Shipping is one of the oldest trades in the world, and one of the foremost measures of a country's economic growth and prosperity. Be it coffee from Africa, dark chocolate from Indonesia or the latest appliances and gadgets from China and Taiwan, your preferences and affinities rely more on shipping companies than you might imagine. Around 80% of all goods today are carried via sea. Global maritime container trade accounts for 60% of all seaborne trade and was valued at $14 trillion in 2019. Among several organisations that manage the flow of sea traffic is Maersk, a Danish shipping company - and the largest container shipping line and vessel operator in the world for nearly 25 years now. 

In the past decade, shipping companies have begun using a host of AI technologies to make processes more streamlined and efficient. While ships - be it commercial or military vessels - themselves have evolved over time to become supremely sophisticated carriers, on-ground support functions and operations were still antiquated. But with the emergence and popularity of AI, critical support functions are getting a revamp. 

For instance, take order handling and bookings. Maersk operates in 130 countries and specifically manages 150,000 booking requests in a month. Last year, Maersk Technology Centre in Bangalore carried out a hyper-automation initiative that saw the development of a homegrown, production system that streamlined order handling and bookings, leveraging AI and Robotic Process Automation. 

Sunil Kumar, Engineering Director - Data Science, Innovation and Digital Automation, Maersk Technology Centre said, "One of the main problems we faced was the flow of order requests through multiple channels. In addition to the Maersk website, customers would send queries to EDI and INTTRA - third party websites for managing incoming orders. In addition to keeping track of multiple channels, customers would also make several amendments to order such as specifications of the vessel needed, the kind of cargo being transported, timelines etc. These variables can be quite challenging to manually keep track of." 

Adding that reading free text was the specifically challenging aspect of constantly amended requests, Kumar explained that this was the case not only because the text was in Natural Language but also steeped in business jargon - and unfamiliar for anyone outside the supply chain domain. "There is a specific taxonomy associated with the domain we specialise in and highly trained operators manage these operations daily. Yet, there is scope for mistakes and errors as these are repetitive tasks," he adds.

So, the Maersk Technology Centre developed an AI module that integrated RPA, named entity recognition, BERT embedding, Regular Expression (REGEX) to provide deterministic solutions to internal users. They used large volumes of training data and historical data from booking amendments processed by teams, trained the NLP model, extracted patterns and hosted this on Azure as an API. Whenever a booking was made, it was routed to this production system where the APIs performed entity extraction and synthesis. This was converted into structured data for the benefit of other teams. 

Primary challenges the team faced included data privacy, the scope of curation, scarcity of linguistic components and addressing pre-conceived notions on the efficacy of intelligent automation. But the benefits experienced from this upgrade were multi-fold. The AI module was processing around 17,000 transactions per week, saving nearly $36,000 per year. In addition, with the AI module in place, the accuracy rate was 97.4% instead of the 88% that the teams would average, added Kumar. Moreover, the team involved in setting up this technology platform were software engineers who worked on automating systems, gaining valuable AI experience along the way. "Our team of developers grew from software engineers to data scientists. Many of their careers have changed due to projects like this," said Kumar. Currently, a team of 120 engineers in India, Copenhagen and Chengdu work on order handling. 

AI-based automation of bookings for transport and logistics was among many other projects taken on by Maersk to improve business efficiency across the board. Now, data scientists and engineers are analysing customer requests made in voice, developing vessel utilisation plans using AI and predicting container deployment based on customer requirements. 

Industries like shipping are powered by complex transactions daily, with little margin for error as even a tiny mistake can cost them millions. But these industries are also data-rich, process-driven and rely on automation to execute repetitive tasks with near-zero error. These are the right ingredients for an AI system to thrive, perform to its potential and aid these industries with significant cost savings.