NerveSight reads every customs document in a shipment — invoices, bills of lading, certificates of origin — and catches errors before they become $4,000 demurrage charges.
Customs brokers manually key data from multiple documents into trade management systems, HS-coding from memory against tariff schedules that change every quarter.
Each classification error triggers demurrage charges, CBP penalties, or re-export costs — none of which appear until the container is sitting in a bonded warehouse at $200/day.
A single customs entry requires reconciling 5–9 documents from different parties, in different formats, in different languages — all before the freight deadline.
Quantities, weights, values, and HS codes reconciled across every document in the packet. Mismatches flagged with source citations before the entry is filed.
Tariff schedules change quarterly. NerveSight keeps pace — every classification checked against the current national schedule, not last year's PDF.
Knows that "Made in Vietnam" on a Chinese-origin component triggers different rules. Applies the correct origin determination standard per importing country.
Every field, every decision, every source. If CBP ever comes knocking, your compliance trail is already built.
The US customs industry is deeply bifurcated. Enterprise freight forwarders (Expeditors, Kuehne+Nagel) have compliance infrastructure. The 15,000+ independent CHBs with 2–12 employees do not. They run on QuickBooks, industry memory, and one person who knows the HS schedule "pretty well."
That cohort processes millions of entries per year. Every tariff change is a potential penalty event. Every new FTA triggers a new set of origin rules. And CBP's automated ACE system is getting better at pattern-matching misclassifications faster than manual processes can keep up.
USMCA, CBAM, and Section 301 changes have made "memorize the tariff schedule" an impossible strategy — the schedule is now a living document.
Mid-market manufacturers with in-house import teams (Phase 2 target) are still running post-import compliance audits in spreadsheets, discovering errors 6 months after entry.
The beachhead — independent US CHBs — is highly concentrated: ~4,000 firms handle 80% of mid-market import volume, all within 8 major port cities.
Every discrepancy caught, every penalty avoided — anonymized — becomes a training signal. After 18 months, NerveSight's error-detection model is trained on more real customs disputes than any government database publishes publicly.
A generic LLM hallucinating HS codes with high confidence is worse than Excel. NerveSight's output is source-grounded: every field traces to a document line and a regulation clause. That's not a prompt — that's architecture.
As brokers adopt NerveSight, their importer clients gain visibility into pre-audits. That importer then requires their next broker to be on NerveSight. Classic two-sided lock-in, built through value — not switching costs.
Before GPT-4V and Opus-class models, OCR + extraction on heterogeneous trade documents had a 20%+ field error rate — worse than humans on complex packets. That's now under 3% with grounded extraction. The quality bar for commercial deployment just cleared.
Section 301 tariffs, USMCA origin rule changes, EU Carbon Border Adjustment Mechanism, India BIS requirements — the tariff schedule is no longer a stable reference. It changes quarterly. Manual classification is now structurally impossible to keep current. This shift happened in 2024, not gradually over a decade.
CBP is running automated pattern-matching audits via its ACE system. Small CHBs are receiving penalty notices they've never seen before. Fear is a real demand driver — and it's new. The first 50 customers don't need to be convinced the problem is real. They already have the penalty notice.
Saksham built NerveSight from a problem he couldn't stop thinking about. Watching import operations run on tribal knowledge and manual entry — seeing penalties arrive months after containers cleared — convinced him that this wasn't a process problem. It was an information problem that AI could finally solve.
He combines deep operational instincts with a technical conviction that grounded, citation-traceable AI is the only way to build something that actually belongs in a compliance workflow — where a confident wrong answer is worse than no answer at all.
"The container doesn't wait for your manual process to catch up. Neither does CBP."