Getting crew on board ran on email. So we built an AI agent for the inbox.

Crew embarkation is coordinated by email.

Our client is a leading cruise line operating in the US and Canada. Before a ship sails, crew have to be scheduled to join it, placed in the right roles, and cleared with the right documents.

That coordination happens over email. Each incoming message has to be read, checked against the embarkation schedule, and answered with the right next step.

No two crew members are the same case.

The documents a crew member needs depend on their nationality, and sometimes on their gender or age. What one person has to send is not what the next person has to send.

And plans change. People cancel, need to reschedule, or fall sick. Every one of those needs a correct, timely reply, and often a new plan.

How we came in.

We worked as part of an enterprise technology partner that serves the cruise line. Our team joined the programme and started by understanding the requirements.

From there we designed the architecture and chose the approach: an AI agent that can work through rules, exceptions and changes, rather than a fixed script. Then we built it.

An agent that follows the procedures.

We built the agent on Amazon Bedrock Agents. Its instructions come from the cruise line's standard operating procedures, so it works the way the team already works, not the way a model guesses it should.

To act, it uses Bedrock action groups as its tools. That is how it looks things up, such as the embarkation schedule and the available staff, and how it takes each next step.

What happens when an email arrives.

The agent reads the incoming email and checks it against the embarkation schedule. It works out what is needed: who is joining, when, and what is still missing.

It reviews the available staff and the groups and roles each person can be placed in. Then it contacts the crew member about their schedule and asks for the documents their case requires, based on nationality and, where it applies, gender or age.

When documents come back, the agent reviews them. If something is missing or not right, it asks for more. If a crew member cancels, needs to reschedule or is sick, it handles that change too. Then it replies.

When it isn't sure, a person decides.

The agent does not have to be right about everything on its own. When its confidence is low, the case goes to a person on the team to review.

That keeps people on the cases that need their judgement, and keeps the routine work moving without them.

Where it stands.

The agent we built takes crew embarkation from the first email to the reply: schedule checks, staff placement, document requests and review, and changes such as cancellations, rescheduling and sickness.

It follows the same procedures the team works to, uses tools to act, and leaves the hard calls to people.