Every ETA is a small promise. Somewhere downstream of that date, a warehouse books a dock door, a planner commits inventory, a salesperson tells a customer “Thursday.” When the date moves, all of those commitments move with it — usually at the worst possible moment, usually by phone.
So it’s worth being precise about where arrival predictions actually fall apart. We looked at 1.2 million ocean shipments from the first half of this year and traced every ETA miss back to a root cause. The shape of the answer surprised even us.
The 72-hour window
First, a framing we use constantly: the 72-hour window. Inside three days of arrival, an ETA should be operational-grade — tight enough to book labor and dock doors against. Outside that window, an ETA is a planning signal, not a schedule. Most visibility disappointment comes from treating a 20-day-out estimate like a 2-day-out one. They are different instruments and should carry different confidence.
Where the error comes from
Inside the window, when ETAs still go wrong, the causes are lopsided:
- Port dwell is the giant. A vessel can cross the Pacific on schedule and then sit at anchor for four days. Any model that ignores berth queues is guessing.
- Transshipment misses are the silent killer: your container arrives at the hub on time and the connecting vessel leaves without it. The ocean leg looks perfect; the promise is already broken.
- Blank sailings — cancelled voyages — punish anyone still reading published schedules as truth.
- AIS gaps matter less than people fear. Position data is noisy, but it’s rarely the reason a date was wrong.
What a trustworthy ETA looks like
An ETA is a promise with a confidence interval — and the interval is part of the promise.
- Publish the confidence, not just the date. “Thursday 14:20, 96%” and “Thursday, 60%” deserve different reactions, and your team knows it.
- Update on signals, not on schedules. Re-predict when a berth queue grows or a gate-out lands late — not once a day at midnight.
- Alert on the delta, not the date. Nobody needs a ping that a shipment is still on time. They need to know the moment the plan slips, and by how much.
The takeaway
ETAs don’t go bad because prediction is hopeless. They go bad in a few specific, watchable places — and most of those places emit signals hours or days before the date officially moves. Watch the signals and the promise holds. That’s the whole product, honestly: one board, watching.