Four tests before agentic AI runs a procurement process alone

Edited for readability from the episode audio. Filler and repetition removed, meaning unchanged.

00:05
PHILIP IDESON: Today on the show, I'm delighted to welcome Liza Tsimbal. Liza is the Director of Supply Chain Management for TransAlta, a publicly traded power generator based in Canada, with customers across Canada, the United States and Western Australia. First of all, Liza, thank you so much for joining me on the show today.

00:23
LIZA TSIMBAL: Hi, Phil, and hi, everyone. I'm happy to be here.

00:26
PHILIP IDESON: It's our pleasure. Let's kick things off by understanding a little bit about your role. If you could give us a quick picture of what you're responsible for, and how procurement fits into the broader organization at TransAlta, that would be really helpful.

00:40
LIZA TSIMBAL: For sure. I lead the supply chain function at TransAlta. TransAlta is one of Canada's biggest power generators, and we've been generating electricity for more than 115 years. We have a very diversified portfolio of more than 90 generating facilities in Canada, in Australia and in the States, and the assets span hydro, solar, gas and battery energy storage. So that's a lot. Supply chain at TransAlta is not a back office function. We are an operational one. We support the full spectrum of everything the company needs, from day to day procurement, outages and projects to IT and corporate, you name it. We purchase all the equipment required for the growth of the company. We do complex negotiations with OEMs. We do everything. And I look at supply chain, and what my team does, through three lenses. It is reliability, because we're a power generator. It is speed, and it's cost. Those can really pull you in different directions, so my job is finding the right balance for the assets and for the risk.

01:59
PHILIP IDESON: And how is your team structured? When you think about your team specifically, what does that look like?

02:05
LIZA TSIMBAL: I have roughly 30 people across buying and strategic sourcing, and we support the entire company. Our supply chain is centralized, but the team is not sitting together. We have some people in the corporate office here, and also at sites across Canada and Australia. The buyers are the closest to the business. They keep the company moving: the purchase orders, the outages, the expediting. The sourcing specialists are more on the complex side. They're building the strategies, working with major suppliers and doing supplier performance management. That said, I don't believe in high walls between strategic and operational procurement, because strategic work has to understand the operational consequences, and operational work needs to understand the commercial context. For the last three and a half years, we've been moving away from being somewhat of a purchase order factory into more of a strategic function. We're doubling down on category management, supplier relationships and early engagement in business decisions.

03:19
PHILIP IDESON: How do you manage those handoffs? It's really important, what you said. You can get into silos where somebody in a category management lead role thinks, I'm responsible for the category, so I'm responsible for the strategy, and I'm not too concerned about downstream. But of course, the decisions they make impact downstream, and if you're not careful, you create a world where they're not connected. How do you make sure everybody is connected? Is it one role that does multiple facets, or do you have other ways of making sure there are connections between folks who are local in the facilities and those in head office?

04:02
LIZA TSIMBAL: This is an excellent question, because I've seen what you've described before: the siloed approach, my turf, staying in my lane. What we do is cradle to grave, and we also flex people across the portfolio. When you think about it, the work doesn't come in neat organizational boxes, and when the business looks at the org chart, they don't really care as much. They just want things delivered on time and on budget, and these days, even ahead of time. So my role is to make sure I understand the full picture, explain the why to the team, and then give them the opportunity to deliver what's required for the business, regardless of their title. It's not about the title. It's about the work that needs to be done.

04:50
PHILIP IDESON: Their experience of working with procurement as a whole. Now, we called this podcast series The Agentic Procurement Diaries to really focus on agentic AI, and I'd love if you could share a little bit about where you are on your journey with agentic AI.

05:07
LIZA TSIMBAL: Phil, we're early.

05:09
PHILIP IDESON: Yeah.

05:10
LIZA TSIMBAL: We're really early, but we're seriously exploring. Assistive AI, through Copilot and other applications, has been around for us for the last several years. We're using it for contract reviews, for first drafts, for scenario planning, maybe even for getting to the better question sooner. But agentic AI is a different ballgame altogether, and this is where the value comes. We're now at the point where we're focusing on practical use cases. For this year, we've identified two areas. The first is requisition to purchase order automation, and the second is contract and purchase order leakage.

05:59
LIZA TSIMBAL: Next year should be spend analytics. That's the roadmap at a high level, and we're at the point where we want to pilot these things, but small. The pilots, in my view, have to be small enough that we can control them, but meaningful enough that we can measure the impact on the work. Once we identify the areas we want to pilot, we also want to be cognizant of the impact on the company, because at TransAlta, we came through a major ERP transformation just last year. We moved from SAP ECC to S/4HANA, and that's been quite a journey over the last three and a half years. The business is also in a very active growth phase, through acquisitions and development, so internal capacity is at a premium. And personally, just to share something with you, I'm trying to avoid the whole AI FOMO, because everyone and their dog is talking about AI. AI is everywhere, but scaled results are not. So I wanted to start with the business case, where I define a business outcome that is clear and well structured, and then decide where the answer belongs: on the platform, let's say SAP and Joule, or whether we want to consider a point solution. That's where we're at right now.

