328. Driving agentic transformation: How AI is rewriting the playbook on large-scale change

We observe all these jobs change, but they change in ways that are cool. You spend less time making a Powerpoint deck and crunching, you know, figuring out some data irregularity, and spend more time coaching that initiative owner on how to make the procurement initiative more impactful. That's cool. And so I I think what we found at all levels of our transformation office, from the most junior person all the way to the chief transformation officer, him or herself.

The role is different, the tools are different. The underlying core job of changing humans is not different, but you go about it in a different way. And that way is generally more fun and exciting and allows people of all tenures to have more impact.

From McKinsey and Company, I'm Sean Brown and welcome to Inside the Strategy Room. That was Chase Covington, co-author of our recent article, The Agentic Transformation Office, redefining the economics of change. And in today's episode, we discuss how transformation workflows are ripe for agentification and how deploying agents allows a transformation office team to focus more on the human side of change. Coaching, generating ideas, pushing for value, and the change management itself, the kind of work that makes for an even more successful transformation.

Chase is a partner based in our New York office and a leader in our transformation practice as well as our strategy and corporate finance practice. Chase, welcome,

Sean, appreciate the introduction.

and Tomer Slaney is a partner in our New York office and an expert in change management and long-term change success. Tomer, it's great to have you here.

Thanks, Sean, good to be here.

So, in your new article, you note that the transformation office, the very function charged with driving massive change across organizations, is also proving to be one of the slowest to adopt AI technology. So, Chase, why is there this somewhat ironic lag?

Yeah, that that's kind of the dirty secret, right? I I think we, uh, what we observed is a lot of transformation offices or PMOs or integration offices, whatever variant you wanna name, you wanna put on that, uh, we're relatively slow to identify, uh, despite often pushing agentic solutions onto an organization, or at least espousing to to do so. And I think that, you know, to some degree, the the core reason for that is they're just busy, really, really busy with the day to day.

There's always a ton of work in a PMO or a transformation office or a growth office, what whatever it might be. And so the insight for us was how do we get those group of clients take a step back and really think through how those workflows can be different. And when we did that, what we typically found is a lot of the grunt work in a transformation office can be agentified in a way that enables the people in that office to operate at a higher level.

So you're not producing a report, you're producing a change outcome in the enterprise. And that's really the key to the whole, generally to the article we wrote, to the what is powerful about agentifying a transformation office.

Thanks, Chase and Tomer. Could you talk a little bit about where AI fits into the broader context of transformations and where you see it creating the most value?

Yeah, absolutely. So when we talk about AI and transformation, really think of it as an entire range of. How do you identify the functions in the transformation office? How do you elevate the work that's being done and the level which it's operating? There's also the entire notion of how do you identify the transformation? How do you make things available to your participants? If that is co-pilots writing business cases, plugging to ERP systems to populate, um, financials and impact, tying it to board documents and all the way on into transformations, really being the or AI being the catalyst for transformation.

So the reason for change for many of the organizations is preparing themselves and rethinking how they remain competitive in an agentified or in an AI forward environment going forward.

And maybe Sean, I, I add one thing here, just as we think about what is an agentic transformation. A year ago, two years ago, I think we were getting lots of inbounds from clients saying, I, maybe I want to do an old school transformation, or I wanna do an agentic transformation. Those are somehow separate things, or or separate asks, uh, maybe even separate approaches from our side. But I think what we've seen over the course of the last twenty four months, that coming together.

The analogy I give you is if you were to tell somebody you're listening to digital music versus just on Spotify, of course, you're listening to music. Uh, or the vinyl. And so the concept of an agentic transformation is just all transformations need to be using agentic levers because that's just one of the most defining technologies of our time. It'd be like saying, Hey, I'm using the internet for my transformation in 2005. Uh, of course, right?

So I think we will reflect that language throughout here. I don't think there's in our minds as much a distinction between an old school transformation and an agentic one, because agentic's just a really powerful lever to execute change at scale in a business. That we're working into all the all the efforts that we're doing

Thanks, and just to dig into that a little bit more deeply, Chase. Does the concept of agentifying the transformation office make it more accessible to more companies, or does it make it more straightforward, for example?

it. It makes it more accessible to more companies because the human resource load required to stand up an effective transformation office is lower. It doesn't mean it's any more difficult in terms of executing change, but just the standing up the basics, we find the barriers to entry there are a little lower. I think what's more interesting, though, is less, hey, more companies can do this now, because you don't need as many people in the TO. That's less interesting.

