Over the past few years artificial intelligence has become the fixed headline of shipping conferences, investor decks and software launches. The interest is not misplaced: the industry produces large volumes of data, from vessel positions and fuel consumption to charter parties and port records, and much of that data is still handled by hand. But the “AI” label gathers technologies of very different maturity under a single expectation. Some applications do measurable work in daily operations today; for others the timeline is measured in decades rather than years.

This article looks at where AI is genuinely used, where expectation runs ahead of practice and what determines the gap, from the point of view of small and mid-sized owners and operators trading on voyage charter. For companies of this size the real question is not “should we use AI?” but “for which task, with which data, and under whose control?”

Key concepts

  • Artificial intelligence (AI): The general name for software that performs tasks normally requiring human judgement, such as understanding text, recognising patterns or forecasting. It is not a single technology but a family of methods.
  • Machine learning: Methods that learn from patterns in past data rather than following rules written by hand. Arrival time prediction and fuel consumption modelling are typical examples.
  • Large language model (LLM): A model trained on very large bodies of text that can read and summarise documents, extract structured information from them and generate new text. It is the source of today’s progress in document work such as reading charter parties and statements of facts.
  • Hallucination: A language model confidently producing information that is plausible but not present in the source. It is the main reason AI output must be verified in document work.
  • Human-in-the-loop: A way of working in which AI produces a suggestion or a draft, and the decision and approval stay with a person.
  • AIS (Automatic Identification System): The system through which vessels broadcast their identity, position, course and speed at regular intervals; mandatory for ships covered by SOLAS Chapter V, Regulation 19. It is the raw data of many AI applications in shipping.
  • Digital twin: A virtual model of a vessel or system, fed by sensor data. Used in performance monitoring and maintenance planning.

The value of an AI application depends largely on two things: the quality of the data and the cost of an error. Where data is plentiful and consistent and errors are easy to spot and correct, AI matures quickly. Where data is sparse or scattered and errors affect lives, property or legal outcomes, progress is slow, and it should be. The assessment below rests on these two criteria.

Real use and expectation in five areas

1. Document reading and data extraction

Real use: Extracting dates, times, rates, ports and clauses from unstructured documents such as charter parties, statements of facts (SOF), bills of lading, invoices and emails is today the most mature AI application in shipping. Language models can substantially reduce the manual data entry done by operations teams and turn documents in different formats into a common structure.

Expectation: A common expectation is that the document is “read and the calculation comes out by itself”. Riders in charter parties, handwritten remarks, poor-quality scans and the risk of hallucination mean the output still has to be checked by a person. For calculations that turn directly into money, such as laytime and demurrage, the sound approach is for AI to read, a person to approve and the calculation to be made by a rule-based, auditable engine.

2. Routing, arrival time and market forecasting

Real use: Weather routing has been an established service for a long time. Machine learning is now widely used, together with historical AIS data and weather forecasts, to improve estimated time of arrival (ETA) predictions and route recommendations. Monitoring port waiting times and congestion also belongs here.

Expectation: Expecting forecasts to be “exact” is unrealistic. Factors whose data is not publicly available, such as port operations, agency communications, cargo readiness and local restrictions, limit forecast quality. AIS data can also contain coverage gaps, delays and incorrectly entered information. Freight market forecasting is another frequent expectation; yet the cyclical nature of shipping markets, driven by supply and demand shocks, expectations and lagged responses, keeps forecast uncertainty high regardless of the method used.

3. Fuel, performance and emissions

Real use: Modelling fuel consumption against speed, loading condition, weather and hull fouling, detecting performance loss (such as hull fouling) early and automating emissions reporting are mature applications. IMO’s Carbon Intensity Indicator (CII) from 2023, the inclusion of shipping in the EU Emissions Trading System (EU ETS) from 2024 and FuelEU Maritime from 2025 have markedly increased the need for data in this area.

Expectation: Models are only as valuable as the data that feeds them. On a vessel relying only on noon reports, with no sensor data, the gain from advanced models remains limited. For small fleets the first step is often not AI but collecting data regularly and consistently.

4. Predictive maintenance

Real use: Early detection of deviations that could lead to failure, using vibration, temperature and pressure data from main and auxiliary engines, is applied particularly on newer vessels equipped with sensors.

Expectation: The expectation of a system that “knows” about failures in advance depends heavily on investment in data and hardware. Missing sensors on older vessels, manufacturers’ closed data formats and the lack of enough failure examples for a model to learn from are the main limits here. Compliance with classification society and manufacturer rules is a separate matter to be addressed.

5. Autonomous ships

Real use: What is actually in use today are short-range, controlled pilot projects and systems that assist the officer of the watch: collision avoidance aids that detect other vessels and objects from camera images, and decision support displays.

Expectation: The expectation of uncrewed, ocean-going commercial shipping is as much a legal question as a technical one. Following the regulatory scoping exercise it completed in 2021, IMO is working on a code for Maritime Autonomous Surface Ships (MASS). Liability, insurance, port state control and the adaptation of international conventions call for a longer timeline than the technology itself. There is no reason today for a small or mid-sized owner on voyage charter to shape near-term decisions around this expectation.

A workable approach for small and mid-sized owners

The common lesson of the five areas is that the value of AI comes less from the technology itself than from choosing the right task and having that task’s data. The approaches that stand out in the literature and in industry practice can be summarised as follows:

  1. Start from the task, not the technology: Identify the work that needs the most manual data entry and where errors cost money, such as document reading, demurrage calculation or certificate tracking.
  2. Data first: Collect noon reports, statements of facts and documents consistently and in one place. The value of AI is built on that foundation.
  3. Human oversight: Have a person approve every AI output at the point where it has financial or legal consequences.
  4. Auditability: Record the reasoning behind every calculation and decision, so that the charterer or an arbitrator is presented with the documents, the calculation and the reasoning, not “the model said so”.
  5. Start small and measure: Begin with a pilot on a single process whose before and after can be measured, such as processing time, number of errors and amounts collected.

Conclusion

In shipping, AI delivers its fastest, quietest and most measurable benefits not in the headline areas but in document and data work. Expectations such as autonomous ships and exact market forecasts face legal and economic limits as much as technical ones. For a small or mid-sized owner, the realistic gain is less manual work and fewer errors under human oversight. In most cases that starts not with a new technology, but with regular and verifiable record-keeping.

This article is for general information and is not an assessment of any particular product or service. Regulatory timelines change; the current position should be confirmed from the official sources of the relevant bodies.

References

  • IMO, SOLAS Chapter V, Regulation 19: Carriage requirements for shipborne navigational systems and equipment.
  • IMO, Resolution A.1106(29): Revised Guidelines for the Onboard Operational Use of Shipborne Automatic Identification Systems (AIS), 2015.
  • IMO, Resolution MEPC.328(76): 2021 Revised MARPOL Annex VI (EEXI and CII).
  • IMO, MSC.1/Circ.1638: Outcome of the Regulatory Scoping Exercise for the Use of Maritime Autonomous Surface Ships (MASS), 2021.
  • European Union, Directive (EU) 2023/959 (amending the EU Emissions Trading System to cover maritime transport).
  • European Union, Regulation (EU) 2023/1805 (FuelEU Maritime).
  • NIST, Artificial Intelligence Risk Management Framework (AI RMF 1.0), 2023.
  • Russell, S. and Norvig, P., Artificial Intelligence: A Modern Approach, 4th edition, Pearson, 2021.
  • Stopford, M., Maritime Economics, 3rd edition, Routledge, 2009.