AI Jetty Planning for Tank Terminals: ETA, Slots, Cost

Can AI predict vessel ETA and plan jetty slots at a tank terminal?

AI can predict a vessel's arrival at a tank terminal hours before the agent's call and propose the jetty and tank assignment, with the planner approving. Rotterdam's Pronto platform cut average vessel waiting time by 20 percent.

Yes, for the part of the day that now runs on a spreadsheet, the agent's phone call and the planner's memory: a model that predicts when a vessel will actually be alongside, from AIS, the nomination, the port call data, weather and tide, and proposes which jetty, which line and which shore tank take it. The planner approves every assignment. This guide says what changes, which data a terminal already has, where the model fails, and what a first version costs.

By Tom Joseph · Last updated: 14 October 2026

What an arrival model does that the spreadsheet and the phone cannot

Predicts the arrival, not the declared ETA. The agent's ETA in the nomination and in the Portbase notification is a promise. The model combines the live AIS position and speed, the previous ports, pilot and tide windows and the terminal's own history with that vessel and that agent into a probable alongside time with a confidence band, updated every few minutes.

Proposes the jetty, the line and the shore tank together. Jetty draught and length limits, product compatibility of the lines, ullage from the tank gauging system, pump rate, the vapour return connection and the product and quantity in the nomination are all constraints. The model proposes an assignment that respects all of them and shows the planner the alternatives.

Shows laytime at risk before the notice of readiness is tendered. For every nomination it shows when laytime starts counting and which assignment pushes a vessel into demurrage, so the planner weighs a sea-going tanker against a barge on numbers instead of on who calls loudest.

Replans when the day moves. A delayed vessel, a slow pump, fog at the pilot station, a tank not yet released after gauging: the schedule is recomputed and the planner sees what changed and why. The confirmations to the agent, the surveyor and the customer still go out from the planner.

AI or the terminal planner

Decision	AI	Terminal planner
Arrival time	Predicts the alongside time from AIS, port call data, tide and history, with a confidence band	Confirms with the agent, decides when to call the surveyor and the jetty crew
Jetty and tank assignment	Proposes an assignment on draught, length, product compatibility, ullage and pump rate	Approves, overrides for what the model cannot know: a customer's preference, a maintenance crew still on the jetty
Conflicts and demurrage	Shows which nominations run into laytime and what each swap costs	Decides the trade-off and informs the customer
Safety clearance	Nothing. The ISGOTT ship/shore safety checklist, PGS 29 and the Seveso duties stay with the terminal and the vessel	Jetty operator and loading master sign the checklist as they do today

The data a terminal already has, and the systems it lives in

Everything the model needs exists at the terminal already; it is scattered. The terminal management system holds the nominations, the jetty schedule and the tank movements: Implico OpenTAS, Emerson Syncade Logistics (which Emerson developed with Vopak and first ran at Vopak's Amsterdam Westpoort terminal in 2012, Digital Refining) or a system built in-house. The tank gauging system holds ullage and temperature per shore tank. Portbase Vessel Notification and the Harbour Master's HaMIS hold the declared ETA and the berth order; in Rotterdam the agent must notify 24 hours before arrival and forward every change of 30 minutes or more (Port of Rotterdam), and the agent is responsible for keeping PCS and HaMIS identical (Harbour Master FAQ). An AIS feed gives position and speed. Add tide and weather from Rijkswaterstaat and KNMI, and the terminal's own record of alongside times, pump rates and laytime statements for the last two to three years.

The build is two models and one screen. An arrival model that learns, per vessel class and agent, how declared ETAs relate to actual alongside times. An assignment model, mostly constraint solving, that proposes jetty, line and tank for the next 72 hours. And one screen for the planner that shows the proposal, the alternatives and the laytime exposure, and writes the approved assignment back into the TMS so nothing lives in a second system. The sector is moving this way: Evos Hamburg launched DigiTank in January 2025, a three-year research programme under IHATEC II that combines digital twins, AI and robotics to reduce control room stress and optimise operations (Tank Storage Magazine), and Vopak reported in March 2025 that algorithm-driven berth scheduling at its Qinzhou terminal replaced manual scheduling and improved berth utilisation by more than 50 percent, a company figure from one terminal that it says it plans to replicate elsewhere (LNG Industry).

