AI and automation are at the top of nearly every terminal operator’s agenda. The pressure to act is real and is coming from boards, customers and competitors. Yet most terminals find themselves caught between urgency and uncertainty, hearing the same question repeated in planning meetings, “Where do we begin?” By Subbu Bhat, Interim President & Head of Engineering, Tideworks Technology
Why AI Fails: The Trust Problem
AI adoption in terminal operations is still in its early stages. The most common applications today, such as safety monitoring and logistics forecasting, are valuable but limited in scope. More broadly, operators are still defining where AI belongs in real operational workflows.
This hesitation is often framed as caution. More constructively, it reflects a natural need to build confidence in new capabilities.
The industry has seen this pattern before. Large-scale automation initiatives promised transformational results but struggled to deliver in practice. Not because the technology failed, but because integration into day-to-day operations took time to mature.
The same dynamic is now playing out with AI.
Deploying AI as a full decision-maker too early leads to a predictable outcome. The workforce needs time to understand and adopt it, adoption stalls, and the investment takes longer to realise its full value.
What is emerging instead, both in terminals and across the wider supply chain, is a more pragmatic model: AI should assist before it recommends, and recommend before it automates. This progression builds familiarity, confidence and measurable results over time.
A Framework for Trustworthy AI Adoption
Five principles should guide how terminals approach AI. Together, they describe not just what to deploy, but how to deploy it in a way that operators will embrace and scale.
1. Audit before you automate
The most valuable first step requires no new technology. Terminals need to understand where time and effort are actually going.
A practical method is the “random moment” time study, periodically asking staff:
- What are you working on?
- Which customer or service is it for?
- How long will it take?
Patterns emerge quickly. Operators can identify which customers generate the most manual work, which services create the most exceptions, and where staff spend time correcting data.
These insights highlight where automation will deliver measurable returns and where fixing a process should come first.
2. Treat data quality as a prerequisite
Every AI initiative runs on the data beneath it. If that data is inconsistent, incomplete, or siloed, the output will reflect it. AI does not filter out noise; it amplifies it.
Investing in data quality, including system connectivity, process consistency, and clear ownership, pays off before any AI application goes live. Terminals with strong data foundations move faster and see better results because they are building on reliable inputs.
3. Sequence adoption: Assist, Recommend, Automate
This is the most often skipped step and the most critical to long-term success.
- Assist: AI supports human decision-making by aggregating data a planner cannot compile quickly. For an incoming vessel call, it can analyse past voyages, weather patterns, crane availability, and gate activity, surfacing a recommendation in seconds. The planner reviews and decides, building confidence through repeated accuracy.
- Recommend: As trust and confidence build, AI begins to provide earlier, more forward-looking guidance—improving yard positioning, labour planning, and reducing last-minute exceptions. The planner still owns the decision.
- Automate: Automation is applied to specific, well-defined tasks only after trust and consistent value have been established.
This progression is about making adoption sustainable. Skip it, and the result is familiar: tools that technically work but go unused because they are not fully embedded into operational flow.
4. Keep humans in the decision loop
At this stage of AI maturity, the goal is not to replace human decision-making, but to strengthen it.
Significant operational gains are available through AI-assisted decision-making while preserving human expertise where it matters most.
Maintaining human oversight ensures accountability and accelerates adoption as teams gain confidence in outcomes.
5. Require explainability
AI must be understandable to be usable.
A black-box recommendation gives a planner no basis for evaluation, only a choice to accept or reject. A transparent system provides context. For example, it might flag that multiple vessels are likely to converge due to weather delays and suggest a labour adjustment, along with the data driving that recommendation.
This visibility allows planners to assess, challenge, and act with confidence. It also accelerates adoption. Systems that are understood are more likely to be trusted, refined, and scaled across operations.
What This Looks Like in Practice
Tideworks is currently working with a terminal in Panama on a machine learning model that estimates container pickup time. The system draws on historical data, including past vessel calls, gate appointment patterns, and trucker completion rates, refining its probability window as new information arrives. When a gate appointment is made against a container and historical data shows 80% of truckers meet their appointments, the system reflects that in real time.
This is the Assist phase in action. The output is a better-informed planner, not a replaced one. The model provides a data set that improves yard planning and labour decisions. It does not make those decisions. Over time, as the model proves its accuracy, trust accumulates, and the scope of what it can do expands.
A similar principle underlies automated gate processing that Tideworks has deployed at West Coast terminals. By integrating appointment data with internal systems, gate transactions that once required a clerk can now be completed without manual intervention for the majority of standard cases. Exceptions still go to humans. The automation handles what is routine and predictable; people handle what is not.
Start with Foundations, Then Scale
The terminals getting AI right are not the ones moving fastest. They are the ones that started with a clear-eyed assessment of where their operation actually loses time and money, invested in the data foundation that AI requires, and introduced tools in a sequence that let trust and confidence grow alongside performance.
The pressure to act is real. But acting without this foundation produces the same outcome every time. Tools that technically function but operationally fail, because they are not fully adopted. Build confidence and trust first. The automation follows.