Still Go Ahead? One Bad Vessel Record Enters the System: How Far Can AI Propagate the Wrong Decision?
🔔 Subscribe to ShipUniverse Weekly →
The Data Error Hidden Inside the AI Shipping Stack
Veson has brought Q88, Q88 Dry, VesselsValue, Shipfix, Milbros and IMOS onto one AI-enabled platform foundation. The opportunity is enormous. So is the importance of knowing exactly which source every decision inherited.
A charterer asks the AI for the best vessel. The platform sees a clean consumption curve, acceptable draft, current certificates, suitable ownership history and attractive voyage economics. The answer is convincing. The vessel is ranked first.
There is only one problem. One source field is wrong.
In fragmented software the error may die inside one spreadsheet or application. In an integrated commercial platform, the same data can be searched, compared, reconciled, imported into an estimate, queried by AI, exported through APIs and used in downstream decisions. Integration removes re-keying. It can also remove the accidental friction that once slowed a mistake down.
This is beginning to look less like a software suite and more like a commercial operating layer
Veson's October 6 expansion brings previously separate information, market-intelligence and workflow products into a common workspace around the IMOS system of record.
The new decision surface
Same workspace · different source systemsEstimates, fixtures, contracts, voyages, operations, finance and P&L context.
Tanker specifications, certificates, inspection history, officer data and vetting context.
Baltic99 data, certificates, loadlines and questionnaires for dry bulk vessels.
Valuations, ownership, transactions, fleet data, AIS-derived trade intelligence and efficiency information.
Orders, tonnage, fixtures, vessel positions, email-derived market information and voyage estimates.
More than 15,000 commodity records covering compatibility, regulatory information, safety and cleaning requirements.
We already have a real example of one-source-to-many reuse
Complete the vessel record once
Veson says answers from the Baltic99 automatically populate more than 300 related charterer and terminal questionnaires.
One incorrect consumption curve shows why lineage matters
Veson itself identifies wrong vessel data at the estimate stage as a problem that can cascade into bunker projections, port costs and demurrage. Take a simple hypothetical vessel whose consumption is understated.
A four-tonne-per-day error can materially change a thin voyage margin
30 tonnes/day for 25 sea days
34 tonnes/day for 25 sea days
The dangerous field depends on the decision being made
Speed / consumption
Can distort bunker requirements, voyage economics and comparative vessel ranking.
Draft / loadline
Can influence terminal suitability, questionnaire responses and cargo planning.
Certificate expiry
Can change a vetting result if a document appears current when it is not.
Ownership
Can change who a compliance team believes ultimately controls or operates the asset.
Loading capability
Can change terminal planning, time assumptions and cargo-handling expectations.
Today's architecture has more containment than the headline might suggest
The important distinction is between a common workspace, a reconciled data layer and automatic cross-product record mutation. They are not the same thing.
| Layer | How data is used | Can CoCaptain change it today? | Important guardrail | Error concern |
|---|---|---|---|---|
| Q88 / Q88 Dry | Vessel particulars, certificates, inspections, officer and questionnaire information. | Current Platform guidance describes CoCaptain as informational. | Account scope, permissions and centralized source modules. | Reuse risk |
| Vessel Insights | Reconciles Q88, Shipfix and VesselsValue fields into a validated vessel record. | Data can be pulled into IMOS. | Type checks, range checks, business rules and field-by-field adoption. | Reconciliation gate |
| VesselsValue | Valuations, ownership, transactions, efficiency, trade and vessel intelligence. | CoCaptain currently surfaces and summarizes data rather than changing records. | Separate data research, validation and analytical processes. | Decision support |
| Shipfix | Email-derived orders, tonnage, fixtures, AIS context and voyage estimates. | Current CoCaptain Market guidance is informational. | Vessel data is reconciled across recent market circulars and third-party sources. | Extraction risk |
| Milbros | Commodity, compatibility, cleaning, regulatory and cargo-safety information. | Current CoCaptain guidance says inform, not act. | Separate commodity records and organization-level access. | Context risk |
| IMOS | Commercial system of record for estimates, fixtures, voyages, contracts and operations. | Yes. Current documentation says CoCaptain can create, update and delete records. | Existing user permissions plus permission/save workflow around changes. | Action layer |
| Connect / APIs | Moves approved records or data into counterparties, BI environments and external systems. | Depends on workflow and integration. | Counterparty sharing is deliberate and API access is controlled. | External radius |
The best defense is not telling AI to “be careful”
Preserve source lineage
A user should be able to see whether a field came from Q88, Shipfix, VesselsValue, an owner record, a market circular or an internal IMOS override.
Reconcile independent sources
Agreement between independently collected records is stronger evidence than multiple screens displaying a value derived from the same upstream field.
Keep high-impact writes gated
Searching should be frictionless. Changing an operational record that drives money, compliance or safety deserves a different permission threshold.
Show disagreement, not just the winner
A validated value is useful, but a decision maker may also need to know when three credible sources disagreed before the platform selected one.
Attach freshness to critical fields
A vessel name, operator, certificate, draft variant or consumption curve can be perfectly correct and still be too old for the decision being made.
Make downstream correction traceable
Correcting the source is only half the job if the old value has already entered estimates, reports, APIs or externally shared records.
Maritime data is messy before AI ever sees it
The sea produces identity problems, stale records and contradictory signals
AIS identifiers can be reused or misconfigured, ships change names and MMSIs, market circulars disagree, ownership structures change, and vessel particulars evolve after modifications. AI sits downstream of all of it.
The risk changes when the platform moves from answering to acting
Veson describes its long-term direction as turning the system of record into a system of action. That is commercially powerful because the AI no longer has to stop at “this vessel looks best.” It can increasingly help users perform the next task inside the same operating environment.
That is also where provenance becomes operational control. A questionable data point used in a read-only search is an information problem. The same field used to update an estimate, trigger a workflow, populate an external system or support a fixture becomes a decision-control problem.
Bad Vessel Record Blast-Radius Simulator
Start with a consumption error that looks small enough to survive casual review. Then increase the number of estimates, reports or API consumers that use the record and change the number of independent validation gates. The model separates the commercial error from the propagation problem.
One vessel field is wrong. Does the fixture still make sense?
The commercial calculation is deterministic. The propagation calculation is an illustrative stress test showing how independent checks can reduce the number of downstream decisions exposed to the error.
The voyage remains profitable in this simplified case, but the corrected fuel burn removes more than half of the modeled margin.
The main control objective is to catch the consumption discrepancy before it becomes an adopted estimate input.