Unlocking Gen AI Potential in Procurement – Transcript

Jeff Gilkerson:

For a long time there’s been use of some level of AI in transaction processing, trying to classify data – all the rules based off what we’re talking about. What’s different, and I think frankly exciting about procurement now is Gen AI allows you to move into more strategic activities.

Announcer:

Welcome to The Hackett Group’s “Business Excelleration^™ Podcast.” Week after week, you’ll hear from top experts on how to achieve Digital World Class performance.

Vin Kumar:

Welcome to our new episode in our podcast series, brought to you by The Hackett Group. I’m your host Vin Kumar. I lead our Gen AI Advisory program. And with me I have two distinguished guests today – Tim Yoo and Jeff Gilkerson. They lead our Procurement Supply Chain Transformation practice. And the topic today is to see and discuss and hear perspectives on how Gen AI is impacting the procurement organization.

So welcome, Tim and Jeff.

Tim Yoo:

Thanks, Vin. Hi, I’m Tim Yoo. I’m a principal. I lead our Procurement Supply Chain Operations Improvement Consulting practice. I’ve been called a lot of things but distinguished very rarely, so I appreciate the introduction. And Jeff is part of my group – leads our Procurement Transformation practice.

Jeff Gilkerson:

Good to be here and excited to talk about it, and also be called distinguished in the same thread as Tim. Very exciting.

Vin Kumar:

I’ll start probably with you, Tim. And give us the big picture on what you are talking to, I know you talk to multiple solution providers, you talk to the CPOs of large organizations. What’s their reaction or what are they thinking about Gen AI – if you can give us a big picture.

Tim Yu:

I’d say that everyone’s interested in the topic. It’s the hot buzzword. Everyone’s interested in learning more and trying to understand what they should be doing – where they should be investing and/or learning about it. Not many people though understand what they need in terms of to be ready for it. Are there things you can do to prepare? We do have clients though that have started their investments, have picked specific use cases and are piloting next-gen AI.

We’re spending a lot of time talking about it because everyone’s interested – it’s the hot buzzword right now. There are scenarios and clients where their board is pushing them to make investments and to make sure that it’s incorporated into all aspects of their business. But I’d say relative to other functions in a corporate world, the procurement tends to be a bit of a laggard when it comes to the cutting edge of technology. And that’s kind of my viewpoint, but they’re starting to catch up particularly in this area.

Jeff Gilkerson:

There’s definitely a lot of activity, Vin. Tim, I know the other thing – one of the things we also often talk with a lot of our clients about – is what are their current providers going to be doing in AI? And I think historically they’ve been a little bit behind. When you actually look at the market, they have a landscape slide that has an ever-growing number of solutions that are out there. And a lot of the big players were not leaders in that, and so these providers filled that space. But what we’re seeing is they’re starting to catch up now. They’re starting to release new features. And so it’s definitely a fluid situation, as organizations are looking at it and trying to figure out where to invest and when.

Tim Yoo:

Well, Vin, I always comment that procurement has used some sort of AI for a very long time as it relates to data cleansing. So in your mind, what is generative AI? And what’s the potential impact for procurement?

Vin Kumar:

I get this question often is, “Hey, AI has always been there, so what’s new about generative AI” and stuff? And what we explain is there’s been a fundamental shift in what the capability of AI is. So I define AI as the machine trying to emulate human reasoning – that’s what it’s trying to do. And the AI prior to Gen AI – what the industry is now calling predictive AI – was basically emulating our left brain activities, doing very much. We are giving the machine a set of instructions – it’s maybe very complex instruction – but the machine would go and do it very fast. And it’s able to crunch a lot of data, a lot of variables, and support us.

And where this generative AI has now come in with this new capability is more what you and I would call gut experience. We have seen so many clients. We have seen a lot. We’ve seen our own experiences and maybe some biases that we have. And now we have to make decisions, and we take these gut-based decisions. Mathematically, they call it probabilistic decisions, and that’s kind of where Gen AI is really good at taking these probabilistic kind of decisions and using unstructured data.

