Every procurement leader today is being asked the same question by their board: “What’s our AI strategy?” But most conversations about AI in procurement stay stuck at the buzzword level, agentic this, autonomous that, without ever addressing the unglamorous groundwork that determines whether any of it actually works.
That’s exactly what covered in one of the Beyond Procurement podcasts with Megha Singh, Director of Procurement Transformation at Micron Technology. With two decades of experience across Novartis, Walmart, and now Micron, where she has led global initiatives impacting over $2 billion in unmanaged spend, Megha has implemented enough AI projects, and watched enough of them stall, to know exactly where the real gaps are.
Below are the questions answered Megha’s answers, exactly as she spoke, because the most useful advice on AI in procurement rarely sounds like a keynote.
What does your procurement mandate at Micron look like typically?
I lead the global procurement center of excellence at Micron. That covers how we deliver excellence in global service delivery while managing the transformational AI projects and also ensuring that we are compliant with the global processes across Micron.
The focus is on how we use technology to optimize both strategic sourcing, which is upstream, and transactional buying, which is downstream, and not to forget the compliance part. So, my team works on everything from supplier onboarding, spot buys, contract lifecycle automations, spend analytics, and it’s a full-stack mandate.
You recently attended the Gartner Symposium. What were some of your key takeaways from the event, particularly around AI in procurement?
The Gartner Symposium, which was held in Florida, is considered one of the epitomes of procurement events. Let me just correct you, I did not represent Micron, but Micron was represented for implementing AI solutions, especially as part of autonomous sourcing, and it was a wonderful experience.
One thing that stood out is how fast the conversation is shifting from “what is AI” to “how do I deploy AI.” Everybody, all 3,500-plus members across industries who attended, had this one question in their mind: “how do I use AI, how do I implement it within my processes, within my functions.”
“AI isn’t magic. It’s math plus data plus processes. If you combine it together, then you create that magic.”
And I felt that many companies are skipping the hard part, they’re building clean data foundations and rethinking workflows. Especially at Micron, we’ve seen success only when we tackled these fundamental issues first.
With AI as the context here, what are some of the promising use cases you see where companies can deploy AI in procurement?
There are a lot of use cases that companies can deploy but let me talk about a couple of things they can take on right away, and I’ll give you examples from both upstream and downstream.
On the upstream side, in supplier onboarding, we use AI to validate company credentials, or flag compliance risks, saving weeks of due diligence. If you ask anybody who is doing supplier onboarding, you can see they’re wasting so much time doing that due diligence, managing risk, understanding compliance, it takes a lot of manual effort. So that’s one place to start using AI to validate credentials. We can also pilot LLM-based models to auto-draft RFPs using historical templates.
On the downstream side, especially for PO approvals, everybody in procurement knows how painful a process it is, as much as creating the requisitions has to go through a manual process, the approval process is also very manual. We can deploy an AI engine that learns from past approvals and flags anomalies, for example, if there’s a buyer who consistently overrides a policy for a particular vendor. Those kinds of instances can be built into your LLM model so the system recognizes the flags, and PO approvals happen in an automated way. So those are the two use cases I can right away think of, on top of my mind.
Is procurement teams mature enough to implement AI, or do they have the necessary knowledge to implement it at this moment in time?
I’ll give you an honest answer. Honestly, in my observation, most are still in the early phases, maybe experimenting with chatbots or using analytics dashboards. But maturity isn’t just tech. You have to build the models that can actually address whether the AI insights are trusted and actioned. Without trust, adoption will stall.
So they’re in the early stages, I think, but on the right track of thinking through, “okay, I want to implement an AI model.” But there’s a lot of clean-up, a lot of clean data that needs to be in place before they implement AI models, and I’m telling you this from my past experience with some great organizations, the struggles we faced while implementing AI tools. We implemented them but then realized the results we were getting had to be modified because the data itself wasn’t that structured.
In that case, how would you recommend a company or procurement team start using AI in their daily operations, realistically?
Start with the real problem, and the people doing the groundwork on the ground are the ones who would know what the real problem is. Anything that’s taking them a lot of time, anything causing a lot of pain, whether it’s PO approvals, invoicing issues, goods not being received on the system, supplier compliance checks, or understanding risk. Just understand where the real problem is. Don’t start with AI without understanding the real problem first.
We have a long tail-spend leakage problem, and that’s a real one. It’s very difficult for organizations to reduce tail spend, and that’s something we can actually start with. Then you can ask, can AI help me with this? Usually yes, but only after you clean up the data and define your thresholds and your workflows properly. So, looking at your organization’s current problem, and talking to your teams about where they’re spending a lot of manual effort, is where you can implement AI solutions.
What are some of the foundational gaps you’ve seen when it comes to implementing AI in procurement?
