One of the most exciting aspects of Artificial Intelligence (AI) in logistics is that there is a huge number of applications impacting the industry ranging from data cleansing, demand forecasting and price optimization to autonomous trucks, last-mile delivery robots, and more.
With the ongoing evolution in supply chain digitization, more companies are already trying to implement AI-driven technologies into their operations in order to become more efficient and cost-effective. A cross-industry study on AI adoption by McKinsey & Co. found that early adopters with a proactive AI strategy in the transportation and logistics sector enjoyed higher profit margins.
That is why Transmetrics decided to make the focus of this year’s conference on exploring both the basics of AI, machine learning and data science, as well as successful use-cases to show how some of the logistics industry innovators are getting benefits from AI already today. The fourth Logistics meets Innovation conference took place in Brussels on May 28th, 2019 on the Vlerick Business School campus. The event gathered 50+ senior executives from logistics and supply chain companies including Amazon, DHL, ECS, Kuehne + Nagel, LKW Walter, Lufthansa Cargo, NileDutch, Panalpina, Pfizer, Novartis, and many others.
Moderated by Tim Phillips, Senior Vice President Development at sennder, the conference featured four insightful keynotes:
– “Precision Pricing with AI” by Jonah McIntire, Founder and Managing Director at TNX Logistics
– “Enhanced Dispatching via Augmented Intelligence” by Nicholas Minde, Senior Vice President Overland Germany at Kuehne+Nagel
– “Augmented Intelligence for Logistics: Behind the Scenes” by Asparuh Koev, Founder and CEO at Transmetrics
– “Machine Learning vs. Rule-Based Approaches to ETA Calculation” by Vincent Beaufils, Director/Chief Digital Officer at LKW WALTER
“Precision Pricing with AI”
During the first presentation, Jonah McIntire, Founder and Managing Director of TNX Logistics, shared that there is a real lack of understanding of how to transfer AI from a lab setting to the practical logistics in a successful way. Mr. McIntire presented a 5-step framework on how non-technical business leaders can evaluate proposals about potential AI-based projects. Below you can see a short overview of the criteria from this framework:
- Is this a high-frequency decision?
Machine learning algorithms improve their accuracy over time with more data and feedback about results from the model. Therefore, it is crucial for the potential project to generate frequent decisions and get clean feedback about the results.
- How data-driven is this decision?
Hard data questions with quantifiable results are a good case for AI. But if the areas to be evaluated are more qualitative, like customer satisfaction, employee satisfaction, or brand image, then AI might not be the best solution.
- Do we have the data quality?
One of the main differentiators between the best-case and worst-case scenarios in AI typically comes down to data quality. If a company expects a good output, it needs to feed the model with complete, accurate, timely, and consistent data.
- How important is Explainability?
It is often difficult or impossible to explain how powerful AI solutions get to the solution. If your logistics or supply chain company requires to know how the model calculated the result, what was assumed when making the decisions, how the decision-making process can be corrected and so on, then you are talking about a simple automation software with hand-crafted rules and not AI.
- Can we tolerate some mistakes
Machine learning tools always start at the least trained moment, so the mistakes are expected, especially in the beginning. Those are not failures but the way machine learning models learn and improve. One way to mitigate this risk is by having a person with the authority to decide whether to execute the software’s suggestion or not.
Mr. McIntire then presented a practical example of AI software for spot truck procurement developed by TNX Logistics. He demonstrated how for each shipment, their AI models calculate a tendering strategy and suggest carrier-specific offers, which include when and to whom to make offers, pricing, and bundling of work.
The keynote concluded with an interesting view on AI entering the workforce and how some people can benefit from the technology that eliminates the worst/repetitive part of their jobs while at the same time posing a risk for others by eliminating some jobs entirely. That is why logistics companies need to pay attention to the organizational transformation in order to help their staff accept the new AI software and take adequate measures to keep employees motivated.
“Enhanced Dispatching via Augmented Intelligence”
In the second presentation, Nicholas Minde, Senior Vice President Overland Germany at Kuehne + Nagel, discussed what makes road forwarder successful and how an old-fashioned road forwarding industry can benefit from AI technologies.
His presentation included several key messages. He shared that road forwarding companies are typically good in knowing what went right or wrong yesterday, but they don’t know if it means anything for tomorrow. Also, on average this part of the industry is quite old-fashioned and no-tech often beats high-tech in it. The main reason is not that road forwarding companies didn’t try to use modern approaches, but that at the end of the day, what will often succeed is a person solving the problem with a supplier or customer over the phone.
Nevertheless, in Mr. Minde’s opinion, road forwarding can benefit from “selective revolution” – using “augmented intelligence”, which combines the knowledge of the logistics experts with powerful computer algorithms. This approach is especially helpful in solving challenges that are not particularly visible, such as truck scheduling or dispatching. If implemented correctly, pragmatically applied mathematics algorithms can be empowering in logistics, and empowering technology can get adopted quickly even in large organizations.
In the next part of this presentation, Mr. Minde gave an example of how Kuehne + Nagel uses an AI-based technology developed by Transmetrics in order to reach better throughput and efficiency for the hubs. It was achieved by having a reliable prediction of loading based on shipment data and a prediction of capacity requirements (“We need X trucks to Y tomorrow”) based on multiple sources such as local shipment data, live network shipment data, and predictions based on historical data. That helps planners come to an agreement if they need to purchase additional trucks in advance and do so more efficiently and at cheaper rates with foresight, which in the past was done based on gut feeling.
To make AI projects successful, Mr. Minde advised the audience to be highly selective from the beginning, to focus only on the parts that logistics companies can’t do themselves well, to find a solution that fits the existing business model, and to not aim for perfect results – they just need to be accurate enough to achieve real business benefits.
