[REPORT]
Artificial Intelligence in Logistics: 6 applications
The COVID-19 pandemic has heightened the need for supply chain players to implement tools that help them make better decisions rapidly. This has driven a significant trend towards accelerating the adoption of advanced technologies like AI to analyze the vast amounts of data they generate, in order to better understand and accurately predict their operations.
AI-driven digitization projects, while highly promising, still carry inherent risks that need to be managed. If poorly prepared, these projects can become a financial sinkhole, with many failing to deliver the expected results.
Why is this? Is an AI project just like any other project? What are the risks to be aware of before starting? And how can they be anticipated?
In this article, following our previous piece, “Artificial Intelligence for Logistics: 6 specific uses and benefits” we outline the prerequisites and the most suitable method for successfully implementing an AI-based tool in a business.
Prerequisites for implementing AI tools in a business
1. Choose the right use case
Artificial Intelligence (AI) is currently a hot topic, but it’s important not to adopt it for the wrong reasons. Market pressure is pushing many industries to integrate AI technologies to stay competitive. While AI is essential for some use cases, it may not always be necessary. Before committing to an AI implementation process (which can be quite extensive), it’s crucial to evaluate its relevance to the specific problem you want to address and the benefits it will bring.
I attended the implementation of a project by a company that wanted to digitalize volume and production forecasting, particularly to anticipate future revenue. By investing a lot of time and money, they used advanced techniques such as Deep Learning and probabilistic forecasting, but the results improved by only about 1%.
In this specific case, the added value of AI is quite limited. The management controller was actually satisfied with the level of precision he was able to achieve without the tool for his needs. Therefore, the challenge is to ensure that Artificial Intelligence is suited to the problem at hand to avoid investing a lot of resources for minimal return on investment.
Nevertheless, the right approach is to identify the specific use cases to address first and then determine which technologies can meet those needs, rather than the reverse—trying to apply AI technology just because it’s trendy or used in the industry.
2. Ensure quality data
Data is at the core of Artificial Intelligence technologies. Without quality data, an AI project is destined to fail. Some companies may have a real use case to solve and want to engage in such a project but lack the necessary data or have incomplete, unstructured, and therefore unusable data. Simply setting up databases is not enough to ensure good data quality.
There is often a significant gap between field reality and the data stored in databases. In most cases, it is the operational staff who manually enter this data. They need to be made aware of the importance of accurately entering the data they collect: every address, every measured weight, every dimension must be recorded and entered into the system. This requires a substantial shift in practices that needs to be anticipated, especially with field teams. It also necessitates a different organization, or even the creation of dedicated roles to maintain data quality over time.
Risks of AI projects in business
1. Reproducing human biases
Implementing an AI project sometimes involves using techniques such as Machine Learning or Deep Learning, which rely on historical data to enhance their understanding of real-world activities. Among this data, there may be errors or human biases that the AI can reproduce, as was the case with the Microsoft chatbot in 2016.
This Twitter account, powered by Artificial Intelligence, was intended to interact with users based on publicly available data and responses previously crafted by Microsoft teams. Within hours, the account became racist and misogynistic, making statements that shocked the world.
In logistics, calculated intervention times might be based on erroneous data. For example, the machine might detect that a specific delivery always takes 10 minutes while others take 5. What the machine doesn’t know is that during these 10 minutes, the driver usually takes a coffee break with the client nearby. The machine lacks the accurate information to analyze the situation and refine its understanding.
2. Failing to manage change effectively
Change management is a key factor in the success or failure of an AI project. Such projects inevitably lead to changes in how teams work, which can create uncertainty or resistance among employees who may see the technology as a potential threat to their jobs. It is crucial to keep the human element at the center of processes. At Kardinal, we focus on drawing a clear line between human and machine in the design of our solutions.
Find our article on the role of humans in AI solutions for industry.
The best strategy is to delegate complex calculations and machine learning from historical data to AI to support human decision-making and allow people to focus on higher-value tasks, such as customer relations, personnel management, and strategic decisions based on in-depth knowledge of the field. AI is not meant to replace humans but to assist and enhance the value of their work.
It is essential to steer the project in this direction and involve subject matter experts from the beginning. The project will be more readily accepted if all collaborators understand its advantages and benefits (e.g., easier daily tasks due to AI and the ability to deliver better performance). As implementing an AI project may involve some constraints (such as being more rigorous in data collection and entry), it is crucial to demonstrate that the tool will address other pressing issues.
3. Choosing an inadequate tool
One of the challenges in implementing an AI project is being poorly supported and not having the right tool or approach. Navigating through the multitude of AI solution providers on the market can be daunting. Some have their own highly technical jargon that may be confusing for industry experts. The key is to partner with a provider who speaks your language and understands your specific problem. If they do not fully grasp your issue, they will not be able to address it effectively.
Whenever possible, testing the proposed solution in advance and in real-life situations allows you to quickly assess its relevance. Successfully implementing an AI project requires leveraging the knowledge and expertise of people. It is crucial that AI adapts to the field, not the other way around.
What project approach to implement an AI tool in a company?
1. Avoid a Big Bang IT approach
An AI project is sometimes associated with a major overhaul of the information system. Some companies might be tempted to embark on a “Big Bang IT” strategy, meaning starting from scratch and completely disrupting the existing organization. However, this strategy is not always optimal. Given the massive scope of such an approach, employees may become disengaged from the project, especially since the real benefits might take some time to materialize. The fatigue from implementing new data collection processes will inevitably affect data quality, leading to potential declines in the accuracy and reliability of the information collected.
2. Involve the employees
Initially focusing on projects with very short iterations that deliver value quickly helps to engage teams more easily with the added constraints of such projects. By demonstrating the benefits of the tool as soon as possible, employees will be more motivated to participate in the project development and will better understand its usefulness. For example, collecting addresses qualitatively allows for better geocoding, which results in fewer delivery errors and less time wasted. Such concrete results applied to the field show the teams the value of the tool and highlight their role in the project: thanks to their quality data entry, performance improves significantly.
3. Start small to go the distance
Our recommendation is to work in cycles rather than engaging in colossal, lengthy, and costly changes. The best way to test the relevance of your project is to implement it step by step, starting with small, achievable use cases within very short timeframes, limited budgets, and a small project team. By demonstrating locally that the project adds value, it can later be scaled up. To achieve this, it is essential to have pre-planned how to measure this value.
Additionally, you can leverage the success of the local project, using positive feedback from those interacting with the tool on the field, to demonstrate the value of scaling the tool to all use cases. This spiral approach involves running very short iterations on smaller, perhaps less ambitious AI projects that quickly deliver a return on investment (ROI). This allows for subsequent cycles that are longer and more ambitious. The change culture develops more smoothly by gradually acculturating employees to this new way of working, especially in logistics where work methods can still be quite artisanal.
In conclusion: Is there an AI project?
Here are the questions that need to be addressed before embarking on an internal digitalization process:
To implement an AI project, you need to:
- Face a challenge where technology can add value,
- Identify that AI is truly the solution to this challenge,
- Have sufficient and, most importantly, high-quality data,
- Ensure you have the means to manage change effectively.