Bad data example

There continues to be a deluge of articles, videos and conferences extoling the benefits of Artificial Intelligence. But there is less emphasis on what businesses must do internally to implement the technologies. One of the major requirements is to use clean data, otherwise it is ‘garbage in, garbage out (GIGO) for IT systems.

The lack of clean data has been highlighted in an amusing example with the imposition of increased import tariffs by the US government. The list of countries liable for increased tariffs includes Norfolk Island,1,600km north-east of Sydney in the Pacific and the remote Antarctic territories of Heard Island and the McDonald Islands. These locations are actually Australian territories, with Norfolk Island relying on tourism and the Antarctic islands populated by penguins, not humans. Neither has exported goods to America. So why was an import tariff applied?

While the world’s media reported on the funny parts of the story, only the Guardian asked the question “why did this happen”? This blogpost summarises the Guardian article.

Mislabelled shipping documents

A feature of international trade that supply chain professionals commonly experience is shipping documents that contain errors, which occurred in this case. Multiple shipments of goods were classified as having originated from Norfolk Island or Heard and McDonald Islands, even though neither the exporting company, the port of departure, or the destination port were located in these territories.

According to US government data, Norfolk Island exported US$655,000 worth of goods to the US in 2023, with its main export being US$413,000 of leather footwear. The Guardian identified two bills of lading for shipments, each containing 3,714 Timberland men’s black ankle boots. These were shipped in December 2023 from the Bahamas to the port of Miami, Florida. However, the bills of lading showed Timberland’s registered address as Norfolk Island (NI) rather than the correct address in the US State of New Hampshire (NH).

Other bills of lading also incorrectly listed Norfolk Island as the country of origin. These included several from an aquarium company, which sent shipments from their base in the county of Norfolk, UK to the US, and steel equipment sent from Novum Structures in the county of Norfolk, UK, to the US. Novum Structure’s address was identified on one bill of lading as Norfolk Island, instead of Norfolk, UK.

Heard Island and McDonald Islands were reported as having exported US$1.4m of products in 2022, nearly all of which was ‘machinery and electrical’ goods. The Guardian identified multiple bills of lading which showed that exports which supposedly originated from the remote islands were mainly steel or plastic goods sent from European countries to the US.

Other territories were noted as having a similar situation:

• Tokelau, a territory of New Zealand in the southern Pacific Ocean, had a shipment of pergola parts that were shipped from Turkey (Türkiye) to the US. They were mislabelled, with the supplier listed as based on Tokelau.
• Bills of lading for shipments from Indian suppliers to the US were mislabelled as coming from the ‘British Indian Ocean Territory’. This is a remote external territory and military base of the UK with few inhabitants.

Errors flow to analysis

The errors in the shipping records flowed through to the US census and trade data system, that was used for calculating the increased US import tariffs. To protect itself, the US Census Bureau, responsible for collating US trade statistics, acknowledges in its trade statistics guide that reporting errors can occur and can “significantly impact detailed commodity statistics”.

It is not clear where in the shipping and receival process the errors occurred; but they were made by humans, when transcribing from one document to another or between computer screens of different software applications. And these things happen continually through the world’s supply chains.

‘Normalisation’ is the IT term for the process of eliminating discrepancies in downloaded data, so that it is consistent for analysis. As shown in this blogpost example, discrepancies with ‘outside-in’ data can occur anywhere in a supply chain process, some examples being:

• Goods destined for Australia are consigned to Austria. Likewise with other locations that have some similarity in their name
• Companies trade under multiple business names, although they are ultimately owned or controlled by one entity, but the names are not consolidated. For example, Ford Motor Company and Ford Australia;
• Dates on documents can be written in the format dd/mm/yyyy, or mm/dd/yyyy, or yyyy/mm/dd and not recognised as the same date;
• Addresses can be written differently e.g New York, or NY, or NY NY;
• Standard industry codes (SIC) to identify an item may be used by some organisations, but not others, so items are incorrectly allocated and
• Descriptions of a product or service can vary, depending on the person doing the data entry

Using ‘outside-in’ data

The important lesson for supply chain professionals is that when using external data (also called ‘outside-in’ data), maximum care is required. Since the early days of MRP systems, there has been an emphasis by software suppliers, member organisations and implementation trainers on the importance of data accuracy within a business system.

Fifty years later, with inputs destined for use by AI, classes continue to be presented, so the lessons are not being absorbed. Too often, data accuracy is considered as a ‘nice to have’, but if not measured is relegated to the hard basket.

Supply Chains consist of input, internal and output steps. Only the internal part has its data collection controlled so that users know that technologies and people will accurately collect and analyse current performance to produce scenario models.

To reduce the risk of errors and discrepancies in external data, implement an ongoing measurement regime and data accuracy target, together with defining how and when it will be achieved. Wherever possible, standardise supplier documents, labelling and packaging specifications and implement barcode scanning or RFID tagging for inbound items.

As the quality and accuracy of ‘outside-in’ data is less known, so confidence is reduced. Therefore, the data must be used in a more macro manner, assisting with the identification of trends that are used in plans developed by knowledgeable Operations Planners.