One of the other things of prior predictive AI used to be is we had to spend a lot of time cleansing data, getting the data in a very structured manner, and that was sometimes a Herculean task for a lot of organizations to get there. But with Gen AI, it really likes unstructured. If it’s structured, it doesn’t do really well. It likes unstructured. It may be able to really use that and help with decision-making, utilizing and creating content based on that.

Tim Yoo:

Structured and rule-based – that’s what it used to be. I remember the early days of spin analytics tools, where we would do these analytics projects for clients to clean their data. And we’d use this technology that we said had this artificial intelligence. We called our AI machine “Raj” because we would send the data to our offshore center in India and a team.

Jeff Gilkerson:

I’m pretty sure this was my first project with Tim, by the way, was me on a spreadsheet trying to create rules to update the data.

Tim Yoo:

Yeah. And then we would try to learn from that each subsequent project, but it was definitely a lot of blocking and tackling.

Vin Kumar:

And I think that’s where I think this is really interesting, and I’ve got actually one of our clients who’s trying to use it for data cleansing. They’re using a Gen AI solution to help cleanse the data, and they’re able to… So it’s very good at converting this unstructured data into structured data, and then you have your regular rules engines and predictive AI tools and all of them to use that.

One question for you Jeff, I know Tim said that some clients are looking at Gen AI and already their use cases. As you interact with your clients, are you seeing anyone using it? And how are they using it, even maybe in a pilot or a POC mode?

Jeff Gilkerson:

Yeah. I think there’s a couple of areas that organizations are starting to use it. And it’s kind of interesting because, as Tim mentioned, there’s always been, or not always, but for a long time there’s been use of some level of AI in transaction processing, trying to classify data – all the rules-based stuff we were talking about.

What’s different, and I think frankly exciting, about procurement now is Gen AI allows you to move into more strategic activities. So analyzing an RFP response, creating an RFP, loading a contract, and pulling the things out of it that are going to be interesting and populating them, creating summaries of those contracts. So you can create information now that used to require someone, the gut feeling and that experience, and being able to know, “Hey, this is going to be important or not,” and it allows you to do that.

So I’d say the area that’s probably the most prevalent right now is I’ve talked to a lot of organizations and worked with that are building applications around contracts. So contract data is notoriously all over the place. Even if you can get it all in one place, which that’s a big if, but let’s pretend even for a minute that we have it in a repository, how do I get the data out of the contract into a structured format that I can then use for basic things like when is the contract expiring, what is payment terms, what are the tiers for any rebates, etc. – let alone any more advanced like clauses and do we have risk, etc.?

But what you’re seeing is now the technology that with the generative AI and the unstructured, it can go in, read that and basically figure out, “Hey, here are the 20 things that are important,” and populate them for you. Now, someone probably has to review that the first couple of times, but it very quickly learns and it builds onto the long language model that they have.

Tim Yoo:

I have a client who’s using it on the contract side to do contract summaries. They’re not using it to try to author a contract but for existing contracts. They’re using it to quickly pull up a two-paragraph summary in the key deal terms. Whether it’s part of a negotiation strategy or funding requirement that used to historically take them hours and require maybe even a little bit of expertise, and they’re using it quite well in that space.

Vin Kumar:

There’s one I know is trying to do, they went through this RFP process collected from multiple… I think this is a company, MTN in South Africa, where they were getting, I think, almost the whole RFP, the quantity was significant, and they used this to summarize it into a dashboard. So it helped them make decisions on what to do.

Is there expectations from the chief procurement officers that this tool is going to make them more efficient or become more effective? Is there a dimension of that they see or they’re expecting?

Tim Yoo:

All the research out there is saying that, and it’s going to transform the way we do business. So I guess that’s my question for you, I mean, I see things like where there’s major head count reduction and efficiency through next-gen AI and procurement.

