Until a couple of years ago, many companies had to rely on research and analyst teams to process and benefit from the data they generated. Artificial Intelligence (AI) has huge potential for optimizing and automating these processes and achieving excellent value improvements by applying data-driven decision making. Camelot expects a major acceleration in AI use as companies are opting for digitalization to remain competitive.
This particularly applies to logistics, the focus area of this blog post. Since AI encompasses a wide range of topics and disciplines, we want to narrow it down to the most relevant methodologies and the data science process flow.
Logistics managers are constantly facing uncertainties related to scheduling, loading/unloading timings, freight costs, freight forwarders, and so on. Wouldn’t it be great if you could adjust and implement the necessary changes instantly, depending on the changing environment? AI offers viable solutions for predicting the demand patterns, determining best loading schedules, choosing the cost-efficient freight routes and carriers.
Data-driven decision making will help companies to gain advantages over slower competitors. In real-life projects, Camelot helped clients experience a much smoother process flow in logistics execution and freight billing. AI was used for better predictions in loading meters estimation, smart tendering, freight consolidation and more (details to follow in the upcoming posts).
Nevertheless, it needs to be stated that AI is not suitable for all situations – not for all use cases and not for all industries. Both technical and business issues need to be considered, as well as the data availability. The approach of Camelot is to access each business problem individually, check feasibility for the data science process, and find a suitable approach according to individual business needs.
Figure 1: The Camelot Data Science Process will ensure a systematic and verifiably proof of value (PoV)
So what is AI all about?
Artificial Intelligence (AI) is not a technique, neither is it a program that can be installed. Camelot sees it as a path of transforming the value chain within a given organization. It always consists of three steps: Descriptive Analytics, Predictive Analytics and Prescriptive Analytics (see figure 1).
The first aspect in an AI-related task always lies in the descriptive analytics category. Here we are answering the question “What has happened and why?”. Next, we can turn to the predictive analytics which uses statistical models and forecast techniques in order to understand the future and answer the question “What will happen?”. After answering this question, we differentiate two AI directions – cognitive AI versus process AI. An example of cognitive AI can be a chatbot design offered by IBM Watson. It allows you to create AI-powered conversational user interfaces and is already used by enterprises across different industries (banking, telecoms, etc.). In contrast, the process AI path is used to automate business decisions. For instance, replenishment automation triggered by alerts. In summary, AI is a supportive technology to deliver value by either automating process tasks and/or by providing superior user experience (cognitive interfaces). Most often, it’s a combination of many different techniques. In any case, however, we need to access many data layers in order to apply AI techniques and eventually transform business operations.
How to start an AI project in logistics?
The first step is always understanding the business objectives. However, this is paired with an initial feasibility estimation from experienced data scientists. Following a defined data science flow helps lift the value through data work.
Figure 2: CRISP
One of the best-known flows is Cross-Industry Standard Process for Data Mining (CRISP), originally designed for data mining projects. Camelot’s Data Science Process, tailored to value chain management, is derived from this basic principle as well. Our experience has shown that the CRISP flow is a very good framework to start with.
Before any AI technique is implemented, the raw data needs to be cleaned up. This process of cleaning and normalizing the data is more time and cost consuming than most companies anticipate.
To achieve the desired results, a clear objective must be identified. Otherwise, you will get lost in data, time and lack of resources due to missing priorities. An effective AI implementation is always a process of considering individual cases, adopting the data analysis methodologies (e.g., Machine Learning) and ultimately transforming the business towards digitalization.
We would like to thank Frank Kienle for his valuable contribution to this article.