Advanced planning systems have evolved through major technological shifts, from legacy systems to enterprise platforms and from on-premises software to software as a service (SaaS). Each shift changed not only the tools planners used but also how planning organizations worked. Now AI is ushering in the next transformation. Planning is moving from fixed cycles toward continuous, AI-enabled decision-making, raising new questions about what it means to plan and what it will take to translate that capability into real enterprise impact.
In this Voices On episode of McKinsey Talks Operations, host Daphne Luchtenberg is joined by McKinsey’s Andrew Shih, an asset leader, and Mike Raftery, a senior asset leader for digital supply chain, to explore what AI could make possible for planning and how leaders should prepare for the next era of advanced planning systems.
The following conversation has been edited for length and clarity.
Daphne Luchtenberg: Welcome to this Voices On episode of McKinsey Talks Operations. This is where we dig deeper into the detail. We bring into sharper focus some of the tactical operational challenges and look at how operational excellence combined with tech and AI has the power to change the game.
Today we’re talking about advanced planning systems, or APS, and the AI-driven transformation APS leaders must now put to work. It seems like each decade, supply chain and APS leaders come face to face with a technological shift that forces them to completely rethink how they work. First, it was moving from spreadsheets and legacy systems to enterprise APS. Then came the shift from on-premises software to software-as-a-service, or SaaS, platforms. Today, it’s the move from software-driven planning to AI-enabled decision-making. The ones that are leading here aren’t necessarily the businesses with the best AI tools. It’s the organizations that are willing to challenge assumptions, retrain talent, redesign processes, and let go of legacy ways of working.
Mike, you’ve spent 25 years in supply-chain-focused roles, and then you led SCM Connections for several years. No doubt you’ve seen the industry change over time. Take us through some of those changes.
Mike Raftery: I’m dating myself, but when I reflect on it, this is probably the third extinction-level event in the industry that’s happened to force change and evolution. I started my career in the early 2000s, when planning software was just getting started. From there, it was growing from MRP [material requirements planning], getting into S&OP [sales and operations planning] processes, and then trying to figure out how to get an end-to-end solution going. About ten, 15 years ago, there was a move to SaaS platforms, driven by simulations and IBP [integrated-business-planning] processes. That changed the game for the last generation. And now, in probably the last six months, maybe less than that, we’ve seen a whole lot of change with these AI tools coming in, where it’s kind of getting through some of the hype and understanding what is really going to emerge from the ooze, so to speak. That’s where we’re at today. It’s an exciting spot to be. It keeps me engaged even after all these years.
Subscribe to the McKinsey Talks Operations podcast
Daphne Luchtenberg: Never a boring moment, I imagine. So Mike, what did users of APS planning systems need to unlearn as they transitioned from on-prem to a SaaS model?
Mike Raftery: This was a huge change for the first, I guess, 1.0 to 2.0 of this generation of platforms. I remember being in a lot of meetings with senior executives. The first year it came out, they were just saying we were never moving to the cloud. Like never. And then a year later, nobody cared. It was just how it is. They got through the challenges, all of the concerns and governance. That was a huge change. It was how do we take it from something we have a server to manage to something that’s cloud managed. Right now, I’m feeling that same vibe, that same “Well, we’re just never going to do that” or “We’re never giving our data to AI.” It’s the same type of mentality, and I think people are working through the governance of that. What does it mean to expose your data? What are we comfortable with? What policies need to exist to do that? A lot of those transitions have a similar flavor, but in a much different, more accelerated, more exciting manner with the move to AI.
Daphne Luchtenberg: Andrew, let me bring you in here because you’ve got a lot of experience working with teams moving from SaaS to AI. What are you seeing, and how is this shift different?
Andrew Shih: I think the biggest shift is really just how the cadence almost evaporates entirely. Before, teams and organizations planned on a daily, weekly, or monthly cadence. And now, with agentic AI, you can have agents watching data and orchestrating constantly. You really can plan when reality changes, not when the calendar shifts. A big shift is about accommodating that, but also setting up your organization, your data, your operating model to get the most out of it. Such a fundamental shift demands that you change your operating model pretty fundamentally as well. That’s probably the big shift I’d call out, the always-on aspect of it.