What an arrival model does not fix

It does not add a jetty, a pump or a tank. A terminal that is physically full on Thursday is full with a better schedule too; the model shows the congestion earlier, it does not remove it. It does not fix nominations that arrive as PDFs by email with the quantity in the subject line; reading those is a separate, smaller job that the audit scopes first. And it cannot predict what no data carries: a strike at the previous port, or a cargo the trader moved on the phone and the terminal heard about afterwards.

It is not a safety system. The ISGOTT 6 ship/shore safety checklist, which separates what the vessel and what the terminal is responsible for from pre-arrival to departure (OCIMF), the PGS 29 requirements on the tank and its overfill protection, and the Seveso III duties that now sit in the Besluit activiteiten leefomgeving under the Omgevingswet are not changed by a planning model and must never depend on one. The model proposes a slot; the loading master still walks the jetty. The model is also only as good as the terminal's own record of when vessels were actually alongside, started pumping and finished: where that lives as free text in the shift log, the first weeks are spent turning it into time stamps.

And the planner stays. A schedule the customer service desk does not trust is a schedule nobody follows; the model earns trust by showing why, nomination by nomination.

What it is worth at a 40-tank terminal, and what it costs

Volumes are not growing for you: Rotterdam handled 428.4 million tonnes in 2025, with liquid bulk down 1.5 percent and other liquid bulk, including chemicals, down 3.1 percent to 34.3 million tonnes (Port of Rotterdam Authority). The margin is in the utilisation of the jetties you have. Take an independent terminal in the Rotterdam port area with 40 tanks, four jetties and 600 sea-going and barge calls a year. Assumptions, stated so you can replace them with your own: a planner and a deputy spend half their day on arrival calls and replanning; one in five calls arrives more than four hours from the declared ETA; and 5 percent of sea-going calls end in a demurrage claim, settled at an average of €15,000. That is about 30 claims and €450,000 a year, before the jetty hours lost to a berth kept free for a vessel still at the pilot station. If better arrival prediction and assignment remove a third of those claims and free one planner-day a week, the terminal keeps about €150,000 a year in demurrage plus the planner's time. The audit replaces every one of these assumptions with the terminal's own laytime statements. These are illustrative figures, not a client result.

Three fixed steps at Crux Digits, each leaving the terminal with something it keeps. An AI audit and strategy of €2,500, one to two weeks, on twelve months of nominations, jetty schedules, AIS history and laytime statements: the terminal keeps a measured ETA error per vessel class, a data map and a scoped plan, and the fee is credited in full if the work continues within 60 days. A Production-ready MVP of €20,000, four to six weeks: the arrival model and the proposal screen run for one jetty group next to the current spreadsheet, and the terminal keeps the code, the models and a measured comparison. Production from €50,000, from eight to twelve weeks: integrated with the TMS, the gauging system and the Portbase feed, replanning in the planner's hands, handed over to the terminal's IT team and its ICT partner. All prices excl. VAT. See AI implementation, machine learning, data engineering and AI for logistics; terminals that also face grid limits on electrifying heating and pumps can read grid congestion in industry.

Frequently asked questions

Can AI predict when a tanker will actually be alongside?

Yes, within a band that narrows as the vessel approaches. The model learns from AIS, port call data, tide and the terminal's own history how declared ETAs relate to real alongside times, per vessel class and agent. The agent's ETA remains the formal notification; the prediction is what the planner plans on.

Does it replace the terminal planner?

No. It proposes the jetty, the line and the tank and shows the laytime exposure; the planner approves, overrides for what the model cannot know and sends the confirmations. Crux Digits builds it so the approved assignment is written back into the TMS, not into a second system.

Which systems does it have to connect to?

The terminal management system (Implico OpenTAS, Emerson Syncade Logistics or an in-house system), the tank gauging system, the Portbase Vessel Notification and HaMIS data in Rotterdam, an AIS feed, and tide and weather data. The audit maps which of these the terminal already stores and for how long.

Does it change our Seveso, PGS 29 or ISGOTT obligations?

No. A planning model proposes a slot; the ship/shore safety checklist, the PGS 29 requirements on the tanks and the Seveso duties in the Besluit activiteiten leefomgeving stay with the terminal and the vessel, exactly as today, and must not depend on the model.

Are tank terminals behind on AI?

The sector is. In 2024, 11 percent of Dutch transport and storage firms with 10 or more employees used AI, against 22.7 percent across all sectors (CBS AI Monitor 2024); in 2025 the Eurostat figure for the sector was 19.6 percent. At Crux Digits a first version costs a €2,500 audit, a €20,000 Production-ready MVP and production from €50,000, all excl. VAT, with code and models handed over.