What I think is more interesting is in any company that's using a TO, your TO can reach more employees. And if you can reach more employees, you can embed change more granularly into the organization, which our research shows dramatically increases the odds of success. That I think is actually the more interesting conversation, right?

Got it? So, Tomer, how is the agentification of the transformation office changing what the TO actually does? Is there a market difference in the typical activities that those in the TO are spending their time on? And what does a typical transformation office week look like?

Yeah, so if if we think about sort of what that catalyst for change was and why to identify, we went and looked at some teams and we have the research here. One example team that we looked at four person team. How are they spending their time? And 80% of that time is spent on reporting, creating materials, preparing for meetings, doing the follow-ups, a lot of the stuff that is both very time consuming, but also very condensed into a certain time of the week.

But also very repeatable, very predictable, very manual. As sort of Chase said, that's the grunt work of the transformation, but it's stuff that you need to do. It's stuff that you need to do and you need to do correctly, and you need to do well. Because a lot of the change management occurs when there is trust in the in the system, in the transformation. This is the kind of stuff that you need to do to maintain that trust.

The materials that are produced are correct, the stuff's on time, the decks are ready. And through both the amount of other tasks that these teams were doing, the people staffed on these TOs, the people running these TOs. Just getting more exposure to AI tools, whether it's cursor or take your pick, and looking at other initiatives in the transformation, saying, hey, these are the things that are being agentified in the company. Start to taking all of that initiative, saying, hey, we can make our own lives easier.

There's also a bit of the tension where longer running transformations towards the back end of sort of the horizon. The expectation is that these transformation offices get smaller and they get more efficient. But in fact, what we actually see is they actually start getting access to more and more data sources as these transformations progress, right? We create the more capabilities, we feed in more data, which is actually increasing the the time required to do reporting, which squeezes out more of the elevated discussion that Chase was talking about, like creating the insight, driving the change, creating the narrative.

So expect it to be doing more with less and getting more of the grunt work to boot.

So, does this help the transformation office move more quickly, or just focus on, as Chase alluded to, more of the change management, which in turn then helps the organization move more quickly,

Within a given week, you are going to move more quickly from when data is available until materials are ready, until insights are starting to be published, until we start looking for the answers to the questions raised by the data. The transformation will still move at the cadence that we expect it to move at, right? Transformations tend to be regimented, moving at a a weekly cadence between the meetings in preparation for executive committees, board meetings, those sorts.

What we are allowing these teams to do, or what these teams are actually allowing themselves to do with these tools, is front run a lot of the grunt work early. When data is available, spend more time on the insights, on the coaching, on the on the actual human side of the change management in preparations for these meetings or these big um events where decisions can be made, where things can be rolled out.

right? And so, as you see more of this agentification of the transformation office in your work, what are what are some of the early hiccups you've experienced?

Yeah, well, I I'll give you one. It's kind of a funny example, but it's it's emblematic. Uh, I think the the technical term is cognitive surrender, which is a really scary way of saying you just let the AI chatbot do your job. A And we actually saw some of this really with more junior colleagues, colleagues of ours, or or even, you know, clients where it's, hey, this chatbot can create something that looks at first blush pretty good.

It's formatted really well, it's got color scheme, and it's kind of structured and all, and that's neat. Um, but then when you review it, it actually turns out to be slop. And so the pitfall is how do you create the right moments for human in the loop to ensure that the output is high quality? How do you treat the agents you're creating as almost like junior colleagues, but still keep the reviews as the information moves up the chain, so to speak?

So that by the time it's going in front of the CEO, you're you're not embarrassed by, you know, some AI AI slop.

Thanks and and Tomer, what kinds of challenges have you seen as teams have moved to agentify their processes?

Yeah, no, like the spe- speaking of in front of the CEO, similar to that sort of the cognitive surrender, there's also the attention surrender. If you know that something else is paying attention, you can multitask. And we actually saw one of these teams develop a good way to turn meeting transcripts into ac- into like lists of actions. So at the end of many of these meetings, we will make sure that we know who's doing what as next steps.

And one of these chatbots, which was looking at the list of actions, picked on a little bit of sarcasm in the meeting. Where the chief transformation officer asked, Uh, a participant, do you just want the CEO to call every single employee and tell them to do their job? And the the chatbot picked up an action CEO to call every employee in the company, tell them to do their job. Which, you know, luckily, you know, it was still very much a human in the loop.