Daphne Luchtenberg: As you think about that, how does the role of the planner change? What new skills have become much more valuable now?
Andrew Shih: Fundamentally, one big one is teaching [the agents], in the sense that when you’re setting up agents to watch and orchestrate your cadence, or to plan for you in real time, you are encoding the context of your knowledge in such a way that these agents can be trusted to do so. That’s one. The second is decision-making, and this might understate the importance of it, but human-in-the-loop intervention will still be super key here. This can’t be a fully automated process. You still need that human accountability, those decision points, that expertise in the moment to make those decisions. The third is really shifting from on-the-ground execution to more of an orchestrating role. Instead of being the one doing the day-to-day work all the time, you’re setting the goals, setting up the guardrails, and deciding what’s automated versus what’s done manually. I’d call it those three big shifts.
Daphne Luchtenberg: So would you say that planners will now spend less time actually planning and more time governing the AI?
Andrew Shih: Yeah. Governing, teaching, coaching, and intervening, all those things.
Daphne Luchtenberg: Interesting. Mike, let me bring you back in here. If that is the backdrop, how do planning organizations prepare for AI? What should they prioritize as they think about reskilling?
Mike Raftery: There’s a lot of concern, maybe anxiety, around that. Do we even have the right team to do it? I think there will be some evolution and upskilling. There are two parts to how the organization prepares. There are going to be a lot of questions around governance models, architecture models, how we actually stand this up, and what an enterprise-grade AI solution requires organizationally. That’s probably one area that needs discussion right away.
As far as the planners go, I’ve heard a lot of different concerns around workforce reduction, and I feel like that’s not at all the case. To Andrew’s previous point, get out of that cadence, get out of that mindset, which is really hard to do if you have planners who’ve been doing the weekend job runs because it took a weekend to run. What if it doesn’t? Why do we have this meeting? Or why don’t we have this meeting? What would you do with this information? What questions would you love to answer if there were an infinite amount of analysis and resources to do it?
Those are the type of questions that, when you start using these AI-embedded tools, really take a pause and a mindset shift to say, what can we do here that we couldn’t do before? Because if you assume infinite scenarios and simulations, your questions become a lot different. Planning becomes less of a cadence, less of “We have a Monday morning meeting because of this, that, and the other thing.” You may end up saying, when these occur, this is who I call, and this is the information they expect. Eventually, that will just become planning. It’s a huge difference in what that task looks like, and I think it’ll evolve over time as some of these innovations come to market.
Daphne Luchtenberg: Andrew, let me bring you in here. Let’s talk about how planners should consciously think about reskilling. What are some reskilling priorities they should be thinking about?
Andrew Shih: I’d say for both planners and planning organizations, probably the biggest shift—and I love Mike’s point earlier about how these changes will happen—from my perspective, is: What are the capabilities needed to enable all these shifts? Probably the biggest one for me is, and this is maybe reflective of a broader shift in agentic AI as a whole, making sure the context is set up, making sure that when we talk about AI or agents orchestrating and enabling all these things, that they’re able to understand the data and the scenario well enough that they can do so reliably. We’ve all heard about the risk of hallucinations in AI, how that can go comically or horrifically wrong depending on the situation.
So we’re setting up the semantic layer, which is one of the things we call it, so that agents have the context so they can actually understand the data in the situation well enough to run reliably. That’s probably the biggest one. The second one, maybe related, is taking a holistic approach. Some people call it a platform approach, others call it an operating system approach, but really, it’s building that orchestration layer so that you’re not building siloed point solutions use case by use case, widget by widget, but taking an overall holistic approach, both technically and operating model–wise, that’ll enable all this.
Daphne Luchtenberg: Mike, let’s take a cleansheet approach to this. If you were designing a planning organization from scratch today, what would you not build into it?
Mike Raftery: A lot of the core skills are there. I think the danger you would have, to Andrew’s point, is that you want people who are working a lot with these models and know how to manage them. So I think what you would want to have are people who can actually challenge effectively. One of the dangers with hallucinations is that these models come back sounding so confident: “Yes, this is the thing to do.” But if you don’t have that governance and expertise to say, wait a minute, that doesn’t sound right, challenge the model. Getting complacent is probably the biggest danger organizations could have. You really need these individuals to be critical thinkers and problem solvers. It needs to truly be somebody who can see complex problems in real time and understand the implications of them. The skills were always found in the best planners, but they weren’t always people in the planning roles. They were the people doing network studies. That’s really where you want to lean in and reward those kinds of behaviors.
