A practitioner’s view of AI in procurement – Are we there yet?

Spend Matters is researching the personal viewpoints and the practical application of AI among leaders in the procurement profession. We are welcoming both opinion and real-life use cases from senior procurement and supply practitioners to bring genuine first-hand insights to like-minded procurement professionals.

If you would like to join the dialog please reach out to us or submit your points of view or observations here.

We recently spoke with a leader of strategic global procurement transformation within a multi-national technology hardware and services company.

AI and Procurement: Are they ready for each other?

In his opinion, AI, and generative AI (GenAI) in particular, is not fully ready for the procurement world, and that the procurement world is not yet ready for it. It’s a tall statement but one that he vouches: “I believe in the provenance of data and information, and that it must be proved. Therefore, a strict process is necessary with suppliers who claim to use it in their products which insists they can declare themselves.”

GenAI is but one form of AI that can be employed, of course — and many smaller tasks or predictions can be made using statistical AI algorithms that have been in place for many years — but it does seem the case that the use of GenAI in particular is inhibited by a lack of traceability of the data source, as our senior research analyst Bertrand Maltaverne attests:

“It’s true that you may never really know the source of the data. There are different levels of GenAI applications in use by procurement solution providers, from basic/shallow to embedded (see our definition of the ‘6 flavours of AI’ here). In the basic use case, and these make up the majority of the demos we’ve seen, while you may think you are interacting with the application, it’s actually an integration with ChatGPT or similar behind the scenes, which draws from the huge corpus of publicly available data it has ingested, the source of which is obscure to the untrained user and carries risk of ‘hallucination.’ But, there are other ways and techniques (like Retrieval-Augmented Generation [RAG]) to leverage GenAI that enable a better control over the ‘corpus’ that is used as reference.

“The irony is that AI companies have spent many years developing AI and embedding mechanisms to prevent it being a ‘black box,’ working on data quality, governance principles, robustness and safety. Then a powerful tool comes along, like ChatGPT, with far-reaching use cases, and it’s the biggest black box of all.”

But uptake, which has been mammoth in the consumer world, is not equalled in the business world, particularly by Procurement. Our interviewee suggests that:

“The procurement industry itself is behind in terms of applications. Fundamentally, it remains conservative, concerned mostly with purchase orders, invoices and control. The P2P process we know today was set up about 15 years ago for one simple reason, not necessarily to make buyers’ decisions easier, but for control. When you think about P2P there are many organizations that still don’t have automated models.”

“This conservatism is especially true for the people ‘working in the trenches’” adds our analyst. “Traditionally, procurement is not the most modern organization compared to other departments. While we do see GenAI infiltrating some applications in the procurement world, it is less so than in other functions. But attitudes are changing.”

The McKinsey report ‘State of AI in 2023’ finds that the functions with the highest adoption of AI (marketing, sales, product and service development, service operations) have the highest usage of GenAI, but that supply chain management is one of the functions with the lowest adoption due, in part, to the fact that procurement and supply chain are areas where traditional AI (data analysis, correlations, etc.) applies more than GenAI.

However, the good thing about GenAI, according to our analyst, “is that you can integrate it pretty easily with your applications, especially ChatGPT. These technologies have been designed from the get go to be used not just online but also through API integration. So it’s a good thing for procurement solution providers that have not yet gone down the AI route.”

So how does the procurement practitioner perceive AI?

In our practitioner’s definition “AI is the bringing together of different pieces of information and different data points and correlating them to facilitate the decision-making process.

“Often though, software vendors claim to offer many AI-based capabilities, but fundamentally, there is no easy way of knowing whether the data they use is traceable, smart or just ‘dead’ information. And, as we said, that’s where the sticking point comes in.”

He explains:

“The procurement role is not complicated. You have a supplier onboarding process, a contract management process, a P2P process and spend data analytics. Those make up the very classic activities. What many procurement professionals really want to know is what are the possibilities of AI and what can we gain from them?

“Potentially you gain two main things:

• The first is knowledge from the market that gives you insights to make better decisions.
• The second is user experience.

“For example, if you knew nothing about procurement but were tasked with bringing a consultant on board, how would you go about doing that? In a large organization there may be 1000s of people who rarely use a procurement tool, other than occasionally to order a keyboard or get a contract through. These users don’t really care how it works, they just want to get the job done.

“So this is where the overarching solutions come into play today. They can guide you through the four main process pillars of procurement we mentioned. They take your task and walk you through it step by step. So I would say this ‘automated procurement’ is more pre-AI, in the sense that it is just facilitating the how-to playbook.”

Our analyst agrees. “Until the ‘frenzy’ of ‘doing AI’ is behind us, we are trying to educate users that ‘there is GenAI and then there is GenAI.’ So we do have to be cautious, because not everything, even if labelled as such, is GenAI.”

There is a lot of buzz around AI, inflated expectations and misrepresentations, so it is essential to decipher it. For a succinct explanation see our short paper: The difference between traditional AI and GenAI in S2P.

What, then, is an AI-based process?

“In the four staples of procurement we described,” says our practitioner, “there are very ‘static’ data points. In supplier onboarding we might want to know for example: ‘what is the status of the supplier being onboarded, has it been blocked by information security or did the supplier just not respond?’ Similarly for contract management: ‘When is the contract expiring? What are the liability clauses? Is there a termination clause? And for spend: ‘What is my spend with certain supplies? How many purchase orders have there been over the year? Were all invoices paid?’ This is one piece of information that sits somewhere that you need to extract, but it doesn’t tell a story!

