BorderRev Valuation runs every declared value against external market data in under three seconds, applying WTO Methods 2, 3 and 6 in sequence and returning a confidence score so auditors can see how strong each finding is. It runs on every declaration, not a sample.
An importer declares a container of electronics at $50,000. Market price is $180,000. Your valuation officer has 30 seconds per declaration. No external price database. No AI. Just experience and intuition.
Manual spot-checking catches fewer than 2% of under-declared consignments. A single importer fraudulently under-declaring high-value electronics over three years can cost a customs authority $50 million or more in lost duties and VAT.
This is a problem of scale. No team of humans can run every declared value against current market data across hundreds of thousands of lines per day.
A seven-stage pipeline turns a raw goods description into a WTO-compliant valuation verdict in under three seconds. Semantic caching means repeat items return instantly without re-running market retrieval.
Local sentence-transformers (all-MiniLM-L6-v2) embed every goods description into a 384-dimensional vector. Near-duplicate items retrieve their cached valuation in milliseconds, bypassing market retrieval entirely.
Agentic retrieval pulls structured price data from exchange prices, manufacturer RRP, and verified marketplace sources. CSS selector extraction and Schema.org JSON-LD parsing produce reliable price observations without LLM hallucination risk.
Prices are normalised to FOB via incoterm conversion tables keyed on origin country, commodity chapter, and transport mode. WTO Methods 2, 3, and 6 are applied in sequence, with the result scored for confidence and returned with full source attribution.
The composite confidence score maps directly to downstream action queues. No manual sorting, because every valuation verdict routes itself to the right team.
Immediate post-clearance recovery action or goods hold. Strong market evidence, clear methodology, defensible in a tribunal.
Declared value is within tolerance of market price. Result is cached so the same consignment is never re-valued needlessly.
Risk-scored alongside other signals: classification disputes, origin flags and trader compliance history.
Specialist valuations officer assigned, with full case context and source data. Decision written back to the knowledge graph.
Every valuation verdict cites the WTO method applied, the source tier used, the incoterm conversion applied, and the observations underpinning the price. When a finding is challenged in a customs appeal, the legal basis is already documented.
The system applies WTO Methods in strict hierarchical order. Method 2, the transaction value of identical goods, is always attempted first. For an observation to qualify under Method 2, three conditions must all be met: the goods must be identical in all respects; they must originate from the same country of origin; and they must have been imported within the last 90 days. Country of origin is not inferred from the goods description alone. It is matched explicitly against the country declared on each reference transaction. If fewer than three qualifying observations exist, Method 3 (similar goods, same criteria) is applied. Method 6 (fall-back deductive or computed value) is the final resort, with full computation trace.
Vehicles are among the highest-value and most commonly under-valued goods at the border. A 2022 BMW X5 declared at $15,000. Actual market value: $62,000. Without the VIN, that is almost impossible to catch in 30 seconds.
BorderRev Valuation decodes the ISO 3779 VIN automatically, reading the World Manufacturer Identifier, Vehicle Descriptor Section and Vehicle Identifier Section, through the NHTSA decode API. Make, model, year, engine, trim, and country of manufacture are extracted and matched to market comparables. A depreciation model factors in age, mileage estimate, and salvage or rebuild status.
Once BorderRev Valuation has priced "iPhone 15 Pro Max 256GB Space Black", the next declaration of the same item, in any phrasing or capitalisation, is matched semantically and returned from cache in milliseconds. No market retrieval, no web scraping, no API call.
Semantic similarity uses local sentence-transformers embeddings (all-MiniLM-L6-v2, 384 dimensions, CPU-only). In steady state, 60–90% of incoming declarations match an existing cache entry, so the marginal cost of a valuation approaches zero as volume grows.
For cash-constrained customs authorities, which is most of them, the economics matter as much as the capability. BorderRev Valuation is designed to operate without paid LLM APIs in its standard path.
Rule-based keyword extraction and NLP heuristics handle the majority of product descriptions. Structured HTML extraction (CSS selectors, Schema.org JSON-LD) pulls price data reliably from market sources without LLM inference. Google Gemini Flash on free tier handles genuinely complex or ambiguous descriptions as a fallback, degrading gracefully if the free quota is exhausted. Local sentence-transformers run on CPU with no GPU required.
We offer a short, no-risk trial against a sample of your historical declarations. We return valuation verdicts with confidence scores and source attribution, and you keep the findings.