Your team stops typing and starts handling exceptions.
This is AI that assists your people — not AI that replaces them. A forwarder's edge is trust and judgment. We take away the keyboard, not the desk.
The one thing to understand first
One shipment produces a dozen documents — the shipper's invoice, packing list, house and master bills, the customs entry, the arrival notice, your sales and purchase invoices. They all describe the same goods. Yet a person retypes the shipper, consignee, weights, values and HS codes into each one, in every system.
So AI's real job in forwarding isn't a chatbot. It's read once → structure → reconcile → populate everything, forever. Every opportunity below hangs off that one loop — and it's the loop KargoZenn is built around.
estimated cost of manual data entry, per employee per year.1
of forwarding professionals say manual entry causes costly errors and delays.1
The map
Each card: the pain, what AI does about it, and where KargoZenn already stands.
Pain: a quote takes 30–40 minutes — read the request, dig out rate cards, surcharges, FX. Slow quotes lose the booking.
AI: drop the customer's RFQ email/PDF → in seconds it's parsed, priced from your rate cards, margin-checked against your floor, and rendered as a branded quote. You review and send.
Pain: re-keying the whole shipment from the shipper's documents into a new job.
AI: drop a folder of documents — extraction reads them and populates the job. Industry document-AI benchmarks report 97–99% field accuracy.2
Pain: extracted names don't match your customers or ports — garbage in, duplicates forever.
AI: reconciles every party, port and vessel to your masters, flags what's new, and asks before creating it.
Pain: a mismatch between the invoice, B/L and packing list — a different consignee, weight or HS code — gets cargo held at the border.
AI: cross-checks the whole document set and flags every disagreement before the job goes to clearance. Not blind data entry — an intelligent second pair of eyes.
Pain: wrong HS code or mismatched values = held cargo, demurrage, fines. Rules differ per country.
AI: suggest HS codes, check invoice vs B/L, list what each destination needs — our deep dive ↓
Pain: carrier portals don't talk to each other, ETAs go stale, delays get noticed too late.
AI: parse carrier and email events into one timeline; predict delays; surface only the exceptions.
Pain: billing backs up; vendor invoices don't reconcile to the job; over-billing slips through.
AI: auto-build the sales invoice from job charges, and match every vendor invoice to the job, flagging what's over cost or unexpected.
Pain: endless email to shipper, agent, customer and carrier — across time zones and languages.
AI: draft replies and status updates, chase missing documents, summarize a job's whole email thread.
Every stage gets harder across borders — and that's exactly where AI flattens the curve:
Two gates at the door
First we check the file is safe. Then we check its facts are right. One guards the envelope; the other guards what's written inside.
Every upload is scanned for malware before it's accepted. A byte-level bouncer — anything infected is blocked at the door.
We cross-check the whole set — consignee, weights, HS codes, ports — and flag any disagreement before customs. A virus scan for your documents' meaning.
The difference that matters: the antivirus blocks, Document Intelligence advises. A mismatch can be legitimate, so it flags for your team to decide — an intelligent second pair of eyes, never a blind gatekeeper.
Spotlight · built for a worldwide service
Customs is where a good AI stack pays for itself and where a naive one gets you fined. Here's the honest picture: what AI does brilliantly, and the line it must never cross.
A Chinese commercial invoice. A handwritten Arabic packing list. A Spanish certificate in a layout you've never seen. Old template-based OCR needs a template per vendor per language — and fails silently when a format changes.
Modern language models read meaning, not positions. Production logistics pipelines process 30+ languages with automatic detection,3 and independent benchmarks put LLM extraction at 97–99% versus 85–95% for template OCR — with the gap widening on low-quality and handwritten scans.4
The HS code is only universal to 6 digits. Above that, every country diverges — the EU's 10-digit TARIC, the US HTS, Japan's NACCS.5 The same product can carry wildly different duty depending on where it lands.
Destination-aware AI suggests the right national extension, estimates duty and VAT for that jurisdiction, and builds the required-document checklist for the country of entry. Research classifiers reach ~89% at the 10-digit level, with a top-3 suggestion hitting 93.9% — used as a support tool, not a decider.6
A single carton mismatch between the packing list and commercial invoice can trigger an automatic customs hold — commonly 5–14 days to resolve.7 Meanwhile demurrage runs ~$75–$300 per container per day.8
Document Intelligence — shipping today — cross-checks invoice ↔ B/L ↔ packing list on every job: consignee, shipper, weights, HS codes, ports and EORI. It flags each disagreement on the review screen while you can still fix it for free.
In ruling HQ H350722 (Jan 2026), US Customs held that AI deriving HS codes beyond six digits for goods being entered is "customs business" — requiring a licensed broker.9 A broker must keep responsible supervision and cannot rely solely on AI output.10
So we build it the only right way: AI assists. You decide.
Getting it wrong is statutory, not theoretical: US penalties for negligent misclassification reach up to twice the lost duties (four times for gross negligence); UK penalties run up to 30% of lost revenue for careless errors, 70% for deliberate.11 That's the downside we're designing out.
"Your team stops typing and
starts handling exceptions."
Not "AI replaces your team." The relationships, the problem-solving when a container is stuck, the judgment at the border — that stays human. We just take away the keyboard.
How we prioritize
Verifiable, human-approved, immediate ROI, low risk.
High value, advisory only, never autonomous.
Bring a real folder of shipment paperwork. We'll turn it into a job — parties, cargo, charges and all — while you watch.
Figures from solution vendors are directional; regulatory, WCO and peer-reviewed figures are authoritative. Where we say "AI classifies," read "AI suggests, a licensed human verifies."