How generative AI fits into a sustainable supply chain.

How Generative AI Fits Into a Sustainable Supply Chain

by Hariesh Manaadiar,
CargoNOW Ocean Freight Features Editor

Two questions are being asked in almost every logistics boardroom this year.

• How do we cut emissions across our supply chain?
• And what are we doing about AI?

In most companies the two sit with different teams. Emissions reporting lands with sustainability or finance, AI sits with IT, and they buy separate tools for overlapping problems.

In this article I go through where generative AI is genuinely useful to a sustainable supply chain, the limits you will hit, and what to have in place before you spend money on it.

What a sustainable supply chain actually means

UN Global Compact, describes supply chain sustainability as the “management of environmental, social and economic impacts”, together with the encouragement of good governance practices, right across the lifecycles of goods and services.

In a freight operation that definition comes down to a short list of things you can count. 

• Carbon released per tonne moved
• How full the equipment leaves
• How far empties travel to reposition
• How much packaging goes around the goods and how much of it is recovered
• How much cargo is damaged, rejected or thrown away
• The conditions people at each link in the chain are working under

For a typical shipper the bulk of the footprint is created in supplier factories, on carriers and by hauliers you do not own and cannot instruct. You choose them, you pay them, you measure them, and that is where your control ends.

 

What is Generative AI

Generative AI produces new content such as text, summaries, tables or recommendations, based on patterns learned from very large volumes of data.

The difference from the analytics tools this industry has used for years is that you ask a question in ordinary language and get an answer in ordinary language.

 

One thing to get out of the way

“Buying an AI tool does not make your supply chain sustainable”, although you will see it presented in board packs as a sustainability initiative in its own right.

Look again at the six measures above. Every one is the result of a decision somebody in your business takes. Which supplier gets the order, which mode the goods move by, what the packaging spec says, and how often cargo is rehandled because the wrong information went out at the start.

Generative AI can inform those decisions and that is the extent of it. A badly designed network stays badly designed, and the technology just describes it faster.

Remember, a model can only work with what YOU feed it.

1. Getting your emissions data out 

Most companies report their own fuel and electricity use without much trouble. The difficulty is everything happening beyond their own gate.

The data you need to account for those emissions sits in other people’s paperwork. Bills of lading, packing lists, invoices, telematics extracts and supplier declarations. Different formats, several languages, and often a scanned photocopy nothing in your building can read.

Generative AI reads that paperwork and pulls consistent fields out of it. Weight, mode, distance, origin, destination, fuel type, material. Once those fields are consistent, the footprint calculation moves from a six month consulting exercise to a monthly routine your team runs in house.

Research published this year found the environmental gains from generative AI came mostly through redesigned, more circular supply chains rather than from faster operations.

2. Cutting out the wasted moves

Empty running, half-filled containers and last minute air freight all cost money and generate emissions. They usually trace back to one cause, finding out too late what is coming down the line.

Scenario work is where the technology earns its keep. Ask what happens to cost, transit time and CO2 if a lane shifts from air to sea and you hold two more weeks of stock. Ask which suppliers ship short often enough to force expedited replenishment.

Those answers still need checking against what your planners know. 

3. Keeping up with the reporting load

Carbon border declarations, deforestation regulations, investor disclosures, customer questionnaires. The load has grown far quicker than the teams handling it.

First draft work is where generative AI performs. Mapping a supplier document against a requirement, flagging what is missing, cutting a two hundred page framework down to the few clauses that touch your operation.

Everything it produces is a draft and needs somebody who understands the regulation to check and sign it. Feed it unverified supplier claims and you will publish greenwashing at speed, under your own name.

4. Designing the waste out before it happens

The biggest wins are locked in before anything moves, when somebody chooses a material, sizes a carton or appoints a supplier.

Generative design tools help at that stage. Repacking a product so more units fit a pallet and more pallets fit a container. Modelling a regional sourcing option against a global one with the freight and carbon numbers alongside.

 

The energy question you will be asked

If you are presenting an AI project as a sustainability win, somebody could ask what the technology itself burns.

The International Energy Agency puts data centre electricity use at around 415 terawatt hours in 2024, roughly 1.5 percent of global consumption, and expects that to double by 2030 with AI as the main driver. It also reports that energy used per AI task has dropped sharply as hardware and software improve.

So test each application before you approve it. Does it remove a physical movement, a wasted kilometre, a spoiled load or a rehandled container.? A dashboard nobody opens fails that test and still runs up the power bill.

Four things to settle before the budget goes in

1. Put a cargo person in charge of the data. Whoever owns the accuracy of what goes in has to know what a bill of lading should say and spot a weight that cannot be right. IT can run the platform but IT cannot tell you that 28 tonnes on a forty foot reefer is nonsense.
2. Start with the worst job in the building. Not the most impressive one. Take the task that is manual, repetitive, high volume and quietly hated, which in most companies is pulling emissions data out of supplier paperwork.
3. Put a name against every number that leaves the building. The drafting can be automated. The answer to where a figure came from cannot.
4. Settle what may go into the tool before somebody settles it for you. Customer contracts, rate sheets, personal data. Write the rule down and circulate it, or you will find out later that your rate structure sits inside a public model.

The summary

Generative AI will not make your supply chain sustainable. It makes the chain VISIBLE, and visibility is what most companies in this industry are short of. You cannot manage a number you are still estimating, and that is what you are doing while the evidence sits in paperwork nobody has read.

So the question for anyone running one of these projects is a narrow one. Is it taking movements, waste and errors out of the chain, or is it producing better looking reports on the same old problems?