07:32
PHILIP IDESON: As you were looking at all the different ways you could deploy agentic AI use cases, what were some of the considerations that went into selecting req to PO and contract and PO leakage as your two test cases?

07:53
LIZA TSIMBAL: Initially, we were thinking: what are the cases where we have a workflow that is high volume, that has defined rules, and that no one really likes doing? Those two cases fit the profile exactly. We have a high volume of transactions. They're rules-based: we know exactly what constitutes a good requisition and what needs to go on the purchase order. We have defined terms and conditions, so the agent can compare the invoice against the contract for leakage. So those were the criteria. High volume, people don't like the workflow, it's potentially reversible if the agent makes a mistake, and it's rules-based. Those four criteria.

08:56
PHILIP IDESON: Was there a specific moment where you felt agentic AI could really help in these areas? These have obviously been pain points for a long time, and I can attest to it from my career. The req to PO process means lots of pain for procurement and lots of pain for the rest of the business. At what moment did you think, I think agentic AI can be the solution to that problem?

09:28
LIZA TSIMBAL: It wasn't necessarily the driver for those two pilots, and it wasn't a lightning bolt. It was just a random Tuesday. We had a technical issue with one of our OEMs, and we needed to find a specific technical piece of information. We knew it existed somewhere. It could have been on an old drawing, in the contract, in an email chain. The problem was finding that piece. I was watching a lot of smart people, days in and days out, looking across documents and opening folders. And this is what crystallized it for me: information that you cannot retrieve fast enough might as well not exist. When I reflect on what my team does in negotiation, the value of information is partly about speed. If you find the answer a week later, you might have already lost leverage, or time. That's why I started thinking about how AI can support us on that journey. That was one angle. The other is that as we get bigger as a company, we cannot keep expanding the support structure to fit the growth of the company. We need to find a way to create capacity in my team, because supply chain talent is scarce. We can strip out the transactional and repetitive work so people can focus on things that move the needle, that are truly value-added. That was the moment.

11:13
PHILIP IDESON: We're all under pressure to do more with less, and to figure out where the efficiencies are. As you started to explore agentic AI, did you find any pushback, or unexpected pushback, to your plans? Or was it, why have we been waiting so long? I'd love to know about the resistance or the encouragement you came across.

11:41
LIZA TSIMBAL: The pushback was definitely there, and for the most part, it was reasonable. The questions being asked were: can we trust the answer? What data is it using? What about confidentiality, cybersecurity, data access? Are we even automating a bad process? And on the people side, of course: does it eventually mean fewer jobs? There is also a business case question when it comes to AI, because an AI demo might sound very compelling, but every leader wants to know the impact on their P&L, on productivity, on risk, and that's all fair. The way I think about it, if I cannot connect the use case to a measurable business outcome, then it's probably not ready for investment.

12:37
PHILIP IDESON: And how do you measure those baselines so you can actually assess the ROI?

12:42
LIZA TSIMBAL: First of all, you need to define what good should look like, which is very hard. Then you measure before and after. That's the only way to know if you're making progress, and whether it's meaningful progress that will justify the investment.

13:10
PHILIP IDESON: How do you think about the role of the human in partnership with the AI? Is there a world where certain tasks can be done entirely by an AI agent, without a human in the loop, or does there always need to be a human because of that trust factor?

13:36
LIZA TSIMBAL: Oh, this is where I really struggle a lot. First of all: what are the use cases where we should use AI, versus where we should not? It goes back to what you asked me about how I picked the use cases. The test I typically use is whether the work is low risk, high volume, rules-based and reversible, and I want all four. That could include routing requests, checking completeness, exactly the requisition to purchase order and the leakage, and monitoring tail spend transactions that seemingly never end. This is where I think we can have full AI automation. Where I would not even think, at this point, about handing anything over to AI is when an activity falls into the other box: when it can create a new material obligation, when it's a major negotiation, when it's a critical supplier, or when it's a decision with a safety implication. At the strategic end, AI should make the human better, not remove the human. But at the transactional end, once we make those transactions boring enough and safe enough, I think the human can remain as oversight, and not in the loop.

15:27
PHILIP IDESON: When you say a process, action or task is reversible, could you expand a little on what you mean?

15:37
LIZA TSIMBAL: A reversible decision: if I cannot unwind a decision made by the AI, then a human needs to be the one approving it. Because I could never go to an auditor or an executive and say, the system did it. So it needs to be easily reversible.

15:58
PHILIP IDESON: So for you, you're always accountable for the actions taken by an AI agent, as much as for the actions taken by a member of your team. And so you want oversight accordingly: where you feel comfortable with an agent doing something, and where you don't, based on that accountability.

16:24
LIZA TSIMBAL: A hundred percent.