I’ll restrict it to three for now.
One is data quality.
- Is your data ready to be implemented for a solution?
- Do you have historical data?
- Is all the data content structured, or is it changing every now and then?
- How data-ready is the organization to implement the AI solution?
That’s one thing. And especially when you’re implementing solutions for vendor master data, it is a killer, because vendor master data is constantly changing, and a lot of organizations do not have a real-time update on that.
The second big problem, especially for large-scale organizations, is very fragmented processes, and the more global the organization, the more fragmented the processes. They’ll have a regional process, a global process, and a local process, and tying all that up takes a lot of time.
For example, if teams use Excel in one region, SAP in another, and there’s no one common ERP model, good luck training an AI model, because it has to tie up with Excel, with SAP, with whatever other ERP the organization has. It’s a big problem for implementation because processes are so fragmented. It’s not a discouraging thing to say, but it’s a thing to reflect on before implementing. Maybe you start slow, region-wise, go local first, then regional, then global, depending on the problem you’re facing.
And the third, very valid one, is mindset. That’s where change plays a lot of role, change implementation, change adoption. How do you change the mindset of people, of leaders, of your stakeholders and businesses? Some still have this idea that AI is a silver bullet, that they can use it and all their problems would be solved. No. The mindset has to change. It’s a toolbox, I would say, and you have to keep evaluating.
So, mindset, fragmented processes, and data quality are the three big killers.
That brings us to something related to your last point about mindset and culture. What kind of internal collaboration is necessary for an AI implementation to work across procurement touchpoints?
I would say the answer is tons of collaboration. You need tons of collaboration when you’re implementing an AI solution. IT is an obvious one, for integration, they’re the ones implementing, testing, and integrating a lot of our processes and tools. But also finance, they’re the ones who control our costs. And then there are legal implications, if you’re implementing an AI solution, you also need to account for the legal implications when you’re automating contracts, or where due diligence is required. Their involvement is key.
We create cross-functional AI squads and keep everyone more aligned. It’s very important, when you’re implementing an AI solution, that you keep your procurement, IT, legal, and compliance stakeholders involved in any solution you’re trying to implement, so that even before you’re investing in such a big deal, everybody is aligned and everybody’s perspective is taken into account. We don’t want to invest a lot into something and then find out we can’t really do it, because even though it saves time and comes at less cost eventually, currently we don’t have the scope, because it’s not legally okay, or we might have a compliance leak. So, that cross-functional collaboration is very much needed before you implement a solution.
Obviously, to implement a solution like this, you’ll need an external partner who has done this before and can help. How would you evaluate a partner, and what are some best practices you can share?
There have been a lot of cases where we’ve implemented AI solutions, I will not name the organizations, but from automating and implementing AI solutions for upstream processes to transactional-level AI solutions, we’ve done it, and I can put together maybe a five-point checklist I usually use for vendors.
- One, is the AI explainable? I want to implement an AI solution, but is my AI capability explainable to my stakeholders? Would they easily be able to understand and comprehend what exactly is needed?
- Two, do we have clean training data to train the AI model? If we don’t have clean training data, we cannot implement the solution, the solution would fail. I’ve been stressing a lot about data, and it’s very, very crucial, even before we implement and evaluate a vendor.
- Three, how would that AI solution be plugged into our ERP models? Is the solution SAP-ready, or SAP MM-ready, or Coupa-ready? How is it tying into our ERP solutions?
- Four, is it configurable? Can we implement a solution, for example a solution for autonomous sourcing, that’s configurable to your SAP Ariba, configurable to your contract management? You need to see if the solution is configurable or not.
- And last but not least, and I think every CPO would look at this, the ROI. What would be my return on investment if I implemented this solution, and not from a five-year range, mind you, because we usually say, “eventually, five years, seven years, we’ll have this much return on investment.” We need to also calculate what our ROI would be in one year. Eventually the five years would definitely look good, but is the return on investment we’re getting right away worthwhile for the effort?
So, that’s the kind of checklist one can use while evaluating vendors, and mind you, there are so many vendors who come back with proposals built exactly around these points.
When leadership wants to get started with this, what are some of the biggest misconceptions these leaders typically have in mind when it comes to AI in procurement?
You usually go to the boardrooms, present your stuff, hear their perspective, hear what they want implemented, and no offense to anybody or any leaders, but this is the expectation you usually hear: they do not understand what can go wrong. They assume AI is like flipping a switch. But it’s more like adopting a new teammate. It takes training, obviously, it takes a lot of trust to build, it needs a lot of tweaking.
So, when leaders say, “I want to implement an AI solution,” they think that once implemented, results are generated, like flipping a switch. It’s not. It’s constant change.
