I’m not an AI wizard — just someone trying to make sense of where this technology really works. 

This week I went down a rabbit hole on edge AI in manufacturing — GPU-powered devices running models locally on factory floors, making split-second calls without cloud connectivity. That’s proper Operational Technology (OT) integration: silicon at the coalface. 

Then I got sidetracked by a very different thought: corporate “transformation,” customer-journey mapping, and the ongoing struggle to deliver value. 

Two Faces of AI Feedback Loops 

1) Manufacturing: AI needs speed at the edge. 
On the factory floor, bearings don’t wait for the cloud. Sensors, yield checks, and quality gates already create fast, closed loops. AI’s job is to work inside those loops — spotting defects, predicting failures, and keeping processes stable. That’s why AI makes its split-second calls on the factory floor, while the training and governance sit in the cloud. 

2) Corporate: AI needs evidence for benefits. 
In offices, the challenge isn’t speed — it’s proof. Too many projects start with vague goals like “more sales” or “better collaboration,” but without hard measures no one trusts the results. AI here depends on something simpler: define success in terms of a customer or employee journey, capture the data that shows it happened, and link actions to outcomes people believe. Until you do that, the model doesn’t matter. 

The common thread: build for your feedback loops 

Put simply: factories need decisions in milliseconds; corporates need evidence they can trust. Either way, the model itself won’t matter if you can’t sense, decide, and close the loop. 

AI is only as strong as the loop it runs in. In the factory, that means bringing compute to the data so decisions happen in milliseconds, then syncing learnings back to the cloud. In the office, it means building the missing loops — defining outcomes, instrumenting the customer and employee journeys, and proving benefits people can see. Either way, AI isn’t something you bolt on; it has to be part of the system you design. 

So what? 

Before you fund an AI project, answer these four questions in a single line each: 

• What proves it worked? (one outcome you already track) 

• Where must the decision happen? (factory floor, branch, contact centre, browser) 

• What evidence will show it? (the data points you’ll capture) 

• What changes automatically when it triggers? (stop, re-route, re-price, retrain) 

• If you can’t answer all four in one line each, you’re not ready to begin the project — build the feedback loop first. 

Why this matters now 

AI is shifting from being a tool you plug in to a system you design around. In factories, that shift has already happened: edge AI makes instant calls on the floor while the cloud handles learning. In offices, the same shift means treating measurement like infrastructure — the digital version of sensors and quality gates. 

The real question isn’t cloud or edge; it’s whether you’re building for the feedback loops you already have, or creating the loops you need. In the factory, signals are clear and the job is speed. In the office, signals are messy and the job is proof. 

Start with how decisions flow — then pick the software.