My personal take is I don’t think it’s going to be that huge because a lot of procurement departments are spending a significant portion of time on transactional nonvalue-add activity. So I think it’s just going to allow them to be more strategic. I don’t necessarily think it’s going to change the head count significantly, but I guess that’s my question to you as well.

Vin Kumar:

One of the things that we are seeing in our research, and we had published this, to say, OK, what the impact is going to be over a five- to seven-year period. So it’s not immediate, but we do see from our research that it’s going to impact significantly taking out that transactional activity that the human was still there. You’re trying to automate using all sorts of solutions, but there was still a human in the loop trying to do this fuzzy logic and kind of stuff, which is going to get replaced by Gen AI. But that is going to happen in procurement we see in the next three to seven years. The first three years it’s more on the effectiveness sides. But the efficiency, we will actually see that the process is going to get significantly more automated.

But majority of the capability of the Gen AI solutions is not that they’re going to be building their own models – anything like that. It’s going to come with these embedded enterprise solutions that they use in procurement will provide them capabilities. And there may be some domain-specific models that they will subscribe to, especially maybe risk management, or there may be some, like we’ve got a client – a company – actually developing a model for sales and use tax in the U.S.

It’s only doing that, and it’ll do the entire process. It’ll determine the product and see, OK, what’s the best way of categorizing the product to minimize the sales and use tax, and apply it and goes directly in the system. You don’t see it, but it’ll be integrated into your process that in the AP department you don’t need somebody sitting and doing sales and use tax anymore.

Tim Yoo:

Now there are sales and use tax engines out there, but they’re rule-based. It’s based on the end user deciding what commodity code, what state, even sometimes deciding if it’s taxable or nontaxable.

Vin Kumar:

That fuzzy logic, which is that person sitting and doing, now is being replaced. And they have training it in that sales and use tax domain. So it’s not using, quote unquote, “a ChatGPT,” which is a general-purpose model. They’re not using that. They’re subscribing to this, which is domain specialized in the sales and use tax.

Tim Yoo:

I could see the same applying to supplier onboarding for risk review because it does require the end user to answer some questions to create the roles to determine what channel and how deep we’re going to review the supplier for third-party risk.

Jeff Gilkerson:

I think to your earlier question, Tim, I think it’ll actually be more interesting on the more strategic activities. I think the Gen AI is going to allow you to be more effective and efficient, and I think organizations are going to choose different paths for, “Well, what do we do with that? Do we reduce head count because we’ve become more efficient?” Or do we say, “Actually because of the technology, we’re also able to be more effective, and so we can actually add value on more of the spend than we had before.”

So historically, we might have only been able to build a category strategy for our top 20% of our spend or suppliers because it’s time-intensive, it requires a level of skill and experience, and it’s expensive to do that. But if I get more efficient at it and I’m more effective, do I keep those resources and now expand, and say, “I’m going to actually cover more of my spend and add more value?” Or do I keep doing the same scope and reduce the number of people?

So I think different organizations are going to look at that differently. And I think it’ll be more interesting on the strategic side because the calculus is not the same as on the transactional where, yeah, I take out a head count and I’m probably not going to reallocate that resource.

Tim Yoo:

I understand that. But Vin, do you really think that’s three to five years away? I mean, I’m still waiting for my flying car, so.

Vin Kumar:

So what we are saying is, “Hey, the next seven years, it’s going to take RTs about impact procurement over…” I think our research says something like close to 46% impact it’s going to have on the cost of procurement as an organization to do it.

But what we see is there is more effectiveness play will start coming in initially, and then the efficiency play will come in a little later when these solutions are embedded. Because it takes some effort to start building your own solutions, if you are trying to build your own Gen AI solutions it takes a lot of effort. And it’s not as localized as what the bots were, but it’s not as command and control of putting a procurement system in place. It’s between.

So it does take effort. But when we see these domain-specialized and what we call embedded capability coming in these enterprise solutions, that’s going to give them the efficiency play. The focus is all going to be then for the companies to build on what they think is core to them and that will be more strategic, which they may have to build their own solutions to look at it.

