We hear a lot about artificial intelligence helping offsite companies with marketing and sales. It can write an email, help organize customer information, and suggest ways to reach new builders. Those uses are easy to show. Type a request, wait a few seconds, and something appears on the screen.
But walk onto a modular factory floor and ask what AI is doing there. The answers become harder to find. Can it keep production moving? Can it catch mistakes before walls are closed? Can it help a manager avoid starting a home when half its materials are missing?
I believe those are some of the most useful questions our industry can ask. Another question belongs right beside them: Who pays while the factory and the AI company figure out whether the system works?
Research Is Moving Forward
The quieter conversation about production does not mean nothing is happening. Researchers are testing ways to improve scheduling, inspect components, and connect designs with factory equipment. A 2025 review examined 52 studies involving AI and computer models of offsite operations. That is considerable research, but it is not the same as 52 factories successfully using those systems every day.[1]
One example comes from University of Alberta researchers studying a wall-panel production line in Edmonton. They tested a system that helped decide when to send more work into the line. The idea was to keep unfinished panels from piling up behind slower operations. Their experimental results showed a 38.5 percent reduction in average production lead time compared with the factory's usual practice.[2]
That is encouraging. Anyone who has watched workers struggle to move around stacks of unfinished panels understands the problem. However, this was a research result from a particular case study. It does not mean a factory can buy software tomorrow and expect the same improvement.
Examples exist beyond research, too. Automated Architecture, known as AUAR, reported that two of its robotic MicroFactories had been delivered to Rival and that autonomous production was underway. The company also reported producing designs ready for its robots in under four hours. Those are the company's own results, and they do not tell us how much of the improvement came specifically from AI.[3]
That matters because automation and AI are not interchangeable. A machine following programmed instructions can be valuable without learning anything. Owners need to understand what a system actually does, what information it needs, and what problem it is supposed to solve.
Why the Factory Floor Takes Longer
In my view, production AI takes longer to prove because it has to deal with the factory as it really operates. A schedule may assume a wall takes twenty minutes to build. The crew may know it takes forty whenever certain options are included. If the system never receives that information, its recommendations can begin with the wrong assumptions.
Then come the everyday surprises: missing windows, an absent employee, a design correction, or equipment that stops working. A useful system has to help through those conditions. Showing that it works on one straightforward job is only a start.
Marketing also gets an easier demonstration. We can see an AI-written advertisement immediately, even though proving it brought in profitable business takes longer. On the production floor, we want completed homes, fewer mistakes, and better results. It takes time to show that an improvement lasts through different jobs and crews.
Some production work may receive less publicity, but I would not assume there is a hidden wave of successful installations. The evidence supports active testing and some commercial use. It leaves plenty of questions about dependable results and cost.
The Hidden Cost of a Free Trial
Now imagine a supplier offering a factory a ninety-day trial with no software charge. That sounds good until the production manager spends hours explaining the line, engineering cleans up drawings, and supervisors begin collecting information. Employees still have their regular jobs to do. Their time costs money, whether the supplier invoices them or not.
The trial might also require cameras, sensors, installation, or training. If testing slows production, there could be overtime. If a recommendation creates a problem, it could waste materials or require rework. These costs are possibilities to plan for, not reasons to reject every trial.
Who pays depends on the agreement. The supplier might provide the software and technical support while the factory provides employee time and equipment. A university project might fund research while still relying on the factory's help. Owners should ask directly who covers damaged materials, extra labor, and delays. Never assume that a free trial includes reimbursement for those expenses.
I would also ask whether we are testing an established product or helping develop an unfinished one. A factory helping invent the system contributes knowledge the supplier may later sell to other companies. That contribution should count when discussing price, support, and responsibility for trial costs.
Agree on the Rules Before Testing
Before starting, agree on a budget, including employee hours, and write down how the factory currently performs. Decide what success means and how you will measure it. Establish who can stop the test and how production will return to its previous process. Where possible, let supervisors check the system's suggestions before those suggestions control live work.
Gary's Observation
Gary Fleisher, modcoach@gmail.com
I want to see AI earn a useful place on our production floors. Factory managers have enough problems without being asked to ignore a tool that might help. But they also deserve clear answers about what a trial requires and how much the learning process could cost.
The supplier may be pleased that its software works while the factory is explaining a late delivery to a builder. Both sides need to agree that success includes what happens to production, employees, and customers. A convincing demonstration is valuable. A dependable improvement is worth much more.
Our factories should participate in developing better tools with their eyes open. Their people, experience, materials, and time have value. Before the first test begins, everyone should understand who is paying to learn and what the factory expects to gain.
Sources:
No comments:
Post a Comment