Back to Hub

The difference between traditional AI and GenAI in S2P

01/23/2024 By

Adobe Stock

This article was derived from an interview between Spend Matters analysts Pierre Mitchell and Bertrand Maltaverne based on research conducted in partnership with SAP.

AI existed for a long time before generative AI (GenAI) emerged. It enabled organizations to tackle more complex scenarios, and its current use cases are making processes much more impactful while augmenting human knowledge and capability. It basically multiplies efficiency through higher process automation. The traditional use cases of AI in procurement include: automated sourcing events for non-strategic, lower-risk events, including suggestion of award scenarios and choice of supplier; optimization of sourcing scenarios; supplier information and discovery management; pipeline automation, classification, cleansing and normalization in analytics; contracts for NLP-based extraction of key commercial information and suggestion of alternative approved clause/language, and determining the appropriate procurement method in P2P.

Recent GenAI innovations, however, have taken this a step further, opening the door to new or enhanced possibilities. GenAI is moving traditional, broader AI use cases into a realm of very particular and case-specific uses, which still come with their own challenges, risks and considerations.

Danny Sack, Director of Product Marketing at SAP, says “At SAP (and in the market in general), we’re always thinking about how AI can decrease points of friction. In SAP® Ariba® Sourcing, for example, there are AI-driven tools that enable practitioners to automate the RFx process. Automation will aid in the selection of suppliers and deciding questions to ask, saving everyone valuable time.”

The distinction between traditional or mundane AI and GenAI remains blurry in the procurement industry as GenAI takes on a life of its own. So it’s important to see GenAI as part of the big AI family picture, but with different techniques, tools and so on. How we distinguish it is by looking at how GenAI builds on traditional large language models (LLMs) that are already adept at ingesting and ‘understanding’ large data sets by uncovering deep data patterns to help predict the outcome the user really needs.

“As we think about AI now and AI in the future at SAP,” says Sack, “there will be a blend of mundane AI and generative AI, integrating established, efficient processes with the transformative power of more cutting-edge AI abilities. The tools will evolve as we closely collaborate with our customers — our customers already require certain advanced AI capabilities to stay ahead in today’s dynamic landscape. We see this with the incorporation of a LLM into SAP Ariba Category Management and the increasing role of Joule across our portfolio.”

It’s also worth remembering that some elements of GenAI have seen use in procurement and supply chain applications for years. These have included conversational user interfaces, such as chatbots, and in research and analysis, including extracting, summarizing and rephrasing information, such as in category management, opportunity detection, supplier information and discovery and benchmarking. New uses build on those capabilities: SAP’s relatively new Category Management solution, for example, uses GenAI to prepopulate market dynamics and creates a starting point for users to visualize the competitive forces that may shape their industry and categories.

So traditional AI and GenAI can complement each other. To help shape our understanding of the role of AI versus that of GenAI in procurement, our analysts discussed the following four topics.

To gain more insights about the use of AI in procurement and context for our analysts’ discussion, please visit our dedicated AI in Procurement page.

How ‘mundane’ or traditional AI helps support the S2P process

What we call traditional AI mainly concerns itself with crunching massive amounts of numbers and data, specifically excelling at processes that do not require generative functionalities. It’s very important in S2P because of the multiple elements, parameters and trade-offs that inform almost every procurement decision. The data quickly becomes simultaneously complex and dynamic, putting strain on a person’s capabilities.

AI models can augment the S2P process, freeing procurement practitioners from mundane tasks so they can focus on higher-value activity. They can, for example, facilitate the auto-classification of spend data in real time so higher quality data feeds the team’s spend analytics or helps with guided procurement processes. This is not an activity that adds a lot of value, nor is it not the kind of activity procurement professionals particularly want to spend time on.

Procurement does not experience this ‘freeing’ alone; procurement’s stakeholders can move beyond bureaucracy. Modeled AI can help stakeholders understand how procurement buys and how other teams should work with procurement — things that exist in our heads and archaic policy documents can be operationalized through a conversational interface. Again, offloading this important-but-rote knowledge frees us from the tyranny of the tactical so we can focus on strategy.

