Generative AI enhances procurement with smarter sourcing, supplier management and decision-making for better outcomes.
Unless you’ve been hiding in a cave for the past five years, you’ll be familiar with the term Generative AI, or GenAI. Yet the concepts behind the term, and those around it, are the source of much confusion. It’s time to unravel the jargon because, if one thing is certain, GenAI is going to impact virtually every aspect of our working lives, and not least, in procurement.
So, let’s start with some definitions to answer the questions, What does GenAI do? What distinguishes it from AI in general? What is GenAI vs ChatGPT? What’s the difference between LLM and GenAI? And what’s with this latest kid on the block, agentic AI?
Generative AI
Generative AI is a form of AI that not only analyzes existing data but also generates new content. It uses deep learning models. Deep learning is a type of machine learning (ML) that uses artificial neural networks to learn from data. The models learn from large data sets to recognize patterns and generate new content similar to the training data used to set them up. Artificial neural networks are inspired by the neurons in the human brain, the nerve cells that receive and conduct impulses. These artificial neural networks can be used to solve a wide variety of problems, including image recognition, natural language processing, and speech recognition.
Building on this foundation, GenAI can be used to write texts, create images, compose music, generate software code, or even create simulations and synthetic data. Generative AI has the potential to revolutionize many areas, from the creative industry to research and development, and, as we shall see in this series of articles, procurement processes such as RFQ and contract generation.
GenAI Versus Other Types of AI
Conventional AI focuses on solving specific problems based on predefined rules, whereas GenAI AI creates new content, like text or images, based on patterns learned from vast datasets. A typical conventional AI will learn from existing historical, structured data to determine the probability of a future event. For example, it is used in predictive maintenance to enable engineers to identify and fix a fault before it occurs.
Table 1 summarizes the differences between conventional AI and generative AI. In procurement, conventional AI has found application in areas such as predictions (for example, predicting the likelihood of on-time delivery of supplies), fraud detection, recommendation systems, and chatbots. GenAI is the upcoming technology. As mentioned, it already finds applications in contract and RFQ generation but areas being explored include operational research.
| Feature | Conventional AI | Generative AI |
| Functionality | Solves specific problems based on rules | Creates new content based on learned patterns |
| Data | Uses structured data for analysis and prediction | Uses vast, often unstructured data for generation |
| Adaptability | Less adaptable, requires human intervention | Highly adaptable, learns and improves over time |
| Transparency | More transparent, rules are predefined | Less transparent, learning algorithms are complex |
| Output | Predictions, insights, or decisions | New content, such as text, images, or videos |
Table 1: Conventional AI versus Generative AI.
What is ChatGPT?
If you have heard of GenAI at all, it is probably through ChatGPT, a generative AI chatbot launched by the firm OpenAI in 2022. It is based on sophisticated large language models (LLMs) that mimic conversation with a human being. ChatGPT is highly effective at generating text in response to questions—text which it can format according to user instructions on word count, style etc. And it will do so while observing the rules of grammar and syntax better than most humans. However, it can also perform other tasks such as composing music, drawing pictures or generating software code.
As a generic tool, ChatGPT can be of assistance to procurement professionals in many ways. For example, it brings efficiency benefits. ChatGPT can quickly generate responses, answer questions, and automate tasks, saving time and resources. In areas such as supplier relationships ChatGPT can be used to provide instant and personalized responses, enhancing supplier engagement and collaboration. It can assist with writing, brainstorming, and creating various content formats. And in a globalized economy, ChatGPT’s ability to translate languages accurately and efficiently can be a huge asset. facilitating communication across different languages.
That said, ChatGPT must be used with great care. It can often generate incorrect or misleading information, especially if it’s not trained on a comprehensive dataset. OpenAI acknowledges that ChatGPT “sometimes writes plausible sounding but incorrect or nonsensical answers”. This phenomenon, common to LLMs, is referred to as “hallucination”. While ChatGPT’s lack of creativity, and accusations of plagiarism, may not be a major issue in procurement, the security risks will be a concern. Using ChatGPT requires the exchange of data and information with the system. This poses potential risks in terms of data protection and security. Users must take appropriate measures to ensure that sensitive data is protected and does not fall into the wrong hands. Finally, over-reliance on the efficiencies of ChatGPT may diminish procurement professionals’ need to apply critical thinking and the emotional intelligence that is vital whenever dealing with other human beings.
