Should supply chain leaders invest in generative AI solutions?

This is a question I am asked quite often during our collaborative projects and other industry engagements. My answer is always a qualified yes based on a supply chain leader’s answer to three basic questions. Let me explain.

1. Does your organization have an established AI policy that will guide your project?

   Generative AI (GenAI) platforms have made AI widely accessible to employees across organizations. This has made it increasingly important for businesses to establish clear policy guidelines that balance perceived benefits with potential reputational risks. Companies I have spoken with in the last 12 months are at various stages of establishing AI-related policies that align with their existing data governance frameworks. In organisations with established policies around the use of AI, there are often limitations placed on the use of sensitive information on GenAI platforms. In businesses without an established policy, supply chain teams are often exploring different GenAI tools without a coherent objective. Without an established policy, there is no clear project approval process that can result in a deployable outcome that would meet business needs. Companies looking for resources on AI policy development would find it useful to explore publicly available policy content created by government agencies (e.g.,  Generative AI: basic guidance | Digital.NSW) and frameworks shared by solution vendors to communicate the trustworthiness of their offerings (e.g., IBM Artificial Intelligence Pillars - IBM Policy).  Given the data-driven nature of operational decision making, supply chain leaders can make a significant contribution to the development of such policies within their organisations. GenAI can also be part of the solution in this context. For example, solution vendor, Dataiku, provides a GenAI-based policy and regulations explorer that can help a company develop its approach to AI governance (AI Policy & Regulation Explorer | Dataiku).

2. Does your problem need a GenAI-based solution?

   If the organization has a policy with clear guidance on the application of GenAI, then a supply chain leader needs to consider whether the problem does indeed need a GenAI-based solution. My concern here is the overuse of GenAI because of the easy accessibility of a ChatGPT prompt. My team and I are currently using both GenAI and predictive AI models in our supply chain projects with industry partners. A simple but useful comparison of the two types of AI models is available here: Generative AI vs. predictive AI: Understanding the differences | TechTarget. Predictive AI involves the use of machine learning to identify patterns in historical data (demand, price, sales, etc.) to make predictions about the future. In our predictive AI projects, comparing the performance of different machine learning models has become standard practice. This is due to the complexity of models at the deep learning end of the spectrum and the fact that the highest performing models are also often the least explainable. The increasing drive for making AI applications more explainable to decision makers has led to the emergence of the field of explainable artificial intelligence (XAI) but there is still considerable progress to be made in this area (DARPA's explainable AI (XAI) program: A retrospective - Gunning - 2021 - Applied AI Letters - Wiley Online Library).

   GenAI differs from predictive AI in the sense that is designed to create new content based on user prompts. GenAI models (e.g., large language models – LLMs) are pretrained on large volumes of unstructured data and can be trained further on organization-specific datasets. While it is challenging to ‘look under the hood’ of many predictive AI models because of complexity, it is impossible to do so for proprietary GenAI models such as GPT-4 (openai.com), the most recent LLM from OpenAI. Use of GPT-4 to generate code for forecasting does not further the goal of producing explainable AI solutions for decision makers. It is also not productive to prompt OpenAI’s chatbot, ChatGPT, to generate a forecast knowing that hallucination is a significant problem for LLMs.  There are also several open-source LLMs currently available in the market and several lists compiled by various sites (e.g., 8 Top Open-Source LLMs for 2024 and Their Uses | DataCamp).  Most supply chain teams would not have adequate bandwidth or LLM domain knowledge to explore the range of models available to identify a suitable solution.

   We do not use GPT-4 for our predictive analytics projects but we do use it in projects involving the extraction of information from large volumes of internal company documents relevant to supply chain management (e.g., contracts, policies, and procedures). Some problems can also be quite easily addressed with traditional natural language processing (NLP) solutions and not require the use of LLMs.

   Identifying the most appropriate solution for an operational problem, whether it is based on GenAI or predictive AI models might be an area where expert advice is beneficial to supply chain teams if such expertise does not exist in-house.

3. Are the right skills and tools available to your supply chain analysts?

   By ‘the right skills’, I do not mean programming capabilities. I would never advise supply chain leaders to turn their analysts into data scientists. To be effective in their roles, supply chain analysts need supply chain domain knowledge much more than they need programming skills. Skills and tools are closely connected since smarter tools require fewer skills. With the growing availability of no-code/low-code AI platforms, supply chain analysts can now generate machine learning-based predictions using a sequence of simple visual drag-and-drop steps. The global no-code AI platform market is worth billions of dollars, so it is important for supply chain leaders to consider their strategic needs carefully and ask enough questions before selecting a platform. Once adopted, such a platform would be a far better option for predictive analytics than allowing analysts to prompt ChatGPT to generate forecasts as an alternative to acquiring programming skills. No-code AI platforms can also facilitate no-code access to OpenAI’s GPT models for projects where these are most suitable. Having a single platform for both predictive analytics and GenAI projects can also make it easier to train your analytics team and create a shared knowledgebase that will benefit future projects. Supply chain leaders also need to foster a continuous learning culture within their teams, so the organization benefits from the rapidly evolving AI landscape.

To supply chain leaders who reply in the affirmative to all three of the above questions, my answer is yes, you should absolutely invest in GenAI. The world is only at the beginning of our journey with GenAI. The possibilities are endless. As long as business leaders are making thoughtful strategic decisions that are mindful of emerging regulations and the company’s data governance framework, there are benefits to be gained from adopting generative AI models for the right projects.