Tim Yoo:

So the embedded solutions procurement. And that I see. I was on actually a webinar yesterday with the chief procurement officer at Coupa. But he brought up an interesting statistic, that their transaction volume is $13 trillion of all their customer spend. That’s a massive amount of data. And if they’re able to embed next-gen AI to somehow intelligently use that data more efficiently to the benefit of all their customers, I could see that being something truly transformative.

Vin Kumar:

One of the challenges actually, that brings up a great point, and that’s something which when we talk from an advisory perspective to some of the CPOs, we are letting them know is companies use a lot of SaaS solutions across multiple functions. So what do you want to do with the data that the SaaS provider is going to use? Are you going to allow them to use it to build their own models to train? Or do you say no, or you want to get compensated for that – what it is.

And that’s something important that we feel the procurement leaders need to be having with their functional owners of the SaaS solutions to do it because that’s extremely critical.

One of the things we thought initially last year was some of the SaaS providers are going to jump on it and start using that, but they were getting a lot of pushback from their clients that they can’t use their data to train a model. A lot of it is apprehensive. They don’t know how it’s going to use what gets up. So there is a little bit of that fear right now. But as the market stabilizes, matures, there’ll be people to say, “OK, we can anonymize, aggregate our data and sell it to a Coupa in this case,” and they can use it how they want and provide some insights to the suppliers to do that.

Jeff Gilkerson:

Yeah. I think the data’s a big concern and definitely an area that we see a lot of … it’s top of concern for a lot of organizations that I talk to on a daily basis. Is even just understanding what is the data behind because there’s a lot of tools out there, but understanding what is the data that this model has been trained on? Because I think there’s still maybe some smoke and mirrors out there in the industry around rebranding of things that have existed for a long time as AI because it’s popular now, but it’s nothing new.

Tim Yoo:

And with increased concerns around data privacy, you’ve got to be careful what data is being made available as well. With all of the cybersecurity risks and data breaches, I kind of assume that there’s going to be even more diligence in that area.

Vin Kumar:

Yeah. So one of the challenges with Gen AI especially is there’s specific regulation coming on how you use this technology. For example, if you’re hiring, is there a bias in the model? So there are going to be regulatory governance come to check if you’re using a model to make sure there are no biases. Same thing if you’re looking at customer data.  So there will be more regulatory governance around using this technology. It could almost be if you build your own models, you’ll have third parties to come and audit your models to make sure that they are in compliance with the regulation.

There is regulation just released a month and a half ago from EU – it’s called the EU AI Act. In the U.S., we’ve got the NIST, have got a risk management, a regulatory framework. There is a governance from the executive order from POTUS, which came out last year to do that. So there’s more regulation still in the works that’s coming, so that’s a big concern.

Tim Yoo:

Do you think that’s going to stunt the market – the increased regulation?

Vin Kumar:

This is not going to stump the market, but it’s going to… There’s an assumption in the market that every time you’re looking at Gen AI, it’s a model. And what we are trying to educate is you’re going to build solutions off a general-purpose model. So you don’t have to worry about auditing that model because the model providers – the ChatGPTs, or Microsoft, or OpenAI services, or IBM – they are going to get their models validated, regulated, and checked. And you are going to build solutions using that.

Now, the question comes, if you build your own model, which most probably in SG&A we won’t do it. It’ll be more on your revenue side, a product, or R&D side. Those models will get regulated, and there will be a governance process that you’ll have to comply with.

Jeff Gilkerson:

It’s interesting, Vin, though, I’ve thought about this with the AI in contracting because it’s one of the… as we talk about regulatory and compliance and stuff.  But I do wonder what happens the first time a contract that AI supported, generated, created part of, there becomes a dispute, and you get into what’s the legal ramifications of that if it wasn’t a person who… does that change responsibilities? How does that impact it? Does it introduce risk into third parties that help generate it? I think there’s a lot of ways that that could go.

