How to Evaluate a Maritime Data Provider: A Procurement Guide

Nearly all the data behind a maritime operating picture originates outside the platform that displays it. AIS comes from transponders and receivers. Flag and registry data comes from administrations. Ownership, casualty, and adverse media records come from specialist sources. The picture is only as reliable as the weakest of those inputs and every one of them is now under pressure.

Of 642 vessel transits logged by Windward through the Strait of Hormuz in August 2026, 393, or 61.2%, never happened. They were artifacts of GPS jamming, while real traffic volume was essentially unchanged. An agency working from raw AIS that month would have counted 642 transits where roughly 250 took place.

Identity data degraded more quietly. By the end of Q3 2026, the IMO database listed 613 falsely flagged vessels, up 39% since the start of the year. Together they total 24.8 million gross tonnes, 95% of it tankers, and 569 of the 613 were built before 2010.

Ownership data remains the deepest gap. 78% of the tankers in Windward’s dark fleet are sanctioned by the U.S., UK, or EU, yet the beneficial ownership of roughly 60% of them remains unknown.

In each case the data was flowing. It was simply wrong or incomplete, and only a provider built to detect and correct for that would have known.

A Working Framework: Information, Data, Intelligence

Three terms get used interchangeably in procurement, and they shouldn’t be.

Information is unstructured facts: a news report of a seizure, a port agent’s note, a registry notice.

Data is information structured so it can be queried and compared: a vessel’s position history, its flag record, its registered owner.

Intelligence is what results when multiple data sources are reconciled, weighted for confidence, and assessed against behavior. It answers a question the operator actually has.

Acquiring data is necessary but not sufficient, because data answers questions nobody is asking. A position feed says where a vessel was at a point in time. A registry says who the registered owner is. A sanctions list says whether that name appears on it. None of those is the question an operator has, which is closer to: is this vessel about to enter my area of responsibility, is its declared cargo plausible given where it has actually been, and does the company behind it connect to a designated entity two layers deeper than the registry shows.

Answering that means joining the position, feeding the ownership record to the behavioral history to the sanctions list, deciding which source to trust where they disagree, and attaching a confidence level to the result. Structured data such as company ownership is also among the hardest to acquire, which is why how a provider sources and maintains it matters. But acquisition is the entry fee. The differentiator is whether the data becomes an answer the operator can act on.

The four questions below test all three layers.

1. What Happens When a Data Source Degrades or Disappears?

Every data type has its own failure mode.

Positional data degrades chronically. Terrestrial AIS receivers cover only 40 to 60 nautical miles from the coast, and satellite AIS adds latency and drops legitimate signals in dense traffic. It also degrades acutely: within 24 hours of hostilities in the Middle East Gulf, more than 1,100 vessels experienced GPS and AIS interference, with positions displaced onto airports and inland locations. 

Across Q3 2026, GPS jamming reached 124,235 distinct vessels worldwide, up 11.7% on Q2, and in the Middle East Gulf it spiked with each U.S. strike on Iran, peaking at 1,699 incidents on September 1 against a quarterly daily average of 315.

Structured data sources fail differently. Data providers change terms, get acquired, or exit the market, and a platform with a single source for a data type loses it entirely.

Partial degradation can be worse than a full outage, because nobody notices it. On May 6, more than 100 vessels were anchored near Fujairah. AIS showed about 10 of them. Anyone looking at the map saw ships and had no reason to think 90 were missing. The same thing happened in the Bab el-Mandeb in August and September. Containerships from carriers on the Houthi target list kept crossing, 142 of them, and 5 had AIS on.

Three failure modes, three different questions for a provider:

• How many sources sit behind each data type?
• How does a missing source get replaced?
• How does the platform know when a source is still delivering but delivering less? 

The request below covers all three:

What to request: Confirmation that each data types are backed by more than one independent source.

A strong answer names redundant sourcing by data type, explains how degradation is detected before a user reports it, and describes how users are told when a data type is incomplete.

Red flags: “We’ve never had a data issue.” One source for any critical data type. A global coverage map offered in response to a regional question.

2. How Quickly Can a Data Gap be Closed, and Can They Prove It?

How fast a provider closes a data gap depends on two things it decided long before the gap appeared: whether it already knows where the replacement data would come from, and whether its platform can plug in a new source without being rebuilt around it.

What to request: One real case, with three dates: when the gap was detected, when the fix was decided, and when the data was restored.

A strong answer provides all three dates and describes what changed for the end user.

Red flags: A hypothetical in place of an example. A fix date with no detection date, which usually means the customer found the problem first.

3. How Is Data Quality Maintained When Sources Disagree?

This question separates a data reseller from an intelligence provider. Any provider working at scale gets conflicting records every day, and more and more of those conflicts are deliberate. 

