Guru Podcast.


