ETA prediction & shipment visibility

Once a shipment leaves, two questions matter: where is it now, and when will it really arrive?

Visibility answers the first. Data comes from GPS devices and telematics in trucks, carrier apps, milestone scans (picked up, departed hub, arrived, out for delivery), port and rail updates for containers, and sometimes toll or e-way bill records. Pulling these feeds together gives one live view of every shipment, instead of phoning drivers.

ETA prediction answers the second. The carrier’s planned arrival time is often optimistic. A predicted ETA uses machine learning on history and live conditions:

• distance remaining and current speed,
• typical traffic for that route, day and hour,
• how long similar trucks usually wait at checkpoints, hubs and customer docks,
• weather, festivals and road disruptions.

Instead of one fixed time, a good model says “likely 4:10 pm, almost certainly before 6 pm”, which is more honest and more useful.

The value is in acting early. If a truck carrying a big store order is predicted to be four hours late, planners can warn the store, re-sequence the dock schedule, or send a few cases from another warehouse before anyone complains. Customers get accurate “arriving at” messages instead of chasing updates.

From naive to predicted ETA

What the model learns from

• Trip history: past trips on the same lane, with actual arrival times.
• Live position and speed from GPS devices, telematics or driver apps.
• Patterns by time: traffic by hour and day, festivals, weather.
• Waiting times: typical delays at toll plazas, checkpoints, hubs and customer docks.
• Rules and constraints: city entry restrictions, driver hours, delivery appointments.

Models such as gradient boosting learn how these combine, and quantile versions give a range (likely by 9:15 pm, almost certainly by 10 pm) instead of one time.

Is the ETA any good?

• Accuracy: the share of arrivals within, say, an hour of the ETA, measured at fixed points before arrival (24 hours out, 4 hours out).
• Bias: do trucks arrive consistently later than predicted?
• Calibration: if the model says “90% sure by 10 pm”, does that come true about 90% of the time?

Acting on it

Predicted delays should trigger actions automatically or through a planner: rescheduling dock slots, informing customers with a new time, re-sequencing warehouse work, or sending stock from elsewhere for urgent orders.

Data gaps and consent

Visibility is only as good as its data. GPS devices fail or get switched off, phone-based tracking depends on the driver’s handset and network, and carriers share data unevenly. Combining sources (devices, apps, toll and e-way bill records through government platforms such as ULIP, carrier updates) fills gaps. Tracking drivers’ phones needs informed consent, and India’s Digital Personal Data Protection Act, 2023 sets duties for handling such personal data.

The yard and the dock

Many delays happen in the last few hundred metres: waiting at the gate, in the yard and at the dock. Visibility that stops at “arrived at customer” misses the hours spent waiting to unload, which drive detention charges and carrier dissatisfaction. Yard and dock scheduling systems, with appointment data, close that gap.

Ocean visibility

For containers at sea, ship positions come from the automatic identification system (AIS) that vessels broadcast. Combining these with port congestion data, schedule changes and transshipment plans gives better arrival predictions than the shipping line’s schedule. Cancelled sailings, port congestion and transshipment delays are common sources of surprise.

From visibility to a control tower

A control tower combines visibility across shipments, orders and stock with rules and people who act on exceptions. The progression is the same as for any analytics: see what is happening, understand why, predict what will happen, and recommend or take action. Most of the value is in the last steps.

Proving the value

Visibility projects should be judged on outcomes, such as:

• fewer customer calls asking “where is my order?”,
• lower detention and demurrage charges,
• better dock productivity from accurate arrival times,
• fewer stockouts from late shipments caught in time,
• lower safety stock where lead times became more predictable.