Daphne Luchtenberg: So people who are consciously challenging the whole time.
Mike Raftery: Yes. The difficulty is going to be where that next generation of expertise comes from. Generally, those who have that experience got it because they were on the phone getting yelled at by Walmart or Target, right? They couldn’t find an order on time. They built this experience over time. I think the challenge for organizations is how do you mentor the next generation so that this talent doesn’t just walk out the door, and you become overly reliant on the AI models. That’s probably going to be a challenge for a lot of different groups within an organization, but to me, that’s really the risk in making this sustainable.
Daphne Luchtenberg: Interesting conundrum. But I assume some companies are actually doing this well and are on a good road. What are the characteristics that have helped them define that success?
Mike Raftery: What we see a lot of companies doing right now is putting their toes in the water with proof of concepts and pilots, and almost too cautious, I would say. But I think that’s also the right approach. There’s a lot of new skills, a lot of new organizations. Andrew, you’ve dealt with a lot of companies dealing with questions around how do we make a pretty cool tool, like you can go and download Claude or whatever, and vibe code an app in an afternoon. I think they’re still dealing with what that means from an enterprise level.
Andrew Shih: Yes, and honestly, that’s a big part of it. I’ve seen a lot of organizations let a thousand flowers bloom, where everyone builds their own little proof of concept or widget or application. What they end up with in a few months is we’ve got a bunch of cool applications, but where’s the impact? And that’s when the awkward silence kicks in. So I’d say one of the key things is to close the loop and measure it. You want to see that what you’re building is actually leading to impact, whether that’s financial, operating model–wise, or something else, and that what you’ve done has moved the needle in the way that you expected it to.
Mike Raftery: A lot of what’s old is new again. It’s not necessarily about getting the coolest app or vibe coding the coolest little widget. That’s all fine and good, but the ones that are really having an impact are doing the hard work and heavy lifting to understand how to make these things enterprise grade. The risk is that you end up supercharging your organization with a thousand spreadsheets everywhere, with these vibe-coded apps that are unsupported. Data governance becomes risky, and data security becomes even riskier. That’s really the downside of that, the cautionary tale. If you let everybody have a Claude Code license and go nuts, well now you’ve lost control of your organization’s data, decision-making, governance, and security.
We’ve had some clients that had experience with that, and they pulled it back pretty quickly. So I think it’s taking—maybe conservative has a negative connotation—but a purposeful approach. How do we actually want this to run an organization and not be just a cool party trick? Those are, I think, nuanced decision points. But the ones that will be successful in the long term are doing things like measuring impact, measuring for goals: What are we actually trying to accomplish here? If you’re just rebuilding your old demand-planning solution in AI, you’ve missed the boat. What is the actual change that is going to move the needle in efficiency and impact? That’s the question they need to be asking, not just, can we do it in AI.
Daphne Luchtenberg: So if you were advising a planning leader listening to this program, what would you encourage them to start doing tomorrow? Mike, let me come to you first with that question.
Mike Raftery: I think it’s understanding what is required of your organization’s road map to leverage the AI capabilities without exposing your company to data risk or governance issues. This all works great until it doesn’t. We’re kind of at the part of the hype curve where people are realizing there are still some hard things. This is complex data, these are tough decisions, and this is going to take some governance and control.
The problem is that individuals are used to having Claude or ChatGPT at home, and they think they can run with it. There’s a bit of entitlement that needs to be reeled back in. In my mind, there are two items here. One is culturally helping people understand the risk of what they’re doing or could be doing. The second is more from an enterprise side: What do we need to invest in, what decisions do we make, and do we take them in-house or outsource these capabilities for governance, security, data storage, and so on? Those are probably the two most pressing questions I’d be working on today.
Daphne Luchtenberg: Andrew, from your perspective, would you add anything to that list?
