The defense and economic-security application
Reviewed 8 October 2026. This article retains the existing commercial-intelligence research and moves it to The Defense Lens for its treatment of maritime situational awareness, supply-chain continuity and evidence quality. The source was a procurement analysis, not operational intelligence about named vessels.
IMO’s AIS rules cover specified vessel classes, including international-voyage ships of 300 gross tonnage and above, certain domestic cargo ships of 500 gross tonnage and above, and passenger ships regardless of size. The scope is not all maritime activity. IMO AIS guidance.
Consequently, a missing AIS observation does not establish hostile intent, and a vessel position does not establish cargo ownership or contact with seabed infrastructure. A public-policy decision should preserve the sequence from observation, to inference, to corroboration, to assessed confidence. Escalation authority and responsibility for correcting a false positive need to be explicit.
The same discipline improves a bank’s trade-finance review, an importer’s supply assessment and a government’s resilience picture. These users may share evidence but need different thresholds, permissions and response processes. No private bank budget is carried over as a government procurement estimate.
Commodity and maritime data should be purchased against the decisions it improves. A vessel map can be visually impressive while contributing little to a credit decision, a commodity balance or a sanctions investigation. The strongest purchasing framework separates price formation, physical-flow observation, inventory estimation, strategic research and control workflows before comparing suppliers.
This report combines overlapping historical briefs into one operating framework. Supplier and institutional names have been removed. Quantitative statements retain their original status as historical claims, estimates or teaching illustrations; they are not verified current contract terms. The core argument is that complementary data layers can justify multiple suppliers, but duplication has to be demonstrated at the workflow level.
Five data layers answer different questions
Price assessments establish a reference around which contracts, valuations and commercial negotiations are organised. Their value often comes from embedded contractual use and accepted methodology. A competing dashboard with more features does not necessarily substitute for a benchmark named in a physical contract.
Physical-flow intelligence attempts to infer what is moving, where it came from, where it is going and in what quantity. Inputs may include vessel-position messages, port calls, draught changes, fixtures and analyst interpretation. The economically relevant output is an estimate of cargo and destination, not the raw location of a ship.
Inventory intelligence estimates stocks in tanks, terminals or vessels. It can combine visible infrastructure, satellite observation and statistical inference. Research and asset intelligence adds longer-term supply, demand, cost-curve and project analysis. A fifth layer supports controls: vessel ownership, sanctions exposure, deceptive shipping behaviour and evidence for trade-finance monitoring.
The distinct jobs of commodity-information layers
| Layer | Reader’s operational question | Main validation issue |
|---|---|---|
| Price assessment | What reference price governs this exposure? | Methodology, licence and contract acceptance |
| Cargo and flow data | What physical supply is arriving or leaving? | Coverage, inference error and revision history |
| Inventory estimation | How are available stocks changing? | Ground truth, observation gaps and uncertainty |
| Strategic research | What changes the medium-term balance? | Assumptions and scenario consistency |
| Control intelligence | Is the cargo, vessel or counterparty problematic? | Identity, evidence provenance and escalation rules |
The layers overlap but are not interchangeable. A broad provider may support many desks; a specialist may produce a stronger signal in one product or location. The historical comparison consequently favoured a broad core platform alongside a narrower energy-flow specialist where the latter offered measurable incremental coverage or a genuinely independent validation path.
Figure 01 · Conceptual framework
From observed signal to investment judgement
Accuracy needs an object, a horizon and a denominator
A claim of “99% accuracy” is incomplete without a defined event and population. Does it refer to observed ship positions, inferred cargo grades, destination forecasts, arrival dates or total volumes? Is the test conducted on all voyages or only voyages eventually resolved with high confidence? Is the prediction evaluated before arrival or after the dataset has been revised?
For a bank financing a cargo, the useful test may be whether the vessel reaches the expected destination inside a contractual window. For a trading analyst, the useful measure may be error in a weekly import estimate available before an official release. A historical database can appear highly accurate if its final corrected state is compared with final ground truth. That says much less about the information available when a decision had to be made.
The proof of concept should therefore preserve the first observation, subsequent revisions and final resolution. Report accuracy by commodity, region, vessel class and confidence band. Include dark or missing observations rather than quietly dropping them. Where ground truth itself is uncertain, show the uncertainty rather than forcing an apparently exact comparison.
The old source included an unnamed storage-accuracy claim. It is excluded from the refreshed numerical comparison because the metric, complete evaluation period and comparison baseline were not independently verified. The correct publication test is an out-of-sample error distribution with the observation timestamp preserved.
Data freshness and decision freshness differ
A vessel-position feed can update in seconds while an inferred cargo remains uncertain for days. Conversely, a carefully maintained strategic dataset may update every few hours and still be entirely adequate for its use case. Comparing refresh rates without comparing the object being refreshed leads to poor purchasing decisions.
The old feed-latency and analytical-refresh claims concerned different products and different objects. This edition therefore does not display them as a competitive speed comparison. Ask separately when the event occurred, when the provider observed it, when the inference changed and when the buyer received it.
Contracts should distinguish delivery latency, dataset freshness and analytical revision. A workable requirement might specify that a defined proportion of updates arrive within an agreed interval, while separately recording the age and uncertainty of the underlying observation. An application should expose stale data rather than presenting the last known value as a fresh observation.
This is especially important when dashboards and application programming interfaces are not identical products. A feature visible to a human user may arrive later through an API, have different redistribution rights or be unavailable in historical extracts. Buyers should test the exact delivery channel they intend to put into production.
