Flexport doubled their customs brokerage profit using AI. But CBP just warned that AI-only classification isn't compliant. Here's what you actually need to know.
The headlines are everywhere: "AI is revolutionizing customs brokerage!" "Automate 70% of your customs entries!" "The future of trade compliance is here!"
Meanwhile, CBP issued this statement: "A licensed customs broker must exercise responsible supervision and control over their customs business, which can include classification. If a broker is relying solely on AI output for tariff classification... he/she likely is not exercising responsible supervision and control."
So which is it? Is AI transforming customs, or is it a compliance trap?
The answer, as usual, is: it depends on how you use it.
The Market Reality
Let's start with what's actually happening in the industry.
The U.S. customs brokerage market is estimated at $5.48 billion in 2026 and expected to reach $6.31 billion by 2031. That's modest growth at 2.88% CAGR. But within that market, something dramatic is happening:
Digital-first platforms are advancing at 10.45% CAGR—nearly four times the overall market growth rate.
Traditional brokerages still hold 77% of the market. But digital-first players are expected to triple their share by 2031 if current adoption trends continue.
Flexport—the poster child for tech-forward logistics—reportedly doubled their customs brokerage gross profit in 2025 using AI automation. They're expecting even higher growth in 2026.
The message is clear: AI isn't coming to customs brokerage. It's already here, and it's eating market share.
What AI Can Actually Do Today
Let's cut through the marketing fluff. Here's what AI tools are genuinely capable of in customs operations:
Document Processing and Extraction
This is where AI shines brightest. Modern OCR (Optical Character Recognition) and NLP (Natural Language Processing) can:
- Automatically extract data from bills of lading, commercial invoices, packing lists, and certificates of origin
- Populate customs entry forms without manual data entry
- Flag inconsistencies between documents
- Process documents in multiple languages
Real results: Some brokerages report 70% automation of customs entries and a 2x increase in processing capacity after implementing AI document processing.
Classification Suggestions
AI classification tools analyze:
- Product descriptions
- Historical classification data
- Images of products
- Technical specifications
They can suggest HTS codes with confidence scores and explain their reasoning.
What this means: Instead of starting from scratch with the HTS, a broker can review AI-suggested classifications and verify or override them.
Anomaly Detection
AI excels at pattern recognition. In customs, this means:
- Flagging unusual duty rates for a product category
- Identifying potential misclassifications based on historical data
- Detecting discrepancies between declared values and market prices
- Spotting compliance red flags before shipments arrive
What AI Cannot Do (Yet)
Here's where the hype meets reality:
Make Legally Defensible Classification Decisions
This is the big one. When CBP audits your classifications and asks "how did you arrive at this determination?"—"the AI told me so" is not an acceptable answer.
HTS classification requires applying the General Rules of Interpretation in sequence. It requires understanding section and chapter notes. It requires judgment calls on essential character for composite goods. It requires knowledge of CBP's ruling history and interpretation philosophy.
AI can suggest. Humans must decide and document.
Exercise Professional Judgment
Licensed customs brokers have legal responsibilities that cannot be delegated to software. These include:
- Responsible supervision and control
- Exercise of reasonable care
- Maintaining records and documentation
- Advising clients on compliance matters
CBP has made clear: brokers who abdicate these responsibilities to AI are not exercising responsible supervision.
Handle Novel Situations
AI is trained on historical data. When something genuinely new appears—a novel product, a new regulatory framework, an unprecedented trade action—AI struggles.
The tariff landscape of 2026 is littered with novel situations: de minimis suspension, expanding Section 301 tariffs, UFLPA enforcement, reciprocal tariffs. Many of these have limited historical precedent for AI to learn from.
The CBP Warning: Why It Matters
Let's revisit that CBP statement more carefully:
"A licensed customs broker must exercise responsible supervision and control over their customs business, which can include classification. If a broker is relying solely on AI output for tariff classification... he/she likely is not exercising responsible supervision and control by failing to verify proper classification."
Key phrases:
- "Solely on AI output" — CBP isn't saying you can't use AI. They're saying you can't ONLY use AI.
- "Responsible supervision and control" — The broker must remain in charge. AI is a tool, not a replacement.
- "Failing to verify" — The verification step is mandatory, not optional.
This isn't anti-technology bias. It's recognition that legal responsibility cannot be outsourced to software.
The Right Model: AI + Human Oversight
The brokerages winning with AI aren't replacing humans. They're augmenting them.
Here's what the winning model looks like:
Tier 1: Fully Automated (with guardrails)
- Routine shipments with established classifications
- High-confidence AI suggestions (95%+)
- Standard products with clear precedent
- Human role: Periodic audits, exception handling
Tier 2: AI-Assisted
- New products requiring classification
- Medium-confidence AI suggestions
- Products with multiple possible headings
- Human role: Review AI suggestions, apply GRI analysis, make final determination, document reasoning
Tier 3: Human-Led
- Complex or novel products
- Multi-component goods requiring essential character analysis
- Products in regulatory gray zones
- Binding ruling requests
- Human role: Full classification analysis with AI as research tool
Questions to Ask AI Customs Vendors
If you're evaluating AI tools, here's your diligence checklist:
- What's your classification accuracy rate? (And how do you measure it?)
- What training data do you use? (How recent? How comprehensive?)
- How do you handle the General Rules of Interpretation?
- Can you provide reasoning for suggestions, not just codes?
- How do you incorporate new rulings and regulatory changes?
- What's your approach to novel products with limited precedent?
- What documentation do you provide for audit defense?
- What are the limits of your system? (Be wary of vendors who claim 100% accuracy)
What This Means for Importers
If you're selecting a customs broker, look for:
Green flags:
- Transparent about their AI usage and its limits
- Clear human review processes for classifications
- Strong documentation practices
- Willingness to discuss their compliance approach
Red flags:
- Claims of "100% automated customs clearance"
- Inability to explain how classifications are verified
- No clear answer on how they handle complex products
- Significantly lower prices with no explanation (might be cutting corners)
The Bottom Line
AI is transforming customs brokerage. That's not hype—it's happening.
But transformation doesn't mean replacement. The technology that processes documents at superhuman speed still needs human judgment to determine if a product is "furniture" or "equipment." The algorithm that suggests HTS codes still needs a licensed professional to verify the classification is correct.
The brokerages that thrive in 2026 and beyond will be those that figure out the right balance: enough automation to be competitive, enough human oversight to be compliant.
TariffLens brings AI-powered classification suggestions to your workflow—with full transparency on reasoning so you can verify, document, and defend every classification decision.