Andrew Shih: Maybe not the list itself, because I fully agree with everything you said, Mike, but I’d layer a lens on top of it, which is, I’d advise a leader to really pick one tough problem in your organization, whether it’s a particular domain or something else, and start with that. Bigger than a pilot, but smaller than a full organizational transformation. The former isn’t always guaranteed to be impactful; the latter could take years. Pick one tough, meaningful problem you can start with, solve that, go through all the things you mentioned, Mike, and then grow that accordingly afterward.
Daphne Luchtenberg: Great advice. I wanted to ask a question of both of you. Let me start with you, Andrew. Help me finish this sentence: If the SaaS revolution changed where planning software lives, what will the AI revolution change?
Andrew Shih: It would change what it really means to plan at all, if that makes sense. Because if I sum up everything we discussed, the shift is almost from producing and executing the plan to more architecting and then governing the system that plans continuously while still preserving expertise. It’s a pretty fundamental shift.
Daphne Luchtenberg: And Mike, how would you describe the AI revolution, and what is it going to change?
Mike Raftery: I think it changes what’s possible. I know that’s a vague answer, but I’ll clarify. Planners are going to understand, when systems are no longer the constraint, and hardware performance and data models are less of a constraint, what questions to ask that they could never ask before. That’s going to be an incredible unlock for organizations. Once that clicks—and it will click—the revolution will come quickly as people understand how these are used in their daily jobs. I think we’re still trying to figure that out. It’s a very exciting time for the possibility of what’s happened in probably the last six to nine months in this space, and it’s going to change even more quickly, probably in the last part of the year.
Daphne Luchtenberg: And I know we’re never allowed to make predictions, but Mike, how long do you think that will take?
Mike Raftery: Sooner than expected. I think it’s Gartner that has that IT software hype curve, where it’s just coming off the peak of expectations into the valley of disillusionment. I think that’s going to hit pretty quickly. Because organizations like ours are starting to address this, we’re asking the questions; we’ve been through these revolutions before, and it’ll happen a lot faster than usual. We’re seeing clients understand the challenge in front of them and tackle it very purposefully, to use that word again. I expect that by early 2027, you’ll see a lot less piloting and a lot more utilization. You’ll see some really strong success stories by next summer.
Daphne Luchtenberg: Andrew, do you agree?
Andrew Shih: I agree. I think the tokenmaxxing trend is dying out. People are getting a lot more serious and disciplined about it. So the time frame sounds about right to me.
Mike Raftery: If I have one thing to add, it’s that there is an environment still changing underneath us. I’m going to put an asterisk on all of this because there are all of these third parties out there, and as new models come out and new capabilities evolve, so will these questions evolve. You can’t set and forget. This has to be a constant change and a constant agile evolution to take advantage of it.
The development cycles are so much quicker than they’ve ever been. Organizations will have to really stay attuned to this and live in that space. It’ll be interesting to see how the ecosystem evolves. We’re used to yearly conferences where everybody gets together and sees what the new thing is about, but that’s way too slow. These things are happening monthly, almost weekly, at this point. So how does the overall ecosystem incorporate new players, like Anthropic, ChatGPT, all these tools that are available to you; they’re going to just be a new player in the group.
Daphne Luchtenberg: I loved what you talked about there, Mike—the curve moving into a valley right now, where there’s quite a lot of disillusionment, but we’re going toward an upward trajectory. What’s also going to affect us is the ecosystem of technology and the constant shifting sands around the technology, the solutions, and the other tools that will be available as we move on that journey.
Today’s conversation has been about AI as much as it’s been about the importance of reinvention. APS leaders have lived through one major transformation, from on-premises software to software as a service. The AI era presents a new challenge: not just changing the technology stack, but rethinking how decisions get made, how teams collaborate, and where humans create the most value. The question isn’t whether AI will change planning but how we will change it. I want to thank my guests, Mike and Andrew, for this wonderful, thought-provoking discussion. It sounds like we’re going to have to invite you back in about six months to talk a bit more about how the sands have shifted. Andrew, thank you for joining us.
Andrew Shih: Thanks for having me.
Daphne Luchtenberg: And Mike, thanks for being with us.
Mike Raftery: My pleasure.
Daphne Luchtenberg: You’ve been listening to McKinsey Talks Operations: Voices On, the place where we talk about operational excellence and how it has the potential to change the game.