Vin Kumar:

That’s a good question, Jeff. And it does come up in our conversations, and I tell clients is, “How do you handle it today? If it was a human. who did that error? What happens?” That’s kind of what you will have to have in place. Is that OK? If you do, your solution made an error and it’s a dispute. You have the same liability that if you had one of your subject matter experts – your procurement staff – did that error. So it’s the same kind of liability exposure you have to do it.

The one I think from external regulators is they want to ensure there’s no systemic biases – systemic issues embedded in your solutions. And that’s what they are worried about. It’s not the one-off. It’ll be treated exactly as if an error was done by a human.

Tim Yoo:

At some point, are governments going to worry about open access to this information?

Vin Kumar:

So that’s one of the things we advise clients is that’s why you’re not using public Gen AI models. You’re not using an OpenAI, ChatGPT, or Gemini, or Llama, or Titan, but you’re going to go through these enterprise Gen AI applications – Microsoft Azure OpenAI Services, or IBM watsonx.ai, or AWS’s Bedrock solution, or C3.ai, or Hugging Face. These are all platforms that protect you and you’re using that, so your data is protected. Your IP is protected. You are also limited liability of exposure to any copyright violations that these providers and enterprise applications provide. So that’s why we highly recommend that you use these rather than using the public domain ones. And there’ll be some companies with certain size of the companies may use the public domain, but you’re exposing yourself to those risks.

Tim Yoo:

We’re doing some interesting work with clients now that are trying to prepare themselves – readiness for next-gen AI – because they have challenges in actually getting access to their data.  So in the procurement landscape if you have point solutions, so you’ve got different solutions for different processes, and many of them aren’t integrated. And many of them, they’re not actually even capturing the right data – they’re just using it for manual processes.

So we’re starting to see the advent of these orchestration technologies, sort of that layer across these point solutions as well as your financial system.

Originally, it was for the user experience, so there’s a single point of entry. It can intelligently understand what you’re requesting, to then know that you’re looking for a contract versus a requisition versus a supplier onboarding. But they’re also powerful in that they’re capturing data. So if you’re designing these systems correctly and you’re capturing, and thinking intelligently about what data you’re capturing in this orchestration layer, it can become very powerful when you start becoming ready to plug into next-gen AI.

Vin Kumar:

That’s a great use case there to share with clients of how you can use that data, which is being captured at the orchestration layer, and use Gen AI to get insights for you.

Tim Yoo:         

Right. And then the next thing that we’re starting to see is more need for skills and training. As things are changing, procurement training really hasn’t caught up.

Jeff Gilkerson:

I was just thinking about the number of clients we have that we’re doing basic blocking skills development competency assessments where there’s gaps in the basic capabilities. And I think Gen AI can help with some of that, but it’s also going to open up this whole other set of skills that the organization needs. And procurement hasn’t historically been the best at developing those and being on the cutting edge of that.

Tim Yoo:

Well, to Vin’s earlier point, we’re going to be asking procurement to be more strategic and be doing more of these strategic activities and more modeling that traditionally in procurement you’ve got people that do that, but you’ve got a lot that aren’t spending a significant portion of their time on those areas.

Vin Kumar:

Hey, Tim and Jeff, it looks like we are running out of time. I want to thank you both for sharing your insights on what you’re seeing in the marketplace, talking to solution vendors and clients. Really appreciate you taking time and having this conversation with us. Hope all of you enjoyed the podcast.

To our listeners, don’t forget to go to our website. Hope you enjoyed this. And thank you, Tim and Jeff.

Tim Yoo:

Thank you, Vin.

Jeff Gilkerson:

Thank you, Vin.

Announcer:

Thanks for listening. If you liked this episode, please share it. You can find show notes, transcripts and related research at www.thehackettgroup.com/podcast. Subscribe on your favorite listening app so you never miss an episode. We welcome your feedback by rating this or any episode, or send us an email at podcast@thehackettgroup.com. The Hackett Group is a strategic consulting and executive advisory firm. Learn how we can architect your digital transformation journey, including Gen AI, at www.thehackettgroup.com.