Using AI For Automation and Risk Detection in Trade Finance Compliance

A Digital Foundation in Trade Finance Compliance

A little over a year has passed since the UK Electronic Trade Documents Act came into effect in September 2023. The legislation was permissive and removed a constraint to the use of electronic trade documents under UK law, thereby supporting the global development of an electronic trade documents ecosystem. The intention was that this would make global trade and shipping easier with reports that it would add up to £1bn of additional value to the UK economy over a 10 year period. The legislation felt long overdue modernising long-standing statutes such as the Bills of Exchange Act 1882 and the Carriage of Goods by Sea Act 1992.

Institutions in the UK are now at least deriving efficiencies from using electronic trade documents for example a letter of credit or bill of lading. Digitisation has provided a foundation for modernisation, however the industry still faces numerous challenges involving manual work flows which undoubtedly is a contributing factor to the difficulty in detecting Trade Based Money Laundering.

The Challenge

The Financial Action Task Force has been clear that the International Trade System is susceptible to a wide range of criminal organisation in part arising “from the enormous volume of trade flows, which obscures individual transactions; the complexities associated with use of multiple foreign exchange transactions, the co-mingling of legitimate and illicit funds and limited resources agencies have to detect suspicious trade transactions”. The scale of the problem is difficult to estimate, but think tank Global Financial Integrity (GFI) suggests overall transnational crime could be worth US$2.2tn per year.

Digitisation of documentation has made it easier for trade finance compliance teams to run regulatory checks for example sanctions, peps and adverse media. Similarly, company incorporation, ultimate beneficial ownership, vessel registration and real time tracking has become routine enabled in part by cloud based API technology. Analysts however are faced with the task of running a series of additional manual checks across multiple stages of a trades life cycle from agreement to contract by the shipper, carrier, insurer and bank. These additional checks often mean processing hundreds of pages of documentation identifying clues that lead to the identification of fraudulent or money laundering based typologies.

Specific Trade Based Money Laundering Typologies Include:

• Over-Invoicing: Charging more than the actual value of goods.
• Under-Invoicing: Charging less than the actual value of goods.
• Multiple Invoicing: Issuing multiple invoices for the same vessel.
• Misrepresentation of Good: Falsifying the quality, quantity or description of the goods.
• Phantom Shipments: Transactions take place with no actual physical shipment of goods.
• Misusing Documentation: Criminals misuse or reuse documentation that they “believe” will go undetected by compliance teams.

The Role of Generative AI In Trade Compliance

Generative AI and Large Language Models could have a key role to play as a ‘fast but meticulous auditor’ detecting financial crime by cross referencing and validating millions of data points related to shipments within seconds.

For example, with good use of data and analytics supported by the latest AI techniques, it should be possible to identify if a particular good is being priced inline with its actual market value to avoid over/under invoicing or being correctly described/labelled and therefore adhering to local import/customs regulations and sanction programs.

Other data points for example commodity description, insurer costs, shipping numbers, export licences, chargers and free text fields can help reveal more complex typologies that often would avoid human manual detection.

A Worked Example

In this example we will show how an autonomous AI compliance agent could be developed to identify over invoicing. This is just one example typology to illustrate the potential and the process.

In this scenario, we will process a bill of lading document which describes a number of components being shipped by ‘Shin Auto Inc’ in Guangzhou, China to the purchaser ‘Acme Autos NL’ in Rotterdam, the Netherlands.

The AI agent could be developed to carry out 4 specific steps:

1. Analyse the goods descriptions within the document and capture english translations in a structured format. The dutch term "REMKLAUWEN AUTO" for example is translated into "AUTOMOTIVE CALIPERS”. This will allow us to more easily match when doing subsequent analysis.

2. The AI agent can determine that the shipment is coming from China, and compute unit costs for each item in the shipment based on the Chinese wholesale market value for November 2024. This could also be converted into a consistent currency.

3. The AI agent can identify which unit costs within the document are 37% higher than the Chinese wholesale market value for November 2024 Euros. In this case, the Calipers are automatically detected.

4. If any of the unit costs are 37% higher than the Chinese wholesale market value for November 2024, the agent could find comparative specific examples of those items for sale at the real market value in China in euros. The agent could collect evidence in the form of company name, address, price and website. This would help the Fincrime analyst be more efficient and effective when investigating the alert.

### Cable Grip Devices

1. **Hebei Xuchi Electric Power Technology Co., Ltd.**
   - **Price**: €0.48 - €0.90 per piece
   - **Location**: Hebei, China
   - **Website**: [allpower.en.made-in-china.com](https://allpower.en.made-in-china.com)

2. **Qingdao Xinquanxi Metal Products Co., Ltd.**
   - **Price**: €1.66 - €2.76 per piece
   - **Location**: Shandong, China
   - **Website**: [xinquanxi.en.made-in-china.com](https://xinquanxi.en.made-in-china.com)

### Automotive Calipers (Remklauwen)

1. **Guangzhou Baitai Automotive Components Co., Ltd.**
   - **Price**: €115.00 - €125.00 per piece
   - **Location**: Guangzhou, China
   - **Website**: [baitai.en.alibaba.com](https://baitai.en.alibaba.com)

2. **Zhejiang VOB Group Co., Ltd.**
   - **Price**: €105.00 - €120.00 per piece
   - **Location**: Zhejiang, China
   - **Website**: [vobgroup.en.alibaba.com](https://vobgroup.en.alibaba.com)

3. **Ningbo Top Automobile Parts Co., Ltd.**
   - **Price**: €110.00 - €128.00 per piece
   - **Location**: Ningbo, China
   - **Website**: [topauto.en.alibaba.com](https://topauto.en.alibaba.com)

4. **Hubei Feiling Auto Parts Co., Ltd.**
   - **Price**: €112.00 - €130.00 per piece
   - **Location**: Hubei, China
   - **Website**: [feiling.en.alibaba.com](https://feiling.en.alibaba.com)

5. The AI Compliance agent has also identified a Russian phone number within the document with a Moscow dial code and summarised this situation:

The phone number +7 495 7946 0958 appears in the document, and this is significant because:
The prefix +7 is Russia's country code
495 is the area code for Moscow

The Office of Foreign Asset Control (OFAC) in the US and the EU have imposed a number of sanctions and trade related restrictions when dealing with Russian connected entities. The presence of a Russian connection evidenced by a phone number in what otherwise appears superficially to be a “normal” document could lead to a breach of a global sanction program resulting in regulatory action and reputational damage.

Results

The AI Compliance Agent has analysed a bill of Lading, determining and extracting the key data points, resulting in the identification of a specific typology consistent with Trade Based Money Laundering affecting both ‘Automotive Calipers’ and ‘Cable Grips’ which represents €1.28m of the total transaction value.

The agent also detected other contextual risk points such as the Russian phone number, which may have been fine in isolation, but when combined with other indicators, may be worthy of investigation.

Importantly, the compliance AI Agent has identified this risk without being explicitly instructed to so by any compliance analyst when asked to review the document. These rules could be automatically applied thousands of documents per day, with no checks missed.

This analysis could have happened in the background without any human intervention whatsoever. However, the alerts and outputs could be made available for an analyst to review in multiple ways including a specialised trade compliance application.

Summary

Generative AI and large language models has the potential to play a major role in Trade Finance Compliance. Its value extends beyonds automation but also in risk detection acting as a supervised ‘compliance guardian’ throughout the trade life cycle alerting the analyst of new changes that impact a risk based approach. The effectiveness therefore of a compliance function will not only be in its people, policy, procedures and governance but also within its “ensemble” of AI agents.