AI in procurement – The machine/human paradox

The leaders of the procurement world are up against increasing pressure to, and indeed have the appetite to, “drive significant and unprecedented levels of value” — so says the results of the latest global CPO survey, performed by Deloitte and with expert input from Spend Matters chief research officer and procurement technology authority Pierre Mitchell. In fact, more than 80% of CPOs, it states, indicated that digital transformation is a top priority, with a specific focus on GenAI, because “… AI tools can help procurement professionals identify trends and patterns that are otherwise impossible to tease out, thereby enabling more informed decisions from supplier selection to contract negotiation.”

So where and how is this great leveller of procurement effectiveness being used in the real world (the practical) and how is it viewed by the CPOs and heads of procurement and supply chain who are on the front line fighting the daily efficiency battle.

Spend Matters is gauging the level of interest in, and infiltration of, AI and GenAI in the practical lives of the procurement function today with some real-life use cases and opinions from senior procurement and supply chain practitioners. Not only will this steer our research in more industry-relevant ways but will bring genuine first-hand insights to like-minded procurement professionals.

We spoke recently with David Schultz, VP, Chief Supply Chain Executive at Westfall Technik Inc. Westfall Technik is a portfolio company of Blackstone; having $1 trillion of assets under management across 2000 companies makes it the largest private equity firm in the world. David has been working closely with them as they embark upon AI, a key focus of the Senior Leadership team at Blackstone. In fact, Blackstone has for many years been pioneering research into what AI can do for business. In support of this vision, Steven Schwarzman, the Chairman and CEO of Blackstone has donated approximately $500 million to AI education and research, including a landmark $350 million gift to the Massachusetts Institute of Technology (MIT) in 2018 and a substantial grant to Oxford.

AI use cases are intriguing – but you need to educate yourself

David doesn’t proclaim to be an expert in AI. But he does believe it’s crucial to become educated regarding the “misnomers, mistruths and mysteries around AI” as he puts it.

“It feels irresponsible for someone driving transformation not to become educated,” he told us. “It’s almost like driving a vehicle on the highway at night with your headlights off and hoping to find the right exit.”

Feeling strongly that the procurement leader needs to understand to a significant degree how AI is reshaping the world of business, in order that they evaluate its importance for their own organization, he recently completed an intensive course on AI in a strategic business environment conducted by the Computer Science and Artificial Intelligence Laboratory at MIT.

His reasoning is that while AI has been around for a tremendous amount of time, albeit in different wrappers, computing power has caught up with it and ChatGPT has put it more in the hands of the everyday practitioner. “While AI won’t necessarily replace humans,” he says, “jobs will change, so we need to upskill and retrain. It may be more likely that humans with AI will replace humans without AI.”

At Spend Matters, chief research officer and managing director, Pierre Mitchell, echoes that:

“For procurement, Gen AI, and the technology around it, is basically a supply market like any other where the technology is used in concert with other supply markets. Since Gen AI is so prolific and being infused into basically every category and into every business process, you had better get savvy with it.

“We need to understand the good, the bad and the ugly of AI to best know how to tap the power of that supply market and how best to apply it. Whether your GenAI comes from a third-party vendor, or your own IT has rolled it into your processes, getting educated so you can make an informed decision on how it can best benefit your organization is very smart.”

This era of AI is still human-led — Isn’t it?

There remains scepticism and doubt around AI and who’s in charge.

“I believe one of the biggest problems in business today,” says David, “is opinion masquerading as fact. So, to have an opinion on AI, you must first know the facts. Hence the training. And in my ‘informed’ opinion, the era of AI that we find ourselves in now is largely human-led — but with machine collaboration. So we can think of the human as the co-pilot, because in many of the sophisticated use cases we can’t let it operate on its own.”

To dissect that statement further: what part of AI is human-led? Our analyst points to the creation of the large language models, the content that is being created and the selection of the people who are training it.

“That little genius we call ChatGPT was taught by someone to get as smart as it is,” says Pierre.

One very real-life concern held by the large tech vendors is how the human makes sure the model is used for the right purposes. The tech developer builds solutions to offer them up to the solution provider. The human must ensure the model can access an entire world of data and bring that intelligence to the user in a way that serves the purpose and the persona of what the end user is trying to do.

“The key to that is building knowledge models that connect language models to data models,” Pierre continues. “The more the tech vendor can be very specific about context, persona, language type and goal, the better the LLM will be able to help with an optimal commercial decision that you can bake into your business steps. It’s the human’s job to construct the LLM in a very precise way that will yield better relevance, better search results and better security. So humans definitely have a key role to play — for now!”

Generative AI use in procurement

From a practitioner’s viewpoint, David still doesn’t see a tremendous amount of GenAI happening in procurement. “However, there is a lot of algorithmic and machine learning use taking place. Consider spend categorization or classification,” he says, “AI will take you so far using the information it has, but a human needs to be involved to categorize certain parts, because sometimes the ML might interpret something incorrectly and classify wrongly. So while there are many things the machines can do, they can’t always be relied upon and be let alone, and certainly can’t drive organizational change or perform the nuances related to strategic sourcing just yet. That takes leaders.

“And there is still misunderstanding over what some solution vendors call GenAI in procurement. Terminology gets used interchangeably, and in many cases, what the user thinks is AI is actually just algorithms and machine learning or an API to ChatGPT. It can be hard for the user to know the difference.

