Talk about AI in procurement tends to jump straight from chatbots to fully autonomous supply chains. In practice there are four stages in between, and the companies that get ahead will be the ones that go through them in order.

In the previous article, we followed a lubricant buyer through a typical Wednesday in 2030. Routine orders ran themselves, forecasts flagged cost rises months ahead, and supplier meetings started at the negotiation stage. The question now is how to get there.

It starts with an honest view of how mature the technology is. AI in procurement develops in four stages, and each gives the technology more autonomy than the one before. The stages build on each other, so skipping one isn't a shortcut. You cannot go from zero to coordination.

Stage 1: Assist

This is where most procurement teams are today, and that's fine. It's the foundation. The task is narrow and clearly defined, and a person checks the result. Matching an invoice to its purchase order isn't glamorous, but it's where a team learns what AI does reliably and where it slips. Every later stage depends on that knowledge.

Stage 2: Advise

At this stage the value moves from saving time to making better decisions. AI brings in information no buyer has time to gather: price trends across ports, delivery performance, and the true cost of small or urgent deliveries.

The biggest gain comes from joining procurement data to technical data. If a lubricant category comes in under budget while the engines it protects cost more to run, that isn't a saving. The cost has just moved to another budget line. Or when you link product cost to supplier performance to examine cost-to-value. Only a system that can see both sets of data will catch it. This is where "the cheapest oil isn't always the cheapest" goes from a gut feeling to something you can prove.

Stage 3: Act

This is the big shift. Autonomy doesn't mean AI following your instructions faster. It means AI making decisions, including ones you wouldn't have made.

For lubricant buyers, that could mean the system places a routine cylinder oil order itself. It picks the port, the quantity and the approved supplier from stock and consumption data, and brings you only the exceptions. That's a very different relationship with the technology, and it needs a different kind of preparation.

Autonomy has a price. Anthropic's engineering guidance on building AI agents warns that more autonomy brings higher costs and the risk of compounding errors. In a process with several steps, a small mistake doesn't stay small. A wrong stock reading leads to the wrong quantity. That leads to the wrong port, then the wrong supplier, and finally an urgent delivery nobody wanted. The lesson is to start with the simplest approach that works, test thoroughly, and set guardrails before giving AI more freedom. Companies that skip this step won't save time. They'll pay for it later.

Stage 4: Coordinate

This stage holds the biggest structural change for procurement. When buyer and supplier systems share expected demand, suppliers can plan production and position stock in the right region before it's needed. Emergency deliveries and last-minute premium pricing both fall. The buyer–supplier relationship also changes, from a series of one-off transactions to a working partnership built on shared data.

What this looks like at scale

These stages already exist in practice. US logistics group C.H. Robinson has worked its way through them, and its published results show what's possible:

• By October 2025 it was running more than 30 connected AI agents. In September 2025 alone, one of them captured about 318,000 freight-tracking updates from phone calls, information that used to be lost in unstructured conversations (Business Wire).
• The company reports productivity gains of around 40% from this approach (C.H. Robinson via Bloomberg).
• In June 2026 it added a Lean AI Engineer to its Lean AI Planner. Staff now set the outcome and the AI improves the operation itself. The company says it now has hundreds of agents and handles 92% of its managed shipments autonomously (Logistics Management).

In Philippe's words: "We're not talking marginal gains here. We're talking about changing the game."

But the lesson for shipping isn't in the headline numbers. C.H. Robinson owns few physical assets, is built around technology, and started building its AI models in 2023. Its results can't simply be copied into a shipowner's procurement team. What can be copied is the order: first single tasks, then connected agents, then systems driven by outcomes. The company didn't skip a stage, and nobody else will be able to either.

Where are you?

Five questions for your team:

1. Do you have at least one well-defined procurement task where AI already helps every day? If not, you're before stage 1. Start there.
2. Do you receive automated recommendations based on lubricant spend, stock on board and oil-analysis data in one place? If not, stage 2 will stall.
3. Have you written down the rules an automated order must follow: approved suppliers, price limits, quantities, ports?
4. Do you know how you would spot an AI mistake before it reaches a vessel?
5. Would your suppliers use your demand forecasts if you shared them?

The companies that reach 2030 first won't be the ones with the most ambitious AI strategy. They'll be the ones that worked through the stages deliberately, one at a time.