Inside Logistics

How AI is changing logistics

Current use cases aren’t about eliminating jobs, as some believe, but about helping supply chain professionals cope with the challenges of a high-mix, low-volume environment.


December 24, 2019
by Jacob Stoller

Blue jeans used to be simple. Fifty years ago, they were loose fitting denim pants made for physical work or relaxing on weekends, and people didn’t wear them to restaurants. Today, there are literally thousands of varieties for every possible style, fit, and venue, and many manufacturers sell multiple brands through a variety of store-based and online retail channels. 

“Nowadays, we’re seeing a lot of people who want to personalize what they get and stand out a little bit,” says Montreal-based innovation specialist and growth advisor Keith Blanchet. “And that turns the offer from companies into much smaller batches by high-mix, low-volume. So we’re having to evolve more technology to adapt to that environment.”

One of the major challenges for logistics providers, explains Ben Humphries, head of global pre-sales at Montreal-based AI solution developer Element AI, is that the technology they depend on is based on earlier high-volume paradigms. “A year ago, we looked at the entire supply chain from raw materials all the way through to the consumer,” he says. “What we found is that all the solutions that exist today are geared for yesterday, when it was a producer-push model through the supply chain.”

Machine learning

The bulk of the AI solutions being developed in logistics utilize machine learning, a subset of AI. Essentially, machine learning apps employ a variety of techniques, depending on the application and data involved, to improve their performance against defined objectives as they learn from the data.

The technology brings two key strengths to this new environment. First of all, it is able to execute tasks or deliver actionable information based on vast quantities of data from diverse inputs such as sensors, GIS devices, hand-written notes, audio files, video, emails, EDI data, or ERP systems. And unlike traditional apps that depend on fixed algorithms, machine learning apps are able to rapidly discover new approaches to highly complex problems, and also adapt in real time to changes in variables such as shipping costs, product volumes, or partner relationships.

AI, however, is not a quick fix – solutions take time to develop, and many application areas are just emerging. Furthermore, contrary to what many pundits are saying, AI is not likely to replace large numbers of humans anytime soon. “I don’t see massive layoffs,” says Humphries. “I think the challenge is that there are not enough people for the jobs that are there right now. That’s going to become even more challenging over time.”

New productivity tools

Much of the focus on AI solutions in logistics is aimed at helping supply chain professionals at all levels be more productive. Interestingly, people at IBM often refer to AI as “augmented intelligence”.

“This is about finding ways to make the work better,” says Jennifer Van Cise, VP global sales, IBM Sterling Supply Chain. “Augmented intelligence, as we call it, is actually about making work easier, smarter, and better rather than just replacing workers. It’s about helping the professional manage through the increasingly complex supply chain world that confronts us today.”

“I see AI helping take some of the more mundane repetitive tasks,” says Humphries, “and empowering and enabling the associate to take on more interesting roles. AI solutions become their assistants and their colleagues to some extent.”

What’s new is how well AI can perform some of the more familiar automation tasks. “Maybe it’s something as ordinary as document processing, such handling bills of lading more quickly in order to improve the movement of product around the warehouse,” Humphries adds.

Many AI solutions learn by imitation. For example, if an associate is engaged in repetitive actions such as filling out routine forms, AI apps can learn from monitoring humans how to autofill those forms, reducing the time and tedium required to execute these tasks.

Some of these capabilities are finding their way into out-of-box cloud offerings such as the Oracle NetSuite’s supply chain software, which now includes a feature called Intelligent Automation. “The system starts to learn how each individual user interacts with specific screens and automatically adjusts the screen layout to become easier to use for that particular user,” says Mississauga-based Gavin Davidson, product marketing director at Oracle NetSuite. The system also gives the user tips on profitability, or the likelihood of an order shipping on time, and such advisory capabilities are evolving rapidly.

AI also breathes new life into existing technologies that are decades old but haven’t worked well in complex environments. Optical character recognition (OCR), for example, which turns hard copy into digital documents, previously only worked reliably with printed text. Today’s AI-powered OCR apps can read the scribble of a harassed shipping clerk, adding a powerful tool to AI’s big data collection capabilities.

AI has also supercharged the familiar search engine. Element AI Knowledge Scout, for example, is a similarity search engine with natural language processing (NLP) capabilities which allows users to converse with the system rather than trying to guess which search criteria to use. “It presents answers in a consumable way,” says Humphries. “If you ask it for monthly figures, it will draw you a chart.” And of course, it keeps on learning.