“The ideal, and looking forward, is that a text string will facilitate the telling of the whole story from beginning to end: who the supplier is, the status, the contract that is in place, when it’s expiring, your spend with that supplier, whether it’s been delivered, whether privacy and security information have been checked, and so on. That’s the ultimate goal.

“So how do you go there? Each tech provider can give you a piece of that string. They have algorithms and can give you use cases. But the common buyer doesn’t know what that algorithm can deliver or how powerful it is. The common procurement leader doesn’t know, and I’d suggest the IT department often doesn’t know either. So the time is right for procurement teams to transition to data-first thinkers. Much of the knowledge we gain and the questions we answer as procurement professionals isn’t reformed into reusable, self-service knowledge. We could take a cue from the services world and start to build knowledge bases that can, in turn, become an AI data source.”

But the point is: there is an intermediate step between automated procurement and AI-led procurement.

The problem statement and business driver

Guided procurement is still quite complex, and users in general still find procurement to be slow. Quite often, while the main operations are as streamlined as possible, procurement wants to step off that traditional path.

“Firms that buy a lot of software across various countries can go through many checkpoints,” explains our practitioner. “But even when the supplier is responsive, it can typically take procurement firms 60 to 90 days to get a contract in place. Bringing a consultant in, by way of example, means getting access levels set up regionally, getting an MSA in place, a SOW, background checks done, equipment organized, and you need to get them onboarded. That can’t happen in two weeks! And although that’s not all dependent on procurement, from the user perspective, it is.

“This is a real-life business issue for many. Before they even consider an AI perspective, all the data — regardless of where it comes from across the different activities — must be brought together. One scenario is to use a data lake and export from there into a visual representation. But that is not AI. That is just correlating information from different sources and matching them to find a trend.

“Another real case lies in all the intake forms that an organization has to obtain and process. How do you have an overarching intake form that is user-friendly with technical natural language that can answer all your questions and tell you exactly where to go and what to do?

“This is where I see AI having a role. The correlation of information is not the difficult part — getting to your game plan is! The big difference between correlating data and leveraging AI to get to some kind of business-like chat exchange is the real move from traditional procurement. And it takes ‘baby steps’ to get there.”

An intermediate step

If you start with natural language processing, basically that means a computer reads text and correlates it with other text. A real-life example might be to get a SOW signed. Through a series of back-and-forth questions and answers the computer will guide the user through getting it done — like an assistant. And assistants might be either a textbox or a chatbot, which many firms have started to employ.

But one hurdle is that the procurement infrastructure within many companies operates in silos. In response, some companies opt to bring all that information into one data lake.

“Basically,” explains our practitioner, “you can bring together all the different data points by using a smart chatbot and natural language processing to merge information. (If you don’t use a data lake, you are obliged to put an integration platform in place which you will need to reconfigure for each solution. That can become very complicated and very costly.) From that lake of data you can retrieve information whenever you want. So, later on, when new technologies are ready and when the algorithm of the AI has developed enough, you can plug them into your data lake to have a source of information whose provenance you know and trust.

“For me presently, GenAI is really a toy for the future and not yet fully operational, but it’s intellectually interesting.”

The future of AI investment in procurement

Our practitioner believes that AI in procurement can become very powerful, but that at the moment, there is certainly a lot of effort needed to get payback, because Procurement has not yet invested a lot of money into it.

“Today,” he says, “consumers want things easy and fast. But Procurement has many gatekeepers, regulations and dependencies. It’s complicated. So it’s difficult to satisfy that user demand. And given that traditional procurement is about savings, getting money back on your investments or reducing costs, the industry of procurement can really only evolve if the CFOs accept value that is not financial.

“For example, if procurement makes the user’s life easier, faster, happier, that expectation is not valued as a payback calculation for finance. So how do you make an investment happen? How do you justify user experience and easiness, or better decision making? And ultimately, how do you value the prevention of risk? These are very good reasons why procurement teams need to become more data-driven and more disciplined in how they standardize knowledge capture, process, orchestration and so on.”

But, above all:

Procurement leaders must get educated

Very few companies have set up an IT organization with AI-related skills, especially with all the outsourcing that has happened over the past 10 years. So, if IT doesn’t always have the knowledge inhouse, who does? “The fact is,” says our practitioner, “someone needs to have some basic knowledge at least of natural language processing. The software provider may sell you a license, but you have to do the configuration. And that setup can cost you three times the license fee. That’s where your whole business case can fall down and cause a lot of frustrations – you need a trusted partner.

“AI is a massive, complex beast made up of many parts, and GenAI is just one of them. Other types of AI algorithms exist that are easier to train, deploy and derive value from. So it’s important that leaders learn the basics of AI to understand what is available and what can be achieved. As a leader, if you don’t get at least somewhat involved, if you don’t have an idea of how things work, you can’t build a solution that fits your organization. You could spend literally millions of dollars working off the rationale that someone else has given you, and still go in the wrong direction. Today, if the CPO stays completely hands off, they risk spending a lot of money on a lot of disappointments.”

For more guidance see An Executive Guide to Evaluating AI (beyond just Generative AI): Use Cases and Adoption Strategies.

To read our extensive material on artificial intelligence in Procurement, including our four-part ‘Autogmentation’ series, visit our dedicated ‘Investigate AI’ page.