How AI is reducing manufacturing costs in unexpected ways

Ask most people where AI will reshape manufacturing and they picture autonomous factories, self-healing production lines and fully automated supply chains. Over the last few months we asked a narrower question: where is it saving real money today? 

We spoke with more than a dozen manufacturing companies in sectors such as automotive, industrial machinery and construction equipment, as well as industry experts and consultants, in order to find out. The wins were not futuristic at all. They were hiding in machine code, customs paperwork, supplier data, inventory records and logistics processes that no one had looked at in years. In an instant, AI managed to dust the cobwebs away to find big wins amidst the attic dust. 

Individually, each win looks like a small improvement. Together, they can unlock millions in savings and meaningfully lift operational performance.  

Two examples stood out. 

Conservative machine settings create hidden manufacturing costs

Every machining centre runs on instructions known as G-code and M-code. These programs tell the machine exactly how to cut drill and shape a component. 

Most manufacturers run the programmes originally supplied by their equipment makers, firms like Mazak or LMW. Those programmes are written with one priority: reliability. A machine builder would far rather run a machine conservatively for 10 years than risk a breakdown from pushing it too hard. So the safety margins are generous and they are baked into thousands of programs across a factory, often unchanged for years. 

The result is that many machines operate well below what they are capable of. One expert told us it is common to find a machine that can run at 1,000 revolutions per minute programmed to run at 400 or 500 RPM. 

AI finds faster machine operating points that retain quality 

This is where AI is starting to help. An engineer can now feed the existing machine code, the part drawing and the required tolerances into a model that flags where speeds and feed rates were set conservatively. Instead of manually reviewing hundreds of lines of code, they get a shortlist of places worth a second look. The aim is not to run machines flat out. It is to find a faster operating point that still holds quality. 

The impact can be real. Several manufacturers in fields including cars and industrial machinery reported cycle time reductions of 8% to 10% with no change to the component design, no new equipment and no change to the line. 

There is a catch worth naming. Push a spindle speed too far and you scrap parts or wear out tooling faster than you save, so the better implementations treat the AI’s suggestions as candidates for an engineer to validate rather than instructions to apply blindly. Used that way, the appeal is obvious: the machine, the tooling, the factory and the workforce all stay the same, while only the instructions get better. 

For an industry that tends to equate higher productivity with heavy capital spend, that is a different proposition. The value comes not from buying new assets but from using the ones you already have more intelligently. 

AI can reduce customs duties through smarter product classification

The second area is customs classification. 

For any company with a global supply chain, import duties are a large share of total product cost. Yet classification is a specialised, judgement-heavy task and how a component is described can carry real financial weight. 

One example involved a track adjuster used in heavy equipment, typically to maintain the right tension on heavy-duty tracks. Because the component contained a spring, it had been classified as a spring, which carried an import duty of around 10%. That classification had sat unchallenged for years. 

The company ran the part’s technical specification against customs databases and classification rules. The analysis suggested it should be classified by its primary function, maintaining tension, not by one internal element facilitating such a function. The revised rate fell from 10% to 2.5%. 

Nothing about the product changed, nor the supplier, nor the process. The company simply cut roughly 6% to 7% from the total part cost. 

The wider point matters more than the single saving. A reclassification is not free of risk. Get it wrong and you can invite an audit or a back-duty claim, so this works best as a flag for a customs specialist to confirm, not a decision to automate. But it points to something large. Manufacturing organisations accumulate thousands of decisions over time, from customs codes and supplier choices to machine settings and process parameters. Most were sensible when they were made. Very few are ever revisited. 

The real opportunity for AI in manufacturing: improving everyday decisions

What strikes us about both examples is what they don’t involve. No one is replaced, no factory is redesigned and no large cheque is written. The work is the unglamorous business of removing waste. 

This is also why these inefficiencies survived so long and the reason is more interesting than: nobody noticed. People often did notice. The problem was that re-examining a decision used to cost more human attention than it was likely to save – so reasonable choices calcified and a factory might revisit a handful of them a year. 

That is the real shift. AI does not change what a good decision looks like. It collapses the cost of re-checking decisions, so a manufacturer can now revisit thousands at once rather than a few. Manufacturing has always been a discipline of continuous improvement, through lean, Six Sigma and operational excellence. This is the same instinct, applied at a scale that was never practical before. 

The first wave of value here is quiet. It is the slow work of finding decisions that were right once and asking whether they still are. The future of AI in manufacturing may arrive not as one dramatic breakthrough, but as thousands of small decisions made better every day.