Taking over the repetition

AI also promises to renew efforts to automate some of the more repetitive physical tasks in warehouses and distribution
centres, such as pick and place, by creating solutions that can adapt to the environment.

“To get ROI from automation in the past,” says Blanchet, “you needed to have high volumes, and then you would automate that task and the perform that task exactly the same way time and time again as fast as possible.”

What we’re likely to see soon, he explains, are smart robots equipped with vision and other sensors that can navigate unstructured environments within distribution centres or warehouses. “That’s where we’re going to start seeing the most return on investment,” he says.

Executive level tasks

Many of the situations being presented to AI solution providers weren’t anticipated a decade ago. A clothing manufacturer with multiple brands and channels, for example, was recently developing a process whereby a customer could return an item to a different store from where it was purchased, even if the second store sells the product under a different brand name. “This gets very complex from a logistical standpoint, especially when you get into the inter-company issues,” says Davidson. “For example, who bears the cost?”

On the strategic side, AI allows organizations to transition from a fixed-rule information environment to one where the rules can change according to varying business conditions and goals. For example, an organization facing new pressures to improve customer satisfaction may need to adjust its policies for routing shipments in order to place greater emphasis on delivery performance.

AI capabilities also make it possible for supply chain professionals to answer to a broad range of corporate objectives. In a recent example, Seattle-based outdoor clothing retailer Recreational Equipment Inc (REI) adopted IBM Sterling Fulfillment Optimizer with Watson to improve its supply chain performance against multiple KPIs. In a video interview with IBM, REI’s SVP supply chain Rick Bingle explained that supply chain professionals need to look beyond just managing costs.

“I would really caution us around the word of optimization,” says Bingle “because in supply chain management, we often think of that as distribution costs, freight costs – really thinking about costs. We have to think about the margin impact, the customer experience, how we can drive revenue.”

“People think of the supply chain as a cost centre,” says Van Cise. “But it’s really the heart and lungs of the business – you can’t move forward without it. So the challenge is being able to balance a myriad of KPIs, even if they are at odds. That is, balance the cost of the supply chain, but at the same time ensure customer satisfaction through quality and timeliness of execution.”

Sometimes executives just want answers to the age-old question, “where is the order?”

Visibility into the supply chain has also been a priority of Oracle NetSuite’s supply chain customers, and in response, the company has released a module called Supply Chain Control Tower, which provides an end-to-end view of the supply chain. Using Oracle’s machine learning platform, the program assesses risks based on purchase order and vendor performance. New features continue to evolve.

Companies seeking to leverage the advantages of AI in their supply chains have many options that range from utilizing out-of-box tools in enterprise software like NetSuite, to building a custom solution with a team of data scientists. Canada boasts a strong community of young AI companies like Element AI, and AI developer platforms like Microsoft Azure are in wide use and have been adopted by a number of solution partners.

It’s all about the data

All AI journeys, however, begin with data. “You need to find where your data strengths lie, and where the gaps are,” says Dr. Alexander Wong, University of Waterloo engineering professor, Canada Research Chair in the area of artificial intelligence, and a founding member of the Waterloo Artificial Intelligence Institute. “For example, there may be a lot of regional demographic information that helps guide stocking information. So you need to ask if you’re collecting the right data related to that.”

It’s also critical to eliminate extraneous data, in part because machine learning depends more heavily on data than traditional apps. “Even if we have the right data, it might get obscured by less useful information,” says Wong. “There are factors that are less meaningful, and they can overwhelm the contribution of the useful factors. Then you end up with bad forecasting and bad predictions.”

Once the data is selected and validated, the next step is establishing an initial set of rules for the app to operate by. “You start out with a rules-based system where certain assumptions are made,” says Wong. “Then you use machine learning to relax those assumptions and learn what they should be.”

The process requires that supply chain professionals determine what they want from their AI app, and then continue to work with it to ensure that it performs according to objectives. Essentially, people will first teach and guide the technology and then learn from it.

“You’re building a framework,” says Humphries, “and you can ramp up as slowly or as fast as you want. Everybody should explore this.”

For Van Cise, the technology is ultimately a tool to help supply chain professionals improve what they do. “At the end of the day, we’re talking about our people,” says Van Cise. “How can professionals excel in this changing world. I’m sure the same conversation was happening 60 years ago with increased automation in factories and warehouses alike. For me this is about the constant need for continuous improvement of your own areas of expertise.”