Integration determines whether intelligence becomes usable
Commodity intelligence becomes more valuable when it joins consistently to internal positions, counterparties, facilities and agreements. This requires stable identifiers and a maintained data model. Vessel identifiers, ports, installations, grades, ownership relationships and company hierarchies need explicit mappings.
Name matching alone is fragile. Ships change names and ownership structures; installations are renamed; cargo grades can be described inconsistently. The bank should retain the provider’s identifiers, its own canonical identifiers and the mapping history. When a relationship changes, the system needs to preserve what was known at the earlier decision date.
An ingestion design should also handle corrections and deletions. If a provider revises a voyage from one cargo type to another, downstream positions and alerts may need to be recalculated. An append-only feed without correction semantics can leave a bank with internally inconsistent history.
Operational requirements include rate limits, pagination, backfills, entitlements, schema changes, versioning and incident notification. Data engineers should participate in the proof of concept. A desk-level pilot that works through manual downloads may conceal the cost of enterprise production.
Figure 02 · Qualitative comparison
Three different tests of useful intelligence
| Test | What to inspect |
|---|---|
| Accuracy | Errors against a defined reference |
| Freshness | Timestamp chain and revisions |
| Usability | Identifiers, interfaces and licensing |
| Value | The decision made with the evidence |
Diversification must be real
Maintaining two suppliers can improve resilience and challenge model error. It can also duplicate the same underlying observation with different branding. The relevant question is how independent the information and delivery paths actually are.
A broad provider and a specialist can be complementary if they use meaningfully different analytical methods, cover different commodities or resolve different failure modes. Two interfaces built on the same position-feed supplier may offer less protection against a common data outage. Acquisition-driven concentration can increase this risk even when customer-facing products remain separate.
The historical reports therefore supported selective dual sourcing rather than universal duplication. Retain a second provider where the incremental signal changes a decision, where it supplies a credible fallback or where it provides independent challenge for a control-sensitive use case. Remove seats and overlapping modules that have no identifiable owner or recurring use.
Ownership changes should trigger a review of data lineage, renewal terms and roadmap dependence. They do not automatically imply deterioration. The important point is to make concentration visible before it becomes an unexpected operational or pricing constraint.
Commercial comparisons need a common perimeter
Historical inferred customer-spend figures have been removed from the public comparison. Average supplier revenue per customer does not establish the price paid by a large bank or a public agency. The refreshed method requires current quotes against the same entitlement, delivery, retention and support perimeter.
The commercial perimeter of a commodity-data purchase
| Cost component | What to specify before comparison |
|---|---|
| User access | Named users, concurrent use, desk scope and geography |
| Data rights | Internal redistribution, derived data, retention and audit access |
| Machine delivery | API quotas, bulk history, streaming and cloud egress |
| Integration | Identifier mapping, backfills, schema maintenance and support |
| Commercial change | Renewal caps, acquired datasets and module bundling |
| Exit | Continued use of historical data and transition assistance |
The correct cost denominator is the completed decision workflow. A lower subscription can be offset by manual reconciliation, weak historical access or restrictive derived-data rights. A higher-priced specialist can be worthwhile if its incremental signal is both used and validated. Neither proposition can be established from a feature checklist alone.
Figure 03 · Conceptual framework
Compare vendors on a common perimeter
A proof of concept that can change the decision
Select a limited number of real workflows before opening the trial. Examples include weekly cargo-arrival estimates, financing-condition monitoring, inventory nowcasting and sanctions escalation. For each workflow, define the existing process, the expected improvement and the evidence needed to accept it.
Use a historical period that contains ordinary conditions and disruptions. Include difficult regions, missing position signals, revised destinations and ambiguous cargo classifications. Ask suppliers to state confidence and exclusions in advance. A trial that consists only of familiar high-coverage routes creates an overly favourable impression.
Score coverage, accuracy, timeliness, integration, control evidence and commercial rights separately. Do not let an attractive user interface compensate for a failure in a mandatory control requirement. Equally, do not impose sub-second delivery on a workflow whose decision horizon is measured in days.
Require a named business owner to demonstrate repeated use. Useful measures include adoption in recurring decisions, time spent reconciling conflicting estimates, unresolved data exceptions and the proportion of outputs that can be traced to evidence. Vendor selection should end with an operating plan, not just an approved purchase order.
The historical AI-workflow forecast
The expanded source forecast that banks would increasingly expect natural-language research workflows within two years and that suppliers unable to deliver them would lose ground at renewal. It also argued that an AI-native challenger could win a narrow desk-level trial if its workflow was substantially better and its underlying data accuracy was credible. Building methodology trust at a major-bank standard was estimated to require three to five years.
These are original historical forecasts, not claims about subsequent market share. They sharpen the procurement argument: conversational access can improve usability, but it does not replace a validated commodity signal. An assistant that confidently answers from an uncertain cargo classification can make the uncertainty less visible. Evaluate whether the workflow exposes evidence, revisions and confidence, and whether its output saves real analytical work. A better interface is valuable when it makes trustworthy information easier to use.
What a comprehensive stack should retain
The merged research supports three durable conclusions. First, benchmark price infrastructure, physical intelligence and strategic research have different sources of defensibility. Second, a specialist can justify its place when its distinctive information survives a disciplined test. Third, the greatest avoidable waste often sits in poorly governed licences, redundant modules and data that never reaches the workflow that was used to justify it.
The purchasing objective is consequently a small, coherent set of information services with clear ownership and testable obligations. The bank should know what each supplier contributes, which assumptions remain uncertain and how an important decision proceeds when the data is late, revised or wrong. That is the point at which commodity intelligence becomes decision infrastructure.