But that would have very much, uh, oh, unattended, been sent to the CEO and tell him to go call everybody in the company and make sure they do their job.

Yes, that's a great example, and not one that I think the CEO would have appreciated. But I think the lesson is to pay attention as technology actually accelerates the transformation. So, Tomer, on that point, can you talk a little bit about the cadence, the pace of the transformation office as it agentifies and and how agentic impacts the pacing? Because every week there's a lot of information that's gathered, synthesized, and then reported on. How does that, really, you know, increase the metabolic rate of a typical transformation?

Yeah, happy to talk about what we, you know, think about as a typical transform transformation. Everyone's going to be slightly different, different dates, different periods. But, you know, generally speaking, um, we are going to be working in a cadence where there are the where. There's the ability to elevate discussions and decisions to an executive committee every week. That's sort of some of the magic of the transformation of driving the rigor and the decision making and the cadence, which allows us to drive change a pace.

And to do that, we have to prepare for this meeting every week. And there's a whole cascade of meetings and new data pull, so data is pulled every week. We then look at the famous week over week change. The nature of this means that from the minute the data's pulled until the meet the the, so the key meetings happen, it's when we tend to crunch all of that. What we call the grunt work, the repeatable, predictable manual work that needs to happen, tends to happen between that data pull and the executive meeting.

And you can move the data pull around the week, it it won't materially change the way that that crunch happens. But unfortunately, from when that data pull happens until the executive meeting is when you actually want the best. The best time to spend time with the data for the insights for the actual decisions that you need to drive from that week. When you do that with last week's data, you always get the questions like, Oh, you know, is this data changed?

And that's used very often used as a way to not make decisions and not drive change. So we're, we're. We've created this structural limitation or structural ceiling, as we've called it, that we compress all of the, uh, the TO team's time between when we're doing the data pull and when we're having the meetings where we need to be using that data. And not leaving time for, as Chase mentioned, that elevated thinking and discussion and coaching and teeing up the decision and the strategic.

So what's or the big insights from that's from that same data pull, which is unfortunate because come Friday and Monday, when the teams continue doing some processing, they come up with all these insights. And those might be either stale or the data might have changed by the following Tuesday. Or whenever you pull the data, which which case they they spend a lot of time just bridging back to last week's data and saying, is that still valid?

By that time, they're crunched again with a data meeting happens. And there's also the, you know, just that sense of like, we know that we need to have that discussion. But the the just the data wasn't ready because we were so busy updating things.

Thanks, Tomer and Chase. Earlier, you alluded to this idea that agentification could also be applied in a project management office, for example. So in your client work, are you seeing this agentic approach being deployed across other major change programs?

All of it, all of it. Whether it's a transformation, whether it's a turnaround, whether it's a merger integration, whether it's a divestiture, a carve out, whether any sort of change program broadly defined. I think this is super relevant for now, the exact tools and how you think about what the objective of the PMO or the transformation office or the strategy out whatever that changes. I I think for sure, how you configure the setup of the of the office and the tools that you're deploying.

But the concept of using these tools to take the people in that office and make them more effective, that's that's pretty universal.

So I'm curious to touch on what the experience is for the people in the organization who are then driving the change. What is their interaction like with this agentified transformation office, and where does the human in the loop remain critical there?

Yeah, so, first of all, there's the folks inside the TO, and I think the change management management there is also very important. If you are an individual leading or setting up a TO, realizing some of the design choices you'll be making in deploying these tools and the change management you'll need from the from the transformation office team, right? The transformation office team is also a group of individual who individuals who will, uh, have to adopt some change.

There's a big risk of, you know, call it false positives. Or just the fact that the tools can create so much more data, and that will, that will actually do the inverse of what you're looking for. That can actually go and create a lot more work for the TO team. If you, if you sort of don't set thresholds or guardrails of like, I want to know about every change from last week. And I want a little speaker note next to every little thing that changed.

Or any time that somebody's updated a forecast and not updated the KPIs, I want to know if that's an error, or if there was an error before that they're fixing, or if that's now, if that's, you know, if that was intentional. The fact that you that you can now go and look at thousands of initiatives and tens of thousands of KPIs and impacts can actually drive a lot more work for your TO team, right? So there's also the change management management, uh, element in there.

Many of us have probably received, even if they are well put together, AI generated content from our teams, clients, assistants. You know how to spot them, you know what those little icons next to all the bullets look like. There is some, there is some hesitation. The bar for this being useful is higher. Just generating more and generating more often to more people is real is rarely the solution, actually. And a lot of the things we'll be talking about in the human in the loop.

