How AI Resume Screening Works for Maritime Roles

AI resume screening for maritime roles has a trust problem. A crewing manager who has read ten thousand CVs knows that “five years on bulk carriers” and “five years in bulk carrier operations” describe two different careers, and has little patience for a tool that treats them as the same keyword. This guide walks through how screening works inside Marine Ideas, step by step, so you can see what is decided by fixed rules, what is decided by a language model, and where you stay in control.

The short version: the job description is turned into three tiers of skills, every resume is scored against those tiers by rules first, a language model then reads each resume for context, and the two results are combined into a ranking with a written explanation for every candidate.

Why generic screening struggles with shipping CVs

Most screening tools were built for software and sales hiring. Shipping breaks them in three ways.

The vocabulary overlaps. “Operations” can mean vessel operations, port operations or documentation. “Agency” can mean a port agent, a liner agent or a recruitment agency. A keyword match cannot tell these apart.
The same title means different work. An operations executive at a Liner company manages schedules and slot allocation. At a Feeder operator the same title is closer to port rotation and transhipment connections. At an NVOCC it is bookings and bills of lading.
The volume is real. The BIMCO and ICS Seafarer Workforce Report projected a need for an additional 89,510 officers by 2026. When supply is that tight, recruiters cast a wide net, and a wide net means hundreds of resumes per vacancy.

A screening process that works for shipping has to understand company type and skill depth, and it has to be explainable to the person who signs off the shortlist.

What you need before you start

A job description. You can paste text, upload a PDF, start from one of your own templates, or import a profile from the job description library.
Resumes as PDF or text files. Bulk upload is supported.
About ten minutes for the first run.

If you do not have a job description yet, start with our guide on how to write a job description for a shipping company. Screening quality depends almost entirely on the quality of that document.

Step 1: Turn the job description into skill tiers

When you create a job, the description is parsed into three lists.

Must-have skills. A candidate without these should not reach the shortlist.
Good-to-have skills. These separate a solid candidate from a strong one.
Premium skills. Rare, valuable extras that earn a bonus on the final score.

The parser prefers the platform’s own maritime skill taxonomy, so “laytime calc” and “laytime and demurrage calculation” resolve to one skill instead of two.

What you should see: three editable lists on the job page.

Common mistake: leaving twenty-five skills in the must-have tier. Nobody matches twenty-five must-haves, so every candidate scores low and the ranking flattens. Move anything you would train on the job down to good-to-have. Five to eight true must-haves is a healthy range. You can edit the lists by hand at any time before you run the analysis.

Step 2: Upload resumes in bulk

Drop in every resume you have received for the role. Two things happen automatically.

Deduplication. Each file is fingerprinted by its content. The same resume sent by two agencies is scored once, not twice.
Reuse. If a candidate’s resume is already in your organisation’s library, or in the public talent pool, you can attach it to the new job without uploading it again.

What you should see: each resume appearing in the applicants list with a status of parsing, then scoring.

Step 3: Rule-based scoring runs first

This is the part most “AI screening” articles skip. Before any language model is involved, every resume is scored by deterministic rules against your three skill tiers.

Deterministic means repeatable. The same resume against the same job description produces the same score today, tomorrow and next month. No model mood, no randomness. The output is a set of matched skills and sub-scores for each section of the resume.

This matters for two reasons. First, it is fair: every candidate is measured against the same list in the same way. Second, it is defensible: when a hiring manager asks why a candidate ranked ninth, you can point to the specific must-have skills that were missing.

Step 4: AI analysis adds context

Once the rule-based score exists, a language model reads each resume against the job description. Its job is the part rules are bad at.

Pulling out the lines that are most relevant to this particular role.
Summarising strengths and weaknesses in plain language.
Raising red flags, such as unexplained gaps or claims that conflict with each other.
Recognising a skill that is described in different words from the job description.

The model does not get the final say on its own. The final score for each candidate combines the rule-based score, the AI analysis score and the premium skill bonus. If the model is impressed by confident writing but the must-have skills are missing, the rule-based score holds the candidate down.

Step 5: Watch the ranking build in real time

You do not wait for a batch to finish. The job page moves through four visible phases: parsing the job description, scoring, analysing and complete. Candidates appear in the ranked list as each one finishes, and the order updates as more resumes are processed.

This is where the time saving comes from. Screening that used to take a recruiter days of reading is done in minutes, and the recruiter’s time moves to the top of the list, where judgment is needed.

Step 6: Read the report, not just the rank

Every candidate has a scorecard behind the number. It contains:

The relevant lines from the resume.
Matched and missing skills by tier.
Scores for each section and each dimension.
Strengths, weaknesses and red flags.
An audit trail of how the scoring decision was reached.

There is also a comparison summary for the whole job: a written account of how the candidates stack up against each other and against the requirements. It is the page to bring to the hiring manager.

Common mistake: shortlisting the top five by rank without opening the reports. The rank tells you where to look first. The report tells you whether to pick up the phone.

The ranked list exports as an Excel workbook or a CSV, including contact details and assessment scores where candidates have taken an exam. Crewing agencies use this to send a shortlist to a principal who is not on the platform.

How AI resume screening for maritime roles stays auditable

Recruitment is one of the uses of AI that regulators watch most closely, and shipping companies hire across many jurisdictions. An auditable process is not a nice extra. The design choices above exist for that reason.

Decision	Who makes it	Can you inspect it?
Which skills matter and how much	You, by editing the three tiers	Yes, on the job page
Whether a resume matches those skills	Fixed rules	Yes, matched and missing skills are listed
What the resume says in context	Language model	Yes, quoted lines and written reasoning
Final rank	Combination of the above	Yes, sub-scores and audit trail
Who gets shortlisted	You	Always

The same principle holds later in the hiring process. Our piece on what an AI interviewer cannot decide draws the line between software that advises and the people who choose.

Troubleshooting

Everyone scores low. Your must-have tier is too long or too specific. Trim it and run the analysis again.

A candidate you know is strong ranks low. Open the report and look at the missing skills. Usually the resume describes the skill in words that differ from the job description. Either adjust the skill wording in the tier or move that skill down a tier.

The same person appears twice. The two files are different documents, for example an updated CV. Content fingerprinting only merges identical files. Keep the newer one.

The ranking looks right but the hiring manager disagrees. Ask which skill they are weighing differently, then move it between tiers. The tiers are the lever. Changing them and re-running is quicker than arguing about individual candidates.

What screening does not replace

Screening tells you who is worth your time. It does not verify that a certificate is real, check references or assess how someone communicates under pressure. Those are separate steps, and they are stronger when they are designed to follow screening.

For knowledge checks, see how verifiable skill certificates work and how an employer confirms one.
For communication and reasoning, see how an AI voice interview works, including what it cannot judge.

Statutory certification under the IMO’s STCW Convention remains a document check that no screening tool should claim to perform.

Try it on a live vacancy

The fastest way to judge a screening process is to run it on a role you already know well, where you have a clear idea of who the best candidates are. Upload the job description and the resumes, compare the ranking with your own, and read the reports for the candidates where you disagree. Try Marine Ideas free. No credit card is required.