- How do you first implement it
- How do you ensure the processes are being adopted
- How do you ensure there’s continuous improvement in the current process
- Who would overlook all the inefficiencies that come out of process change and gaps, rather than chasing a dashboard instead of outcomes?
More or less, leaders want to look at a dashboard, “okay, these are my numbers, these are the outcomes,” but the real problem actually lags, where on the ground you see, “okay, I implemented an AI solution, I took all my kudos, this is done, my scale-up is done.” But what happens after that?
How prepared are you to handle the constant improvement, the constant challenges that come, and how are you ensuring it’s being adopted and you’re not going back to your older ways of working?
How do you balance AI or automation with human judgment, especially given the decisions procurement leaders make can impact the whole supply chain and cost?
I think usually we humans are afraid of AI. That’s a general tendency, or an unspoken fear amongst a lot of us. “AI, it’s there, what will happen to my role, will I survive, will I be able to adopt?” And let me tell you one thing, and I always use this with my teams: let’s not fear AI. I would say artificial intelligence plus human intelligence will create wonders. Human intelligence will never fade off. So, AI plus HI is going to create wonders.
We use AI to flag decisions, not make them. It is humans who are making the decisions. Especially in sourcing, the tool might come up with several quotes at a go, but it’s a human who’s making the decision. So human intelligence is still there. For example, if a supplier is flagged for geopolitical risk, AI cannot take a decision there, we still need our category manager to decide about the nuances. It’s still human intelligence playing the bigger role. So, to answer your question, AI plus HI is going to be the future.
“Let’s not fear AI. Artificial intelligence plus human intelligence will create wonders. Human intelligence will never fade off.”
Zooming out a bit from the daily ops, where do you see procurement heading in the next three to five years, especially with AI as your co-pilot?
With my observation over the last couple of years, there has been a big shift in procurement from transactional to predictive, and this is going to continue as we progress. Procurement will become less and less transactional, with more AI processes and agents taking over the transactional work, and become more productive on things like demand forecasting, how to negotiate probably standard contracts, and even maybe an AI tool can simulate some risk scenarios before a supplier is onboarded. So, I feel we will go away from more transactional to more predictive.
Think of AI as your procurement analyst that never sleeps, unlike us humans, who take breaks or go back to the process. Having an AI tool implemented as one of our colleagues, who will never sleep, that’s a 24/7 support, so when we come back as human intelligence in the morning, we have all the work done by our agents, which is an AI, ready on our desk, which is making our life easier.
What advice would you give to a procurement leader who is just getting started with AI, something practical, not just theoretical?
My advice to any leader who’s starting, or thinking of implementing something, is to start small. Solve something real. Find out what problems your team members are facing and show the value. You don’t have to wait for a bigger strategy or long-term vision to have something created. Just start, don’t wait for “I’ll have this perfect data, my data will be ready in like six months or eight months, I’ll have it all clean.” Don’t wait for perfect data.
And don’t go alone. Find your internal champions, whether it’s part of your IT team or your operations team, people who are very excited about implementing your AI solutions, because you need partners who are excited about this whole concept. A more enthusiastic team enables you to succeed more, rather than a more fearful team who’s afraid of implementing AI solutions. So be very careful about choosing the team, and collaboration with your respective functions is also important.
So, I would suggest starting small. Start with the real problem. Don’t wait for the bigger data plan, don’t say, “I’ll start the journey after three or six years, whenever my data is ready.” I’m not ready. Just start.
“Don’t wait for perfect data. Start small, solve something real, and show the value.”
Conclusion
Megha’s account of AI adoption in procurement lands somewhere most vendor pitches conveniently skip: AI isn’t magic, and it isn’t a switch you flip. Here’s what stands out from the conversation:
- AI is math + data + process, in that order. Skipping the unglamorous groundwork is exactly why so many pilots stall after the first round of applause.
- The same three “killers” show up again and again: data quality, fragmented processes, and mindset. None of them get solved by simply choosing a better vendor.
- Clean vendor master data isn’t optional. It’s constantly changing, and most organizations still don’t have real-time visibility into it.
- Fragmented systems across regions quietly sabotage AI models. Excel in one market, SAP in another, no common ERP, means no clean training data to work with.
- Mindset is the hardest gap to close. Treating AI as a “silver bullet” instead of a toolbox is what breaks trust and stalls adoption.
- Cross-functional buy-in isn’t a nice-to-have. IT, finance, and legal all need a seat at the table before a single dollar gets invested.
- The real fix is starting small. Solve one real, felt problem, show the value, and build from there, rather than waiting for “perfect” data that never quite arrives.
- The future isn’t AI replacing procurement, it’s AI plus human intelligence. Let AI flag decisions; let people make the judgment calls that carry real risk.