Most of what we've seen from best practice is these tools generating content and insight for the TO team. And the human in the loop is still not just, um, doing f- uh, quality fact, context checking. They're also, you know, uh, applying some common sense of does, uh, does everything need to be sent out? Does this need to be personalized? Uh, especially when it comes to executives. And and while you know, setting up an entire different structure to run transformations in every in every function, you know, I I think will backfire.

You are, as we say, in crunch, the same executives coming to the meetings from the BUs. If they are not getting what they need from answers, insight from the TO will likely at some point, once they get, you know, hammered a couple of times, stand up their own function or capacity within their teams to generate that. Those, um, individuals are prime candidates for getting some of this automate automated stuff. You know, they're very much an extension of the transformation office.

So if if you've already got those as far as cha- change agents in your organizations, those folks would be sort of the prime candidates. Where we do see the benefit of sending more automated communications.

Can I, can I just real super fine point on this? Don't use your AI bots to talk to the business. That sucks. The the the purpose of a transformation office is to drive change that's human to human. Cause if you when your TO is interacting with the quote unquote business, they're pushing for value, they're counseling, providing feedback. No agent's gonna do that, so use your agents to make your TO people better, not create more work for folks.

Got it, thank you. And and Tomer, can you talk through the areas of work where you would say agentification really has the most impact?

Yeah, so we we started talking about sort of those data, the data pulls, the creating, creating the materials, creating the slides, creating the, you know, the first pass of things. I think we've, we've discussed that quite a bit there. There is the offloading of the weekly task, like go check this, go update that, go send, you know, send this into that repository, do do the data cut. One thing that we've actually seen teams push on and create a lot of value out of is once these materials been created, right?

The emails have been sent, et cetera. Uploading everything and letting your agents, like creating agents that will go and check this week's materials, this week's uploads, this week's downloads versus other things. And they they tend to catch, and they are able to go through just so much more than your team can. They can go through every last slide on every last appendix and identify, hey, is there a place where this week's insights contradict last week's insights?

Is there a place where the data that we're showing in the transformation is contradicting the data we showed in the MBR last week? Is there a place where, um, where we saw a change in initiative that we expected to see a change in the the pendant initiative, and we didn't see happen? So as these agents get trained better in a in a single transformation context, this is the kind of work that makes the makes this team do better, not just do more and do faster.

So we're seeing sort of the whole spectrum of that, of that work, being agentified, from pulling the data, doing the admin work. Um, we're seeing we're seeing agents being trained right now on meeting attendance and agendas. So if last week we agreed that this is the action and this person needs to come speak to that, is that person been invited to the meeting? We have somebody share sharing their change story, somebody being recognized if there's an expectation they're in the meeting.

Like it was usually the TO's team to go and make sure that we have the right agenda and we have the right people and these people have been invited, et cetera, et cetera. Now, you know, Outlook MCPs allow folks to even go and say, Hey, have the invites actually been extended to the right people for the right meeting? Or have we sort of missed that? We put them on the agenda, but not the invite. And and show up at the meeting like, where's Jenny?

I was like, Oh, oops,

Maybe I'll jump in, Tomer. I think the risk here is if you're new to this, you're gonna have your TO teams create a whole lot more stuff. The decks will be three times as long. All that that's not that helpful. When I counsel clients on it, I use a same length deck, sharper insight, so use the tool to get to that. It's easy to create a lot of slop, uh, but that's generally unhelpful. So keep that in mind when you think about what work, you know, moves to agent versus stays with human.

And so, Chase, what are some of the ways that you help make sure that that happens in your teams?

Honestly, the simplest way to do it is have someone pitch to you the material they've created. Cause it'll quickly sort out what's like, Hey, I've actually thought about this. Or I just typed something in and hit enter twice, and, you know, emailed it. So have it almost Shark Tank style pitch and ask a little bit of kind of probing questions. Do you actually understand where that data point comes from? Where am I gonna use this in counseling someone from the business?

Those questions typically pretty sharply get to the meat of what you're trying to communicate, not the kind of fluff that that goes around it.

I've got one more, I've got one rule. My teams you can use any tools you would like, as long as they're sort of approved to use by our risk, the client's risk. But if you put something on a page, it's yours. The same way you would never, you would never say, you would never respond to, Where do you get that information? That Google, right? It is an unacceptable answer for somebody to say where you know who who wrote that.

