Where AI Stops in Chartering

We’re an AI company that doesn’t think AI should make most of the decisions.

Strange thing to build a product on. But the more time we spend around shipping, the more convinced we are of it.

Take freight-rate forecasting.

You can train a model on years of Baltic data, FFAs, and historical fixtures. It does reasonably well a month out. But stretch that forecast out to six months and things start to fall apart. One study, linked in the comments, saw errors grow roughly 12x.

Partly that’s the ordinary decay of any long-horizon forecast. But it’s also because the biggest moves in this market often come from events the model simply couldn’t have seen coming. A canal closes. A new round of sanctions lands. A war changes trade flows almost overnight. You can have all the historical data in the world, but you can’t learn a pattern for something that hasn’t happened before.

And that’s why the important calls should stay with the charterers. Which counterparty to trust. Which fixture to walk away from. When a rate is about to turn.

But look at everything surrounding that decision.

Finding the Q88 buried somewhere across four inboxes. Checking a vessel position against three different sources before confirming a laycan. Rebuilding the same comparison sheet because another offer just came in.

None of that needs a charterer’s judgment. It needs accuracy and speed. And, from what we’ve seen, it’s where an enormous amount of the day disappears. That’s the part AI should take on.

Human judgment makes the call. AI clears the path to it.

That’s what we’re building at Precise.

Sources:

– Freight-rate forecast accuracy over longer horizons:  https://doi.org/10.1057/s41278-019-00121-x

– For a broader literature review on ML forecasting limits in shipping:  https://doi.org/10.1057/s41278-025-00334-3