The data types below show how, and what the conflict costs the analyst when it is not resolved:

Ownership: Dark fleet ownership is hidden on purpose, behind layered shell companies and nominee structures built to slow down attribution. So two ownership databases will often name two different companies for the same vessel. One lists the registered owner. The other lists a beneficial owner several layers up the chain. Both are correct. A platform that shows whichever record arrived last gives the investigator one name with no way to tell which kind it is.

Identity: At the end of Q3 2026, 248 tankers in Windward’s dark fleet were broadcasting the flag of a fraudulent registry, 71 of them VLCCs. Each is a vessel whose AIS declaration and registry record disagree. 

Flag changes now run in both directions. In Q3, 84 dark fleet tankers reflagged to Russia, more than four times the Q2 figure. Windward identified each one by its MMSI change rather than by a registry notice, and no falsely flagged tanker has loaded a Russian cargo since mid-June. A platform that waits for the registry record to update will show a flag Moscow has already replaced.

Position: The August Hormuz data shows the scale: more than six in ten logged transits were false.

Declared data: About one in three AIS-visible Bab el-Mandeb transits now fills the destination field with a statement about the ship instead of a port: “ALL CHINESE CREW,” “NO LINK TO ISRAEL.” The data arrives in the right format and passes every automated check. It just isn’t a destination. A platform that reads it as one logs a voyage to a port that does not exist.

What makes these cases dangerous is that the bad data often looks good. Dark fleet vessels frequently maintain AIS records that are consistent but inaccurate, allowing them to present as compliant.

Method matters as much as sources. At Windward, a data operations team applies maritime domain knowledge to conflicting inputs and assigns confidence scores before data reaches the platform. 

Behavioral checks add a second layer: before treating an AIS gap as a risk event, Windward runs a speed-and-distance feasibility test to confirm the vessel could physically have reached the area. And because records change, vessel risk models run daily across the global fleet and assessments refresh as ownership changes and new signals emerge.

What to request: A walkthrough of one real conflicting record: a vessel with competing ownership records, or a declared flag that does not match its registry. Ask how the platform resolved it and how the analyst sees the result.

A strong answer names a team, a method, and how confidence is shown. The analyst should see one record with a known confidence level, not two records and a judgment call.

4. Can New Data Types Join the Same Vessel Record?

AIS is cooperative by design: a transponder can be switched off in seconds, and position and identity data can be falsified with widely available tools. Even without deception, a study published in Nature found that 21 to 30% of transport and energy vessel activity is missing from public tracking systems.

In contested environments the gap becomes nearly total. On May 5, Windward identified 167 commercial-size vessels across the Strait of Hormuz area, 146 of them dark. A SAR collection on May 6 identified 97 vessels near the northern Hormuz corridor, only 3 transmitting AIS.

By Q3 the gap was the operating norm. UKMTO’s Hormuz transit count rose through the quarter while AIS- and satellite-visible transits fell from 20 a day in July to 13 in September. By September, 52% of transits were night convoys under U.S. escort, running with AIS off and outside satellite observation windows. No single source shows the strait. Windward’s Hormuz assessment reconciles UKMTO’s count, AIS, and satellite analytics into one picture; the convoy share is the measured difference between them.

This is why SAR, EO, and RF are already fused into Windward’s operating picture. The same principle applies to non-sensor data: ownership, casualty history, and adverse media are only useful when they attach to the same vessel record as its behavior. A vessel’s movements, flag history, and beneficial owner answer different parts of the same question.

New data types will keep arriving. A provider organized around one data type has to rebuild to add the next.

What to request: The integration history of the provider’s last new data type: what it was, when integration began, when it reached users, and whether it feeds the same vessel record.

A strong answer comes with dates and shows new data correcting or confirming existing records within a single vessel view.

Red flags: New data displayed as a separate tab or overlay. Ownership and behavior held in different modules. Roadmap slides in place of integration history.

How Windward Approaches Maritime Data

The figures in this guide come from the same approach it describes. The Hormuz transit correction, the Fujairah anchorage counts, the SAR collections, the Q3 Hormuz convoy assessment, and the false-flag and reflagging tracking were possible because no single data type is treated as ground truth, conflicting records are reconciled before they reach the analyst, and degradation is measured rather than assumed away.

Data quality is a design decision made before the gap appears. It is not visible in a demo, but it is visible in the answers to these questions.

Organizations evaluating maritime data providers can run both known-event tests with Windward’s team against their own area of responsibility.

Written by Ziv Gidron, Director of Content. Reviewed by Sabrina Edwards, Director, Data Partnerships, Windward.

See Windward in Action