And you say, cursor, that sort of is a use anything you want. But if it's on a slide, it's your materials, and that is the rule for every, every single person on any one of our teams.

Makes a lot of sense. And Tomer, you just mentioned the word risk. So there's the risk of accepting what you get from an agent without questioning it. There's also the risk that information within the TO might leak outside of the organization or even within the organization. So the TO, by its nature, is privy to very sensitive information. And I wonder how you help keep tabs on that from an agentification perspective. And how does using centralized tracking software, including our wave transformation platform, help to mitigate this risk?

So, earlier days of of a lot of this agentification, many material creations updates were driven by extracts. A team would extract data the same way they would for everything else and feed it in. In pre agentification, there was the common sense of saying, Hey, this is actually has the word confidential in the title, you know, this kind of sense of information that we would not like to share. And that didn't make it into the materials. That control is not necessarily built into any, you know, vibe coded agent that somebody's come up with, or a automated update.

So for we've started, we start by putting putting rules in place. As you're downloading, extracting information, there's a certain way to do it. But more more recently, you know, especially for the transformations using wave, the wave AI team has built to start building something called wave apps, which allows allow this is it's actually designed for this. It allows you to work with data in wave. A You don't keep extracting data, right? Even extracting that data onto shared drives.

There's that, as you said, there's a potential there's quite a bit of sensitive data in these in these, uh, transformations, right? So we have eliminated that need to go and just make weekly huge data extracts, which has everything and all the everything and all of it in it in one place and plug straight into your performance infrastructure. But also those wave apps have the ability to set your visibility to match an audience that you're preparing something for.

So that you you've built in the guardrails that you're not doing that. And I think we are generally speaking, seeing that if in the past, you know, tech teams were building applications and building software, those tech teams have been trained on guardrails, risk control. As this is being more democratized and we're seeing everybody vibe code from every function, the guardrails, the the sort of the the mindset of the risk has not caught up.

All right, so we talked earlier about how agentifying the transformation office frees it up to focus more on the human to human or change management aspects of the transformation. But why use agents to achieve that? Couldn't you just add more human capacity in the first place, or is there really an ideal size for a transformation office? And you can't make it much bigger than that.

Can a TO be too big? Is is maybe, you know, good, provocative question. I think the way I no, the way, the way I'd approach that is saying that the size of the TO should be driven by the number of work streams, stakeholders and interactions that TO has. And it it's very much sort of the coaching, and that is the big driver of more work, right? Making a bigger team to make more slides. You're gonna fall into the trap that you need people to read more slides, you'll have a team to, you know, you'll have to then grow a team to read the slides.

So very much on. If you think about the core function of the TO before you start expanding it, and whether that is distributed at the work streams or the BUs, or whether there is in the S in the TO with coverage to those work streams from coaching. That is very much the value added work. That is the, um, the work of, you know, through talking through the evaluation, problem solving, how to move initiatives forward, unblocking, um, problems in the delivery.

If you are just putting in more people to look at more data sources to generate more data, you know, fast forward, you're gonna need even more people in the future. Because the date, the data and the transformation only goes one way, there's always more of it. So as we as we looked at the as we looked at the agentification and we're back to that same sort of four person team that we looked at earlier in implementation, that team developed four main agents that they were using from materials creation, the admin, the checks.

And by doing that over over a period of time, actually in a in a period of time where the team was expected to get smaller. What they were able to do is one, cut down the team without traditionally, historically, they would probably just offloaded that work onto more more work for fewer people. They were able to offload a lot of that work into the into the agents, and they were able to free up an entire individual's worth of time.

And at the same time, they reduced more than that amount of capacity. And all of the functions which we you know consider the value adding functions, whether that is refilling the pipeline or thinking through, you know, the niche end cases of valuation. And you know what can be done about these, and how do you actually make these progress, and how do you make these work? Uh, and how do you refine business cases? They were able to do more of that and do more on the inside creation at the same time.

So do more of the valuable stuff, do less of the less valuable stuff, and and shrink the team at the same time. So

Chase, anything you'd add here,

Yeah, my my, my take on this is TOs, PMOs, IOs and IMO are generally pretty small and should be. They shouldn't be cottage industries. I think most folks intuitively understand that. Um, taking a small group and making it a little bit smaller doesn't really make a lot of money for your enterprise. That's not really where the value comes from. Where the value comes from is having that small group of people spend time on what matters, which is coaching, providing feedback, pushing for more value, doing interesting analysis.