“Augmented Intelligence for Logistics: Behind the Scenes”
Asparuh Koev, Founder and CEO of Transmetrics, started his keynote from pointing out the fact that the current hype around AI in logistics is mostly connected to exciting technologies like automated warehouses, self-driving trucks, drones, delivery robots, and more. Those are technologies that can replace some manual jobs that don’t require complex skills. For high-skilled positions like logistics planners, Mr. Koev suggested that AI technologies should empower employees instead of replacing them.
There is a lot of room for technological improvements in the logistics planning area. Today, planners typically use their TMS together with Excel and often make decisions based on experience and gut feeling. That leaves about 10% optimization potential – not something to ignore in a low-margin industry. To improve logistics planning, companies should use AI, though not in the sense of “Artificial Intelligence”, but “Augmented Intelligence”. Augmented Intelligence combines inputs from human planners (experience, responsibility, customer service, flexibility, common sense, etc.) together with Artificial Intelligence technology which is left doing the repetitive and tedious work, using all the available data, updating models with recent data, providing fall-back decisions if planner is not available, and more.
At the current state of the art, AI helps planners in real logistics operations by using intelligent predictive alerts. These might include predicting Time of Arrival for trucks based on traffic conditions, predicting container repair requirements based on GPS tracker shock detection, predicting goods damage and insurance claims based on temperature sensors, predicting high future shipping demand based on multiple variables, predicting warehouse employee sick days based on public holidays and weather, and more.
The next step for AI in logistics planning is the so-called “Human-in-the-loop AI” approach and that is what Transmetrics already offers with their predictive optimization solutions for cargo transport. It works by training AI algorithms that offer suggestions to the planners who then make a choice whether to accept them or to change them. Afterwards, the algorithms capture final decisions from the planners, compare the outcomes between human and AI suggestions, and use this input in order to further train the AI. Mr. Koev also shared concrete Transmetrics use-cases. For example, they help a large chemical company with matching customer orders to concrete ISO tanks based on a number of variables. The AI-driven software then provides a set of up to 20 optimal tank assignments per order and logistics planner makes the final decision.
After discussing several challenges of the “Human-in-the-loop AI” approach, Mr. Koev concluded the presentation with the ultimate goal for AI in planning, the so-called “Pilot in the plane cockpit”. This is the concept where AI automatically does all the calculations and suggestions, and planners only interfere when needed to account for unforeseen factors. Mr. Koev shared that Transmetrics is already successfully testing this approach for several customers – to optimize dispatching and steer trucks based on actual and predicted orders in the FTL/LTL sector, and to optimize the logistics of empty containers in the container shipping industry.
“Machine Learning vs. Rule-Based Approaches to ETA Calculation”
In the last presentation, Vincent Beaufils, Director/CDO at LKW WALTER, talked about the importance of AI-powered calculations for Estimated Time of Arrival (ETA) based on the example of LKW Walter. The main goals for the company were to improve customer service and transport monitoring, but also to increase automation and to drive internal optimization since ETA is key to generating triggers in the supply chain.
Mr. Beaufils then demonstrated how LKW Walter used the Internet of Things, Big Data and Artificial Intelligence in this project. Calculations involved multiple factors such as distance, speed profiles on roads, real-time traffic, loading/unloading points, waiting time profiles, 3rd-party data, weather, customer requirements, driving time and drivers’ rest, and more.
ETA calculation is structured as a pipeline, where different models are used for calculating each step of the pipeline. For loading/unloading/POI waiting time profiles, LKW Walter predicts drivers’ behavior by using classical statistical models. To predict the remaining driving time, the company uses several approaches. The first one is a classical hybrid rule-based system, which includes combinatorics and heuristic process. The second approach is Machine Learning, which works well on the linehauls where a lot of data is available. Each machine learning model there is divided into sub-models, and initially training each of the sub-models makes the process really complex.
Mr. Beaufils concluded the presentation with an argument that if you compare rule-based vs Machine Learning approaches, no model is better than the other, and logistics companies need to balance between the two of them. LKW Walter noticed that simpler algorithms (k-Nearest-Neighbor, random forest) performed better than neural networks and deep learning for the situations with lower data quantity. For the geographical areas with more historical data available (e.g. Western Europe) or for B2C logistics companies, more complex algorithms and machine learning tend to give better results. Nevertheless, even though Artificial Intelligence and machine learning models are very resource-intensive to set up and train, once it’s done, the models mostly just produce the results without the need to work on them further.
Panel Discussion and Conclusions
The keynotes were followed by an interactive panel discussion, which provided both speakers and the audience with an opportunity to explore additional aspects of Artificial Intelligence in logistics. In particular, conversations were around the topics of data quality issues in logistics, challenges of selling such technologies into companies, costs and ROI of such projects, and more. The panelists shared an opinion that it’s not that expensive to trial these technologies. The founders of both TNX Logistics and Transmetrics confirmed this argument by the facts that TNX uses the approach “pay if it works” with their customers saving about 4% on average on their transport, while Transmetrics offers their software on a monthly subscription basis with their customers saving on average 8-10% of the transport costs. The panelists representing large logistics service providers also raised another argument related to the project cost consideration – what will happen in the near future if your business doesn’t invest in these technologies now?
The conference concluded with a dinner reception during which the participants had a chance to network with other logistics and supply chain executives as well as to discuss the topic in an informal atmosphere of Vlerick Business School.
Overall, the event provided further insight into the practical issues of Artificial Intelligence in logistics both on the technological and business sides. We hope that as a result of the conference, we managed to inspire the attendees (and now our blog readers, too) and helped them understand where to start with the implementation of AI-driven technologies in their own logistics businesses. See you next time at the fifth edition of Logistics meets Innovation!