Processing the myriad of information generated at every step in the supply chain has always been a headache for shippers and customers. Now, though, increasing consumer expectations for fast, low-cost delivery have combined with shipment backlogs caused by COVID-19 to cause an explosion in the volume and variety of information that organizations must process.
Artificial intelligence (AI) can reduce costs and improve customer service by extracting more information more accurately and at lower cost, to assure timely delivery and meet regulatory requirements.
Accurately and quickly processing information in documents, including bills of lading, commercial invoices, and arrival notices, has become a strategic priority not just for the back office, but also for senior management in virtually every company. This information ranges from customer records to compliance forms—such as for rules around handling potentially dangerous goods—and information about the country of origin and the contents of each shipment.
Stakeholders use this information in many different ways, including for customs reporting, informing customers about deliveries, and analyzing information about the movement of goods to monitor market trends and streamline operations.
Collecting and processing the information is becoming increasingly difficult due to the growing complexity of supply chains, the increasing amount of required documentation, and the profusion of forms and the data formats used in them. Adding to the complexity are the facts that documentation changes hands multiple times as it moves through the supply chain and that, even within a single company, different functional units such as finance, accounting, customs, or operations require different information.
Failure to provide accurate information, in the proper format and at the right time, to customers, governments, and business partners can cause immediate and severe harm to the business. Negative consequences might include:
Traditional document processing in supply chains leads to unhappy customers, excessive costs, and sometimes the loss of essential business licenses. Even the use of “intelligent” optical character recognition (OCR) to extract information from physical documents requires far too many expensive, error-prone manual steps to meet today’s needs.
Many firms currently use one of two versions of OCR, each of which suffers from significant drawbacks.
The first technique, generic OCR, extracts all the text from digitized documents into a raw list of words from each page or image. This raw text output is not particularly useful, since it still requires manual data entry to match the raw text to required fields such as, for example, an invoice number or a description of shipment contents.
Managing and scaling this manual data entry to meet today’s business requirements can be extremely costly—and because it is error-prone, it still doesn’t deliver the required accuracy.
The variability in document layouts and formats limits the accuracy of data extraction with generic OCR to about 50% to 70% for simple documents and only 10% to 50% for complex unstructured documents. That means a lot of missing and inaccurate data.
The second technique, intelligent OCR, can extract the required data fields from a document. But it requires the manual creation and maintenance of templates and rules to define the location of the fields for every document layout. This painstaking work can become incredibly difficult to scale as organizations expand their use cases and must extract data from forms associated with hundreds of business partners. Even the use of custom templates can fail to account for edge cases or exceptions, reducing the accuracy needed to assure timely shipment.
When faced with nonstandard formats or alternate document layouts, even “intelligent” OCR will miss or mislabel data. The use of nonstandard, alternate formats is more common now, as shippers work with new customers, as customers require new or different types of information, or as regulators require new information. In addition, the need to manually monitor data quality and hard-code changes in templates leads to higher costs and delays.
Advances in AI are helping companies use shipping information for competitive advantage, reducing costs and improving service in ways impossible with either generic or intelligent OCR. These advances include:
All of these capabilities are provided by today’s technology, and more, including:
One U.S.-based freight forwarder and customs broker is using AI to cut the time required to process documents—bills of lading, commercial invoices, air waybills, and arrival notices—from more than two days to only seconds. Over 95% of the documents it processes don't require any human intervention.
AI allowed the company to meet its customers’ needs for rapid delivery without the need for templates or having to hire expensive ML engineers or data scientists.
As the model continues to process data, its accuracy continually improves. This company uses AI to ingest more document types and further reduce turnaround time while keeping accuracy near perfect.
AI and ML allow every player in the supply chain to turn data from a source of added cost and delay into a competitive advantage. AI and ML increase revenues, profit margins, and customer satisfaction by extracting more information more accurately from more documents, and more types of documents, at lower cost.
Unlike either generic or intelligent OCR, AI/ML eliminates the need for custom templates or custom code to accommodate new fields and does not require manual data checking. By providing more accurate information, it reduces the risk of shipments getting stuck in customs; missing the ships, planes, or trucks on which they should travel; or triggering extra inspections or the loss of business licenses.
Best of all, AI/ML helps businesses meet and exceed their customers’ expectations for faster, more accurate, and more reliable delivery despite rising shipment volumes and supply chain disruptions.