Brexit & trade wars are causing a customs crisis. What’s the solution?
Answer: Broker expertise + technology
The world of logistics is in turmoil. UK has only one year left till it officially leaves the European Union. In the US, President Trump comes up with a weekly wish list of items to include in tariffs and then China and US go round and round deciding which ones to choose for actually implementing.
According to a report by Institute for Government, a UK-based think tank, around 180,000 British businesses that operate only within EU will be faced with making customs declarations for the first time in the post-Brexit world. Around 200 million declarations will need to be made costing £20–£45 each.
That’s a colossal 4–9 billion pounds in expenses.
Not counting the increase in costs, where are these businesses going to find so many experienced customs brokers?
As our previous piece on the complication of categorizing pasta mentioned — customs classifications are a convoluted system, influenced by history, economics and geopolitical considerations. Changing tariffs will influence products all through the supply chain. And most companies do not know their supply chain. A survey by Business Continuity Institute shows that over 69% of companies do not have full visibility over their supply chains.
The increased costs of reporting along with the lack of visibility in a globalized supply chain means many companies are unprepared to deal with the rapid shocks that they’ve recently been subjected to.
Laying out the plot — how HS codes work.
So why hasn’t technology fixed this already? To answer that we have to go back a bit to the origin and taxonomy of HS codes.
The HS codes or Harmonized System of tariffs is exactly what it sounds like — a way to align different countries’ systems of product classification so that global trade can hum along. It consists of six digits. The first two denote the HS chapter, the next two the heading and the last two the subheading. Chapter indicates broad categories (eg. Cereal) which heading and subheading expand on the product characteristics in more details (eg. Rice milled but not polished).
That’s all well and good. But what about Arborio rice milled in Italy which is very different from the Basmati variety milled in India and Pakistan? Ah, that’s where country-specific variations of HS codes come in.
For the United States the country-specific portion is referred to as HTS codes. The first 6 digits of HS codes are international but the next 1–4 digits are country specific and vary between countries. This means that a Chinese supplier might give products properly classified but with the Chinese tariff codes which then need to be converted into the US tariff codes while importing.
The inefficiencies in the system
As you might have guessed by now, a customs brokers job is not easy. It requires years of experience to navigate a ever-changing complicated world of customs regulations.
Often the solutions implemented to solve these problems oversimplify the complexity of the problem. Take for example the US Custom’s Automated Commercial Environment (ACE) that was supposed to streamline data flow from the Customs & Border Protection agency. Over 3 years and $1 billion dollars later the system is still not fully functional. Quoting the National Customs Brokers and Forwarders Association of America (NCBFAA), the ACE has —
“several critical issues with elements such as remote location filing, currency conversion for duties and value declaration, and insufficient automation of the invoice interface.
Communication is also an issue with the messages that ACE sends out. The NCBFAA is calling for a complete list of ACE messages and their meaning. The authors described the messaging system as “duplicative, inconsistent and prone to incorrect interpretation by CBP and stakeholders.”
Or take Canada for example. In a recently published tariff classification audit the Canada Border Services Agency (CBSA) revealed that 66% of the cases were non-compliant. This is a shockingly high number of non-compliance considering the dollar value of world merchandise exports was US$17.20 trillion in 2017 and it only going to grow further. It is such a massive amount that even the most conservative estimate will place the losses at many billions of dollars in value.
Canada’s solution? To pre-publish a list of verification targets! Yes, they’re that nice. For 2018 CBSA announced the list of categories that will be prioritized for trade certification. I guess their hope is that importers of hair extensions and bicycle parts will fix themselves if they know of the extra scrutiny this year.
So what is the solution?
There’s no single solution
It shouldn’t come as a surprise to anyone that a system which is designed to allow trade and communication between multiple countries, their agencies and its industries is not simple. And to fix such a system doesn’t require a solution, it requires multiple smaller solutions.
One such solution is Semantics3’s Categorization API.
Improving a legacy system
In our surveys with various logistics companies, we noticed something odd. The same product was being assigned up to 10 different HS codes and even clearing customs with such variations. This points to the (unsurprising) lack of uniformity in a purely-manual system. Compared against a machine based classifier which would have a repeatable output given specific input a manual-only system is slow and costly. However, a machine that consistently gives the wrong code is useless too.
There are also finer distinctions between products which AI systems might not be able to grasp. For example, apparel when knitted falls within Chapter 61, whereas apparel when not knitted falls within Chapter 62. With purely machine-powered systems, such finer distinctions might go unnoticed resulting in expensive errors.
Considering the accuracy/cost implications associated with either approach, we need to look at walking a fine line, a balancing act between the two paradigms.
A combined system
A suggestion system which provides both chapters, while highlighting the distinctions, could enable a human reviewer to pick the right code.
The time spent classifying a given product, in a purely-manual system could vary anywhere from 1 minute (for an expert given familiar HS chapters) to over 4 minutes (for beginners in the more arcane codes). Compared to the seconds it takes an AI-system, the cost/time savings of an AI system clearly stand out.
Intelligent systems can also identify the differences between high-cost and low-cost errors. By layering an AI system with human input, specific products can be (re-)sent through multiple classification methods to ensure a higher level of fidelity in the output. Our client, a 3PL provider called Aeropost, found that their classification time per item fell from an average of 2 minutes to under 25 seconds when their employees could scan products to get categorization suggestions.
Many simple solutions
HS codes classification, logistics optimization, tariff wars — none of these are simple topics and they do not have simple solutions. But the key to solving a complex problem, very often, is doing it in small parts. Our AI-categorization automation is just one such solution of many.
By combining AI-powered systems with human curation, a more robust classification engine can be built. Together, the combined system can tackle (in small parts) the customs crisis in an increasingly fragile world filled with global trade wars and uncertainty. After all, a small solution that works well can have a huge compounded impact.
Checkout Semantics3’s Categorization API here.
Written by Anjali Krishnan and Ramanan Balakrishnan.