“On top of that, you have to be careful with Gen AI,” David continues. “You have to verify the source of any claim or evidence, especially in the legal world. So you need expertise in how you prompt the large language model, how you pose the question, what you ask it, and it’s not just about how you get the answer you’re looking for, but how you minimize or eliminate the LLM from hallucinating. So some companies are moving towards a closed model of LLM, one that is fed only with information they trust. It’s true that all the other information out there on the web can be very powerful, but it can also be polluted.”

“At the end of the day,” adds Pierre, “the LLM is a text-prediction engine, you will never totally get away from hallucination. It is trained on a pool of data and is basically predicting from all the deep connections, pattern matching, indexing and the neural-network-based activity to give you what you want out of it. There will always be some level of guesswork as to what the user really wants. And the most impactful thing that you can do, besides scrutinizing the results, is to make sure that the corpus of knowledge that it is drawing from is relevant.”

A great use case for AI

Having said that awareness, training and understanding the true nature of AI in procurement is important, David believes it is a business imperative to embrace it. One of the areas his team has begun deployment is incorporating the use of AI autonomous negotiation bots to drive improved working capital. Able to carry out thousands of negotiations at once, they are training it to learn supplier negotiation styles and use prompts that will result in driving value for all parties involved.

“You can’t efficiently scale humans to do similar tasks,” he says, “but it’s a great use case for AI.”

“We are starting with tail spend, that means that if a mistake were to happen, it wouldn’t be company threatening. It also gives us a chance to understand the tech better and where humans do need to be in the loop. It can help tremendously with the whole build-up to the negotiation, give you the information you need and the ability to slice and dice and rack and stack it in diverse ways. That gives you more, better and quicker insight. There is also evidence that many suppliers prefer to negotiate with conversational AI; they feel more engaged because AI can converse with them far more frequently, which can lead to more opportunities. Keep in mind, that many of these suppliers do not have any regular communication with their customers, so even this new and novel type of engagement is widely embraced.”

“Training your LLM on tail spend is definitely a good first initiative,” says our analyst, “it has high volume of events and low complexity. It gives you lots of data points and feedback opportunity. And that’s important, repeated feedback on ease of use, how well it works, speed, etc. means you can learn from users at scale then build AI interactions around those learnings. Once you have more variables and more context, you can start to build it out. It’s a good way to create mass adoption.”

“So this is one area of AI use that we are exploring to improve productivity and to ensure we’re being competitive on more than just our strategic spend,” says David.

“I do believe however that humans and their ability to empathize will continue to be essential in a relationship environment. I don’t see AI replacing very strategic negotiations just yet – but, as we say with technology, it’s very difficult to see around the corner.”

Spend Matters Insider can help you learn how vendors use AI in their solutions.

Advice to other practitioners

“One thing I have learned,” says David, “is that you can’t do it alone.”

Some firms opt for third-party support, or joint development. Others, as large as Blackstone for instance, have a multitude of skills, ideas and capabilities within their own four walls. “A lot of the time,” says David, “the real asset is the data. And when you are lucky enough to have that from hundreds of companies, you are already ahead of the game. So our approach is it to take the lead from what certain technology providers can offer, then layer that with our own subject-matter expertise.”

For those still wary of implementing AI, David offers the following:

• It’s ok to jump in – just use common sense, trust your instincts and experience. The actual cost of doing nothing may be quite substantial.
• Don’t be afraid to try it out – it’s ok to make mistakes, just not crucial ones – start small.
• Don’t assume that because we’ve all learned a lot about supply chains over the past three years and we’ve read a lot about ChatGPT that we fully understand it. Get educated as much as possible, be cautious and mitigate the major risks, but we mustn’t let that paralyze us from taking responsible action.
• Stretch your capabilities and knowledge by leveraging your suppliers – both the suppliers you currently buy from and your tech providers, so choose the right ones for your business.
• Look outside your own procurement walls. Bring in the sales and operations teams, bring in your commercial team, and don’t stop there, because you’re at the nexus of a very important part of a change that’s going to affect your entire business.
• And don’t forget about change management, it’s the key to rapid adoption and ultimate success. The technology will fail if the humans don’t embrace it.
• All change initiatives begin with creating a shared need, shaping your vision, and mobilizing commitment. This remains true even with the best technology at your side.

“I think AI is the fastest tech acceleration we’ve ever seen in our lifetime,” he says. “While it doesn’t mean you have to absorb everything, there are certain tenants of it that you should at least try to start deploying in your business, otherwise you will be left behind. I equate the deployment timing of AI to ‘dog years’. Every year is similar to 7 years in the normal course of business. Thus, if a company tells you that what you need is on their roadmap in the next 6 months, this could be the equivalent of 3-4 years in another segment!

“And as Pierre explained, AI is a supply market. In my experience navigating the web of tech providers can be a headache – there are so many companies doing great things with AI, but how do you know where to find them? Where do you start? What are all their capabilities and functionalities? How do you know which ones would make good partners for your business? So get smart, do the research, be the person who makes this happen, not the one who watches or wonders what did happen.”

Spend Matters ‘Meet AI’ series is a good place to find out how AI is being used in procurement solutions, free on basic registration.