16:25
PHILIP IDESON: How do you feel your team is embracing agentic AI? You mentioned we should be using these technologies to empower our teams, to give them the ability to do more with the time they have and work on more strategic activities. And of course, there's always fear about what it means, because the role might look very different in the future. Is that something you found when you socialized the direction of travel? I'd love to hear what you found from the team themselves.

17:08
LIZA TSIMBAL: Again, we're still very early. The team hasn't seen the results of the pilots yet. There is definitely an element of fear, because, to your point, what will their role and their job look like in one, two or three years? But the way I'm thinking about it, the strategic end of supply chain will become more human, because we will automate a lot of the transactional work. People will have time and capacity to focus on things they probably enjoy more than the transactional and the repetitive, where they can learn more, and that will be more meaningful for their own career development. That's on the team side. But when I think about the job profile of the future supply chain person, we need to be very cognizant of the changes we're making. If we're going to automate much of the junior work, how do we build senior judgment? We cannot automate the apprenticeship out of procurement and then wonder where the next generation of commercial capability comes from. That means we need to be more deliberate about building commercial acumen, supplier exposure, decision skills and AI literacy, or maybe even hiring for a different mix of capabilities.

18:41
PHILIP IDESON: So could that also mean not selecting every single use case that could lend itself to AI agents, because you want to create a pipeline of talent with that experience? Just because you could use an agent to do it doesn't necessarily mean you should, when you think about the talent pipeline and the experiences you want to give people.

19:12
LIZA TSIMBAL: I really like how you summarized that. In my view, the goal should not be to have the most AI. The goal needs to be making better business decisions, faster, and with the right controls. Whether you use AI for a piece of it, say one third of your transactions, will be very specific to the environment and the industry. It cannot be one size fits all.

19:38
PHILIP IDESON: I want to leave with a final question and ask you to look into your crystal ball a little. This is probably the hardest question, because we don't really know where things will be in 12 months, never mind three or five years. But what do you believe will be true about procurement in three years' time that may not be obvious to us all today?

20:08
LIZA TSIMBAL: This is an excellent question, and something I've been thinking about a lot, because everything in procurement changes on a weekly basis: tariffs, counter tariffs, you name it. When I look into my crystal ball, there are three things. First, as I said, I think AI will make the strategic end of procurement more human, not less, because we will be automating the mechanics. So the premium will go up on commercial judgment, on influence and on strategic supplier relationships, because that part is not going anywhere. Second, I think the scarce skill that will be in demand is the ability to define what good should look like, because you cannot automate something you haven't articulated yet. It could be the rule, the threshold, the specification, you name it. In my view, people who can articulate their judgment will be much more in demand than people who rely on their intuition alone. And third, procurement will move more upstream and become more connected to the operating system of the business. When I look at my own industry, we're seeing unprecedented supply chain disruptions and lead times measured in years. You cannot wait for a project to be fully baked and then come to supply chain to source it. The teams that can connect maintenance planning, inventory, supplier intelligence and project demand, and tie it all together so they can shape decisions earlier, will be ahead of the curve. That will be exactly the competitive advantage they bring to the company. Those are the three things.

22:28
PHILIP IDESON: Can I ask a follow-up that ties some of these together? As we've grown up in procurement and supply chain, it's easy for our mindset to be constrained by what we were brought up to believe the role of procurement and supply chain is. How important is it for that talent to see past those walls and start with a blank piece of paper? We now have the ability to do far more than we used to. Where does procurement start, where does it stop, and how is that different from what we've been used to?

23:12
LIZA TSIMBAL: That's extremely important, Phil, and I think you nailed it. It's the ability not to apply our preconceived notions and automate a potentially bad process, but to start from scratch, from a clean slate, and build something fit for the business, where the process is not dictating the activities. You look at where the company is going and how you can help the company accomplish its goals using whatever tools you have in your toolbox. It could be AI. It could be clean data, and sometimes clean data will be enough. Do you really need AI when you can search? It's difficult to tell right now, but I think we can anticipate a lot of changes in our industry, and in procurement as a whole.

24:09
PHILIP IDESON: I think we become architects. We now have all these tools at our disposal, and the question is how we fit together the right tools for the right moment, the right stakeholder and the right project. The answer is not what it used to be, which was, I'm going to throw a sourcing process and an RFP at that. It's, how can I build something that feels custom for this moment in time, in a way that is still scalable. That's going to be hard, and it requires a different mindset.

24:41
LIZA TSIMBAL: A different mindset, and the mindset should be: start with the outcome in mind and build everything else from that, as opposed to starting with a process and trying to fit the outcome in somehow.

24:53
PHILIP IDESON: Liza, thank you so much for joining me today on the podcast. It's been a pleasure. We'll link to your LinkedIn profile in the show notes, so anyone who would like to reach out can get in touch. The show notes will be at artofprocurement.com/podcast. One last time, it's been a pleasure to speak to you today.

25:11
LIZA TSIMBAL: Thank you, Phil. It's been a pleasure as well.