That's the beauty here. I I look, if one takeaway from this conversation could be, hey, you could do your TO stuff with a smaller team. Yeah, sure, although I think that's honestly pretty, pretty limited in terms of value creation. The value creation here is at being a better TO.

indeed. And it gives those executives in the TO a much better handle on how things are going and a much better ability to connect with the individuals who are actually going to be driving the change, right?

and get you happier TO members, right? Like, let's let's be real, folks join a TO to drive change, not to make reports. They'd rather be doing that.

Absolutely. And so, Tomer, can you perhaps take us through a deep dive into some specific examples of how agents are being used in the TO right now? What does it look like in action?

One is the I talked about the daily control, control, quality control, right? Every night, and this has been trained, so refined. Every night we have an agent that runs on a transformation, that transformation plugs into your performance in the infrastructure, or you can give it a download. This was built originally, we just wave download, or with just data downloads. And it goes through all of the plans in in your transformation and says, Hey, what changed yesterday?

And if this changed, am I expecting other changes? And there's there are changes that we always know are interlinked. Is there a case where an element has changed and the the domino effect of follow-up changes did not happen? And can we actually catch that now? When pub somebody is still sort of in the mindset of, oh, I'm updating this now, I know what this is about. The same person is still on that, on that, you know, on that task at the moment and catching those early.

So and pro it then went through prioritized list on magnitude of changes. What is the number of follow-up effects we wanted, right? So, especially if you have dependent initiatives, uh, you can, you can tell your your transformation infrastructure will know the you know initiative. One moves two, three and four are dependent. Cannot start before one finishes. Uh, it will check all of those and really tee up the transformation office to be able to reach out to the individuals.

And it does, you know, a couple of things from also a change perspective. One one, it tells people that the transformation office is actually on top of it, that they care. It allows the transformation office to reach out to folks when they see a change before they have to explain a variance. So I'm gonna put myself in your corner, right? I will help you make that change before it makes it into the executive materials. Let me help you out here.

It creates a lot of goodwill between the transformation office and and the participants. And you know, this is this is the one where I'd I'd caution that, you know, you can actually create a lot of too much information for for the office. So very much one where you want to control the filters and the sensitivity and the prioritization. Because you don't want to, you know, pick up every two cents of change and say, go explain that.

So that that's one sort of good example where I think has been been probably the most well received. It was the one that created the most additional work up front for the team, um, one where they had to train it quite a bit to get good. But when they started seeing that secondary benefit of the kind of relationship it allowed them to create with the with the work streams, with the folks in the with the participants, actually created a lot of excitement within both the transformation office and the work streams.

That this is going on, this is happening

And Tomer, do you create that sort of filtering, if you will, up front, as you're agentifying the TO? Or do you sort of see the results come in and say, okay, we've got to dial this back because we're getting too much detail here, it's not helpful.

well in this. In this instance, it was very much dial it back, and in most instances we've, we've seen, it's been the dial it back. And transformations are, you know, each one's of a different size and different different magnitude. Um, one thing that we we built in very early was prioritization. So even if you have a list with a which is too long, you start going from the top. At some point you get into stuff which is meaningless and you stop.

So even if the data exists and you, you, you know, and the expectation is not that you check everything off the list every day, that can work. You don't have to spend all your effort playing around with filters.

And so can we talk now a little bit about creating all these PowerPoint decks. I I think everybody would love to hear how agents are helping with that aspect of the work.

So, de deck drafting is the first place that most teams go to. It's the biggest time constraint on a team. It's the one thing that when you get wrong, you get wrong. It's, um, if you look at the weekly work plan of every transformation office, this tends to be it's the data pull, analysis, draft, the deck that goes into it. It is the most repeatable, predictable, grunt worky work that a transformation office does. Go and take this data, put it in these materials, update the comments.

Um, and I'm I, I very are very purposely calling it drift, uh, deck drafting and not deck making. Um, the first part, before you actually start bringing the insight and the context and the analysis into the decks, is where these agents have been doing very well. Especially when you're in the back end of a transformation, you've got fifty weeks of decks to look at. You have a lot of data to look at. You can actually train an agent to do a pretty good job of doing most of the work up front.

It does still require you to go through every single page and every single insight. You will also like the teams get to know week over week where the agent does a very, you know, better job than not, where there are still holes in the in the insight. What it does very well, though, the teams have never done as well, is it will also go through and read with a fine comb, fine tooth comb. Last week's deck and the week before and the finance reports and your, uh, revenue revenue reports.