Large Language Models (LLMs) Versus GenAI
A large language model (LLM) is a type of AI model designed to understand and generate human language. It’s trained on massive amounts of text data to predict and produce coherent responses. A generative pre-trained transformer (GPT) is a specific type of LLM — that’s the “GPT” in ChatGPT. It uses a neural network architecture called a transformer and applies techniques from natural language processing (NLP).
Today, most LLMs are built using transformer architectures and similar training methods, but the terms LLM and GPT are not interchangeable. All GPTs are LLMs, but not all LLMs are GPTs. Today, the largest and most capable LLMs are GPTs, but GPTs are very diverse. To use an analogy: an LLM is like a motor vehicle — a broad category. A GPT is a particular type of vehicle, designed for a specific purpose. Thus, GPTs can be as different as a tractor or a family saloon. Without diving too deep into this rabbit hole, individual models like GPT-4 are closer to specific makes and models of car. They are refined versions with particular capabilities, performance characteristics, and use cases.
Agentic AI – the “New Kid on the Block”
As we have seen, generative AI focuses on creating new content such as text, images and videos, or even music and software programming.
By contrast, the emerging technology of agentic AI is a can make decisions and perform tasks without human intervention. The AI agents automatically respond to conditions to produce process results. In procurement, for example, they can be used to make decisions on purchasing or choice of suppliers based on changing prices, risk factors, demand forecasts etc.
The Evolution of Generative AI
Perhaps more than any other technology, GenAI has been like traffic entering and then emerging from a traffic block. Slow but steady progress over a long period of time followed by a sudden outpouring! Its origins can be traced all the way back to the work of the Russian mathematician Andrey Markov, in the 1900s. Early text generators (like randomly constructed sentences) used Markov chains to choose the next word based on the previous one or two. We can see the application of this concept today in predictive text (and some of us still find it rather primitive!) In the 1940s and 1950s an American mathematician, Claude Shannon, made numerous contributions to the field of artificial intelligence, and co-organized the 1956 Dartmouth Conference, which is considered to be the discipline’s founding event. His Theseus machine was the first electrical device to learn by trial and error.
The first natural language processing system was named ELIZA after the character in Dr. Doolittle. Created by German American Joseph Weizenbaum in the mid-sixties, this chatbot mimicked a psychotherapist using pattern matching and scripted rules.
So-called expert systems emerged in parallel for applications such as the diagnosis of diseases. Based on if-then logic and manually curated rules, these were a key phase in AI development, although they did not yet have learning or generative capabilities.
Developments picked up pace in the 1980s and 90s with statistical natural language processing and machine learning. The n-gram language model, a purely statistical model of language, formed the basis for large language models. Decision trees, support vector machines (SVMs) and classical machine learning supported the shift to data-driven approaches—but were still not generative.
The same period saw the emergence of hidden Markov models (HMMs). These are essentially Markov chains with a hidden state that generates observations. In addition to natural language processing applications, HMMs proved useful for time series analysis, i.e., predictive analytics, such as forecasting stock prices or analyzing weather patterns, as well as robotic processes.
The 2010s saw the development of technologies such as recurrent neural networks (RNN), which are capable of generating sequences such as text and music. These were used for applications such as text generation and machine translation. For example, in 2016 Google Translate transitioned from statistical machine translation to RNN technology. Variational autoencoders (VAEs) and generative adversarial networks (2014–2015) provided a foundation for non-text GenAI.
In 2017, eight authors published a research paper called Attention Is All You Need, which set out the deep learning architecture known as the transformer. This was the key breakthrough in natural language processing that really unleashed the GenAI boom of recent years. OpenAI published its first versions of GPT3 in July 2020, providing the power for ChatGPT, launched at the end of 2022.
Table 2 summarizes the evolution of GenAI.
| Era | Milestone | Contribution |
| Early 1900s | Markov Chains | Probabilistic modelling of sequences |
| 1940s–60s | Shannon, ELIZA | Statistical language, scripted bots |
| 1980s–2000s | N-grams, HMMs | More powerful language models |
| 2010s | Word2Vec, RNNs, GANs | Vector semantics, generative models |
| 2017–Now | Transformers, GPT | Foundation for modern GenAI |
Table 2: The evolution of GenAI
What’s Driving GenAI Adoption in Procurement?
We’ll get into the details of GenAI applications in procurement in future articles. But first let’s consider, “Why has GenAI become such a hot topic?”
The adoption of GenAI in procurement is being driven by several converging pressures: the need for greater speed and agility in sourcing, an explosion of both structured and unstructured data, rising expectations for user-friendly digital experiences, and an ever-growing focus on risk, compliance, and sustainability. Traditional procurement tools, while effective in standardized scenarios, often struggle to adapt to the increasing complexity and volume of information that modern procurement teams must process. This is where GenAI comes into play, not simply as a time-saver, but as an enabler of deeper insights, better decisions, and more responsive procurement operations.