How traditional AI can help beyond typical S2P processes

AI offers more opportunities to get some outside-in intelligence within the S2P process. AI can pull data from different sources inside and outside of a company, which matters when you realize how much our data sets are growing and with them the elements that we must consider when making a decision. Procurement needs to be able to see patterns within the data, the signals through the noise, and where there are outliers and trends worth paying attention to. After all, if done properly, procurement is an outside-in process, specifically as our understanding of supplier market intelligence, supply risk and what is happening in the markets informs how we find different suppliers, how we engage with our supply base and how we design our extended supply networks.

Having a machine that can learn from an extended network of information, which includes people working in the marketplace or users of a SaaS provider’s install base, and train itself with this data leads to a scaling effect in which the tool can be used for these types of analytics. In fact, if you look at GenAI, that’s what it’s doing with LLMs. These models, however, do not train themselves entirely. There should always be humans behind the scenes to ensure the content is appropriate, de-biased as far as possible and the predictions and prescriptions the tool makes are accurate.

How traditional AI and generative AI can complement each other in procurement software

Generative AI and mundane AI complement each other and the broader enterprise software systems within which they operate. Alone, AI does drive value, but the most value comes when it engages with the context of a business’s processes.

Take contract management, for example. LLMs with GenAI can understand the obligations, risks and commitments within the legalese. The best results, however, come when an AI can address your specific context. If you wanted to author a contract, you could train an AI engine to pick the closest template as a starting point. You would be better served, though, by one that understands the patterns of different clauses and asks for more information on the product or services you want to buy. After filling out a questionnaire, the AI would have the information it needs to recommend.

The other context of a business’s broader processes is that the end users are human and will remain human. What differentiates GenAI is how it connects with the ways we communicate by using outputs that are visual or human-styled text. That speaks to us, unlike the numbers or awkward text outputs of traditional AI. So, the storytelling and communication capabilities of GenAI can work very well as a complement to the analytical capabilities of traditional AI to draw out the data hidden within business systems

Where AI has, and will have, the greatest impact on S2P modules

AI can bring capabilities and solutions that go beyond base rule automation and even smart rule automation. We see what is called ‘copilot’ a lot, but a copilot is still someone or something supporting you, making recommendations. The tech has the potential, however, to recreate agents, which are like a virtual worker or colleague that acts with more autonomy than traditional AI. Agents can have a great impact in S2P, especially in processes heavy with knowledge work. In category management, for example, you need to have a simultaneous grasp of outside and inside elements to recommend potential strategies.

A second, more directed impact is predictive analytics. Like LLMs that predict text, predictive analytics recommend solutions to emerging problems. Unlike LLMs, these will not be general-purpose. Multiple models need training with multiple domains, intents and personas to give the AI the correct context for decision making. Hooking up these more focused AI engines is necessary as you move from the more mundane areas to the more critical and strategic ones.

The key to doing this is really to start small, to develop a kind of ‘long-term quick-fix’ that delivers some business value. Then, working with your external providers, which include technology providers, services providers, consultants and other specialists to grow that small tactical adoption into a broader strategic transformation.

What’s next?

The recent rise of GenAI technology use in a business context has overshadowed the large body of other non-generative AI and machine learning tools that also help build up a collective set of purpose-built analytical capabilities. However, both traditional AI and GenAI tools have shaped, and will continue to shape, the effectiveness and impact of the S2P process. But it’s important to understand why both technologies are needed for broader S2P and supply-side automation. “The most important thing to think about is that it will be critical to start this journey for your company,” says Sack. “Even if it’s just a small test, you should start learning how your org can best leverage AI in the future. Your solution providers will be able to provide guidance and get you the best path for your company. The most important thing is to take that first step.”

You can do that by reading: An Executive Guide to Evaluating AI (beyond just Generative AI): Use Cases and Adoption Strategies.’

AI - Artificial Intelligence