And it will flag, not necessarily change, it will flag where there are inconsistencies. And then as you as you start thinking about drafting the um, the prep notes for your chief transformation officer or for your um, executive, uh, members, participants in the meetings, you can actually start preempting, saying, hey, by the way, and this differs to that. Either we know why, and you should hear the talking points, or this may come up in the meeting. You should probably be able to talk that and explain it.

So not only does it draft it a bit faster and takes off a meaningful amount of the grunt work and the time, the work the team doesn't love doing, it actually does also that. Uh, back to the change management, create a lot of goodwill with executive participants, uh, specifically by pointing out to them where there might be quite a a need to do some explaining.

And Chase, if we call out this importance of keeping the transformation human, how do you ensure this automated drafting of the key outputs doesn't impact on that? In terms of TO members being really sure of their content and and truly owning it,

Most of these agents we're gonna talk about are pretty obvious, like it's obvious, hey, we should use AI to, like, make some of these decks we present. Like, Yeah, of course, you don't need McKinsey to tell you that, uh, you probably are already doing it. But what I suspect you're gonna encounter is your team members won't really know what they're saying. They'll be to that cognitive surrender because, ah, someone else wrote that. And so I think, like many of these agents in the TO, it's obvious you should use them.

It's not hard to configure them, but to make sure they actually create the value that we're talking about is relatively difficult. So, you know, how do we, how do we, how do we go about that? It's all about being thoughtful about creating human in the loop moments, right? When do we have a review? Who does the review? How does that review cascade through the members of the transformation office team? I think that's really critical to be thoughtful.

It's different in every context. I'm not there's not, there's not a five point framework for it. We can give you principles, but it is different in every context. But it requires real active thought. It's not, it's not the afterthought.

super. And Tomer, you mentioned earlier this idea of using agents to conduct consistency audits. Can you dive into that a little bit for our listeners and and what it actually looks like?

Yeah, I think we we spoke about this one quite a bit. Agents that are able to make sure that week over week things are are not dropped and we are consistent week over week. That if we add a action to follow up on something from three weeks ago, it actually shows up on the agenda. If there's somebody invited to the meeting, they're actually showing up there. Inconsistencies with MBR decks, it does require you to feed in quite a lot of this information, like depending on, like Chase said, sort of politics in the organization.

Does the TO actually have access to all of these different decks and functions? But when the teams do, there's a real unlock here that just prevents the derailing of conversations of, oh, you're saying this. But my chief revenue officer said something very different in yesterday's revenue review, or the CFO showed me a different number for this. Yes, you know, last week in the MBR. And you start having discussions about bridging information rather than about doing something about about gaps and misses.

And there's a real benefit to that here. I think we've we've already spoken quite a bit about sort of how that comes to happen, but effectively, just if you have the more information to feed it, feed it the information and don't automate the emails.

The consistency audit's a big one. This is where the most junior people on the TO spend the most of their time. And it's kind of this, this detective effort, the degree to which we can provide tools to make this not faster, cause, like that actually is to make this more insightful and accurate, that's really powerful.

So, before we wind up our episode, um, let's talk a little bit more about some of the key takeaways. Tomer, what would you wanna call out for our listeners as they think about agentification in their transformation office?

So, as you think about what you want to agentify, we've just had some examples like th- this is going to be true for everything, not just a transformation office, is understand what you're doing and how it fits in before you start doing anything. Don't automate for the sake sake of automation. So get get to grips with what are the actual workflows, what is the team actually doing, what, what is the the flow? As we've, you know, we we.

We had a chart, a chart up there saying, Hey, here is when you pull the data, here is this meeting that turns into that meeting. Actually have a good understanding of what that looks like. Set up the rules of what, what the agents can do and can't do, so that governance right before you start building the the actual agents. Secondly, as you think about roles, responsibilities, capabilities, orchestrators. We used to say that you want, you know, good project, project owners in the TR- in the transformation office, analytical capabilities.

So start rethinking that, right? So if you would like these folks to not just run, but also innovate and build. As you're staffing your TO team, even from the front, it's hard to come and staff data scientists to come and do the work and do it to the TO, right? All the success that we have seen has been organic, but it's honestly been fortunate that we've just had the right profiles in the right TOs. And they got to feel the pain, they got to see, sort of what is their grunt work, what does it take?

And then sat down and could actually reimagine what it would look like if they could, could automate, you know, quote unquote, for their own benefit, right? Invest in the change management, both within the TO, right? Like I said, you can actually create quite a lot of work and resistance in the TO and be upfront about it. Say that, you know, we're gonna do something and it's gonna be bumpy, or there's somebody on the TO wants to do something.

