Companies have had a doozy of a time adapting to chaotic and unpredictable trade policy. But now, artificial intelligence has entered the chat.

If there is anywhere in this economy where AI could clean up a mess, it’s tariffs. Getting a product into the U.S. was already complicated, before President Donald Trump piled on even more import taxes.

“Every single product that is imported gets classified into one of around 20,000 tariff codes,” explained Todd Smith, CEO of KYG Trade, which produces AI trade software.

Every product has a tariff code, and the government has had a book of them since 1989.

“It’s the size of a phone book,” said Richard Weiner, senior counsel at law firm Sidley Austin. Among those more than 4,400 pages, you can look up the tariff on MRI machines — usually zero. Basmati rice has a tariff of 38 cents per pound, but not if it’s from Jordan or a few other countries, in which case, it’s different.

Sometimes it’s hard to tell which category your product even falls into.

“A smart watch can get classified as, you know, watches. It can get classified as computer equipment, as telco equipment, or even a medical device,” said Priya Rajagopalan, president of product tech at FourKites, which helps companies use AI to manage supply chains.

Oh, and by the way, if companies get it wrong? “They face civil or criminal penalties,” said Weiner. “Criminal is unusual but the Customs Service, they have guns and handcuffs, so you don’t wanna get it wrong.”

On top of everything in that 4,400 page U.S. tariff book are 700 other revenge tariffs the U.S. has on various products it says were dumped into U.S. markets unfairly, and there’s free trade agreements with their own additional rules.

Then came President Trump’s tariffs.

Tariffs by ingredient, like steel. Tariffs by country. But also exceptions for some countries for some products. Plus, they change.

“It’s made it a lot more complicated cause the changes are literally day to day,” said Jenae Ciecko, CEO of Copper Hill, a customs consultancy that’s developed its own AI-powered trade software.

To get the tariff right, importers and manufacturers have to know where a product is actually even from. Sounds easy, but it’s not.

“A manufacturer in Mexico can't, you know, import everything from China, for example, repackage it and call it Mexican origin. That doesn't work,” Ciecko said.

So companies are having to look at where their products are from, and where their parts are from, but also where the parts of their parts are from.

“Electronics is, of course, a classic example of where the ‘bomb,’ as they call it, the bill of materials, truly explodes with hundreds of components per item, and each of those have, you know, many, many vendors behind them,” said Priya Rajagopalan with Four Kites.

The crazy thing is a lot of companies traditionally do figure all this out, but not very efficiently — and not at the speed with which the new tariffs have come down.

“A lot of that work is done by passing unstructured PDFs, emails, Excel sheets via emails, up and down, you know, the supply chain to suppliers and consultants,” said Todd Smith of KYG Trade.

This is one place where AI enters the picture.

“With AI, we're able to now ingest all of this unstructured data and make sense of it really fast,” said Smith.

AI, in some cases, can identify a product and its tariff with just a photo. AI can also help manage the sheer volume of tariff problems. Up until recently, packages worth less than 800 dollars didn’t get tariffed; now they do. About four million of those so-called de minimis packages arrive in the U.S. per day. They didn’t get counted as taxable imports before, now they do.

“Let's say a company previous to the de minimis rule, had 20,000 imports a year, but with the de minimis, change has 30,000 it makes it definitely creates more efficiency for that to be more manageable,” said Jenae Ciecko at Copper Hill.

Now that pretty much every country has a new tariff, a lot of them different, layered on top of old tariff rules, supply chains have become a lot more difficult to figure out, and AI is helping there too.

“What we are seeing is supply chains around the globe reconfiguring themselves right in front of our eyes,” said Madhav Durbha, a group vice president of industry strategy with Relex Solutions, which uses AI to manage supply chains.

For all the glittering new AIs, it’s the old fashioned AIs — machine learning, algorithms, smart math-y software — that are doing a lot of the heavy lifting of calculating the best source with the best price and the best tariff.

Imagine a big box store trying to redo its supply chains.

“They carry anywhere from 20,000 to 100,000 items or so in any single store, right? When you think about the number of stores and number of item combinations, it could potentially be 10s or even hundreds of millions,” Durbha said.

Humans are still necessary — for now — to make decisions, double check, and point the AI in the right direction. AI is proving a valuable tool when paired with humans, but still a tool.

“I just don’t think AI solves everything,” said Jenae Ciecko at Copper Hill. “Not yet. It is not some magic sauce.”

But in the new age of tariffs, it is doing some magic.