Procurement generates huge amounts of data, and LLMs with connectors make it possible to unlock data silos for generative AI applications. GenAI also makes it possible to bridge the divide between structured data, such as exists in ERP systems, and unstructured data, such as the content in emails and PDFs that form the basis for informal agreements and formal contracts. This is crucially important for accelerating procurement processes such as RFQ generation and contract management.
Benefits of GenAI in Procurement
Thus: technological progress only occurs and achieves widespread adoption when there is a business or social need. In today’s world, being slow means being vulnerable. In business, markets are more volatile, customer expectations are more immediate, and innovation cycles have dramatically accelerated. Taking decisions faster, and acting on them faster, is a business imperative in all activities—and procurement is certainly no exception. But speed is hard to achieve when you’re battling through fragmented systems, siloed data, complex processes, and a shortage of human capacity.
We will explore specific benefits of GenAI in procurement in another article, but the overarching benefit is that it significantly reduces friction, especially the friction caused by information overload. Procurement plays a central role in helping organizations respond quickly and responsibly by summarizing, contextualizing, and extracting meaning from masses of unstructured data, whether it’s scanning hundreds of supplier sustainability reports or interpreting shifts in commodity pricing.
Challenges and Ethical Considerations
The rapid march of AI can be unnerving, don’t be alarmed. GenAI is imperfect. It can never fully replace human beings, although it will certainly reshape and redefine the world of work, including the procurement function. We already alluded to the tendency of GenAI to “hallucinate”, in other words, instances where the AI model produces outputs that are factually incorrect or nonsensical, even though they might appear plausible and logical. If the user’s input to the AI is unclear or lacks sufficient context, or there is insufficient information, the model may make incorrect assumptions or attempt to fill in gaps in the information.
For example, let’s say a category manager asks a GenAI assistant: “Which supplier in Category A has the lowest ethical risk score based on the latest ESG data?” The assistant may immediately come back with the answer that Supplier X has by far the best score, with not a single blot on its copybook. Yet this might be because Supplier X has no recent reports or media coverage, maybe because they operate in a jurisdiction with poor transparency, minimal press freedom, and limited regulatory enforcement. This isn’t a sign of ethical excellence; on the contrary, it should be an immediate red flag. The hallucination here arises from confusing absence of evidence with evidence of absence.
This is where human intelligence trumps artificial intelligence. A procurement professional with the right training and domain experience will apply their critical faculties to dismiss such advice, or, better still, avoid walking into such traps in the first place by framing more sensible questions!
There are other ethical concerns with the use of GenAI. Because it is driven by historical data, it will tend to provide answers that replicate or even amplify past biases. This presents a challenge to any organization seeking to diversify its supplier base, whether this is to reduce risk or to pursue ESG policies.
A further issue is transparency, particularly in the public sector or in compliance-heavy industries. If a GenAI assistant recommends a supplier, rewrites a contract clause, or identifies a compliance risk, can the rationale be clearly explained? “Because the AI said so” is not an answer that any auditor will find acceptable! In such instances, human involvement is vital.
Finally, AI lacks the soft skills and tacit knowledge that are the foundation of supplier relationship management, such as an understanding of supplier culture, regional nuances, or the interpersonal dynamics. A buyer always wants to get a good deal, but a procurement professional understands the difference between a strategic supplier and a supplier you just want to beat down on price. Automated correspondence with suppliers might come across as aggressive, disingenuous or manipulative if not reviewed carefully.
All of this has massive implications for jobs and careers.
There is a common misconception that GenAI is only of relevance to “creative” industries. In fact, its transformative potential in traditional industries like manufacturing is arguably even greater — precisely because these sectors deal with complex systems, structured and unstructured data, and repeatable processes ripe for optimization.
Conclusion: We’ve Come a Long way in a Short Time!
While the foundations for GenAI were laid over decades, its development over the past few years has been breathtaking. Evolution has become revolution.
There is much to take in, so we hope this article has been a good introduction to the issues. We will explore them in greater depth. We want to show how and why, over the coming years, generative AI will continue to be integrated into JAGGAER software. This in turn will impact the nature of the procurement profession. We believe this will make a career in the discipline more exciting and intellectually challenging than ever.
GenAI brings efficiency and insight, and most of all speed to procurement processes. But it must be deployed with deliberate ethical guardrails, and it must be used in conjunction with human judgment, experience, and tacit knowledge.