There's two or three other folks who are going to, you know, live with the consequences and be upfront in saying, you know, this is going to be a learning process and we're gonna have to iterate and we're going to have to see what we get. And the first one's gonna be clunky and we'll improve. And lastly, when it comes to the automated emails, when it comes to, as Chase said, if it's on the page, you made it, um, it, it's very much.

Make sure that you build that trust in the system, make sure that everything that comes out of out of your TO is of the same standard that you have been used to. Don't, you know, put slop out there, because if you start eroding the trust in the TO, you start eroding the trust in the transformation programs. And that that domino effect is very unhealthy.

Thanks, Tomer and Chase. Are there any particular AI platforms that are really well suited to agentifying the transformation office?

The vast majority of the stuff we observe in a transformation office can be done by open source models, and open source model's getting quite good. For me, it's less of like, hey, go use this model, that's actually not it. I think it's less, to be totally honest, what is the platform to use, cause they're all pretty reasonably good. And more about how you customize the workflow to work in your context. That's where I see failure modes, I don't see failure mode on, oh, we picked the wrong model.

The other thing, maybe just to point out, there is consistently using the same models. So things that were built for one solution and trained on one set of decks do well when you keep the same the same LLM running behind them. Actually, when teams have gotten, you know, well, we had we. We now have access to a newer, fancier LLM, have actually, um, seen a regression in what they, what they get. Because a lot of this is knowing what to expect, right?

You have planned your your, your workflow around. I'm gonna do this, the model's gonna do that, I'm gonna get X. If X changes materially, cause you've changed your model, you might actually go backwards in the time required it takes from you. So, as Chase said, the open source models tend to tend to do exactly what you need here. But also don't don't keep upgrading because you're gonna keep changing what you're getting as an output, maybe not what you expect to get.

So, before we end, can you share a story of the most inspiring outcome or impact that came out of your work with teams that agentified the TO of a client that you worked with? Maybe Tomer, we can start with you.

Yeah, f- for me, it was actually very much around the experience of the folks on the team, right? For for me, it was a team on, you know, year two of a transformation of implementation and having a, you know, having a a sit down over lunch with the team. And they're saying, actually, we're a lot more excited about the work that we're doing. We feel like we are doing better work because we are able to engage in stuff that, you know, this is why we joined the team.

This is what we were hoping for. And realizing that a lot of that is possible because they are not spending forty five minutes of every hour copy pasting materials from data dumps into presentations.

There you go, thank you, Tomer and Chase, you get the last word.

You know, my thought on agentifying a TO is similar to my thoughts broadly on agentification. You know, there is when we first started doing this, I thought to myself, Oh man, does this mean we no longer need all these junior people? We get rid of all these? No, that's not the case. That's not what we observe. We observe all these jobs change, but they change in ways that are cool. You spend less time making a Powerpoint deck and crunching, you know, figuring out some data irregularity, and spend more time coaching that initiative owner on how to make the procurement initiative more impactful.

That's cool. And so I I think what we found at all levels of our transformation office, from the most junior person all the way to the chief transformation officer, him or herself. The role is different, the tools are different. The underlying core job of changing humans is not different, but you go about it in a different way. And that way is generally more fun and exciting and allows people of all tenures to have more impact quicker.

So that that, to me, is my my down arrow on what the agentification has done for TOs. And I I suspect we see that in other areas of of corporations as well.

Wonderful, Tomer Chase, thank you so much for taking the time with us today. This is a great conversation

Thank you.

Thanks for having us.

and thank you to all of our listeners for joining us today. We hope you enjoyed the conversation and we welcome your feedback and ideas for future episodes. Just email us at itsr at mckinsey dot com. You can also share your ratings and reviews on any podcast player, with many thanks to all who've already done so. We appreciate all the comments and feedback we receive each week and encourage you to please keep them coming. And if you enjoyed this episode and you'd like to subscribe, you can easily follow our weekly series on any podcast player, where you can also access our entire library of more than 320 episodes.

Finally, if you'd like to automatically receive our latest publications and insights, we encourage you to visit our transformation practice page at mckinsey dot com slash mt, as well as our strategy and corporate finance practice page at mckinsey dot com slash scf. You can also connect with us on LinkedIn, and we've included links for all of those in the show notes today. Thanks again for listening. We look forward to having you join us again next week inside the strategy room.