Sponsored by: Deloitte | Driving Enterprise Transformation with Agentic AI

Summary

Deloitte's China Widener argues that the real barrier to enterprise AI ROI is not the technology itself but knowing where to start, since most AI pilots are not built to be enterprise-grade from the beginning and therefore cannot scale.
This video introduces Deloitte's AI Navigator framework, a three-module approach covering task-level value identification, workflow and operating-model redesign, and use-case selection, designed to move organizations from fragmented pilots to sustained enterprise-wide AI transformation.
Achieving enterprise-wide agentic deployment requires aligning the CHRO, CIO, and CFO on a redesigned operating model, because AI agents fundamentally change how workflows function rather than serving as drop-in replacements for existing processes.

Chapters

FAQs

Why do enterprise AI pilots fail to generate ROI?

According to this video, pilots fail to generate ROI primarily because they are rarely constructed from the start to be enterprise-grade later. Organizations focus on solving a specific productivity problem in a pilot, but the design choices made for that narrow scope cannot be extended to enterprise scale, and the problem being solved often turns out not to be an enterprise-level problem.

What is Deloitte's AI Navigator and how does it work?

AI Navigator is Deloitte's framework for agentic AI deployment, structured around three modules: task-level value identification (pinpointing where agents create value), workflow designer and operating-model redesign (rethinking how work gets done), and a use-case hunting module that identifies the right starting points. The framework is designed to connect AI investments to trusted data and measurable ROI from the outset.

How do AI agents change enterprise workflows differently from traditional software?

This video emphasizes that AI agents are not a drop-in replacement for existing tools—they require the operating model itself to change. Unlike software that automates a defined task within an existing process, agents reason and act across workflows, meaning the human roles, approval steps, and accountability structures around those workflows must also be redesigned.

What executive alignment does Deloitte say is needed for agentic transformation to succeed?

Deloitte describes a case study in which the CHRO, CIO, and CFO must all be aligned on the agentic transformation vision before enterprise-wide deployment can succeed. This alignment is needed because agents affect talent (CHRO), technology infrastructure (CIO), and the financial case for investment (CFO), and misalignment at the board level is one of the key reasons transformations stall.

Full transcript

[00:08] Good afternoon. It seems that we are live and started cuz my countdown clock has started. So, first of all, Tina Wadhwa with Deloitte, and we're going to spend the next 19 minutes and 42 seconds having a conversation about what it means to take AI to the enterprise. More
[00:22] importantly, what we're really going to spend time talking about is how to actually effectuate getting AI to the enterprise and all the things that are involved that actually aren't about the technology, though the technology is
[00:34] incredibly important. What we know from the work that we've been doing in the studies and surveys we've conducted over 2025 is that it's actually not the stopper. The stopper are all the other things. And so, we're going to talk a bit about
[00:47] that. The other thing we're going to talk about cuz I'm going to keep you interested for at least these next 20 minutes is this idea that cattle and AI at the enterprise have something in common, and we're going to talk about that, too.
[00:59] Now, you might be thinking, "Cattle, really? I'm a cattle rancher when I'm not doing this. Um and so, the reality is some of what actually has prompted this was some work with that cattle, and that's the connection we'll talk about.
[01:12] So, let's talk about this in terms of what we know. We've all seen the headlines. AI is full of promise. People haven't seen it in the bottom line and their ROI. I had a CFO say to me, "It is everywhere except
[01:24] my bottom line, and I don't know why, and I need to get it there cuz I got my CEO, I got my board, and I got the street all asking me the same set of questions, which is when will I see the
[01:36] ROI?" The reality is when you take a step back, the real problem has been where to start. Now, I have no doubt everybody in sitting in one of these chairs right now has has a pilot somewhere in your
[01:48] organization or several or a hundred. There've been lots of pilots because organizations have done this in lots of ways. But, the challenge is pilots are almost never constructed from the start
[02:02] to be enterprise grade later. And the problem is when you focus on the productivity only and you focus on it in a pilot to solve a specific particular problem, getting it then to be
[02:14] enterprise grade and ensuring the problem you're solving for is actually an enterprise problem is where it starts to break down. So we took a step back and said, "So how do we then deal with this?" There's a lot of smart people. The way we are all
[02:27] educated and entertained in the world is through technology today. So why are we stuck on this particular thing? And it turns out the being stuck part is really a function of a couple of things.
[02:39] Not being clear about why you're identifying. Not being clear about where you should identify. Lacking clarity on the actual value you're likely to get because you
[02:52] actually need to do some real work on the front end of that to appreciate the value. And then the last two pieces of the puzzle are what's my new world look like because the reality is if you're going to identify, you're literally
[03:05] going to change the workflow that you have today because an agent doesn't act like a human. Which means the workflow you have is insufficient and ineffective. And if you
[03:17] only try to identify at the edge, you will not get the outputs and the outcomes that you want. The last part of the puzzle and so this is interesting and you might find this to be an interesting fact. Two things we
[03:29] determined to be true. One is upwards of 80% of companies today have said their their their communication to us through survey is that they either are on the beginning edge of their agentic strategy
[03:42] or they don't have one at all. So think about it. Nobody's got a map. That's problem. The second thing that was an interesting statistic coming out of the a similar survey is that 60 upwards of 65% of co-
[03:56] companies have said, "And oh, by the way, even if I had that strategy, my infrastructure ain't ready for us to be an agentic organization." So, the last piece in the puzzle is how
[04:08] do you solve for that? You infrastructure decisions, you have technical debt, you have a direction and a path and a road map. That evolution has to be accounted for. And then now let's bring in something that's sort of
[04:21] risen in the last 6 months, which is And oh, by the way, we got tollgating, we got tokenization issues, and all of that together has caused many organizations to be stuck and to be stymied in their movement. So,
[04:36] the beauty is what we're going to talk about in AI Navigator is how you get unstuck, and how end-to-end you have a disciplined process that allows you to approach all of that in an asset
[04:48] predicated on data. What I've said to clients when I talk to them is, "I'm going to move you from the art of what's possible to the art of what's probable in a way that you can stand up to all your stakeholders, whether that's the
[05:01] board or the street, and say you know why you expect the value you're going to get, and you have data that supports it's the return you should see." So, with that, let's talk about it. And let's see if I can get the uh clicker to
[05:13] click. All right. So, we've talked a bit about the why, this notion of arriving at having a greater impact, this idea of what the future state is, because the underlying and underpinning in all this
[05:27] is you are going to impact the humans in your organization. If those humans don't have trust, they are going to do what humans from time beginning have done, which is cross their arms and say, "I do
[05:39] not trust what I do not understand. I am not moving." This is no different, and if if you have been in a relationship anywhere in your life, you know that you can't solve an emotional problem with an objective conversation. You have to meet people
[05:53] where they are. That's why it matters to be able to reshape where people are headed and what the future looks like. So, what are the four modules that are involved?
[06:06] I need to stand here to make that work. You can't get second one yet. The first module is what we call AI identifier. It literally looks you can do this by capability. You can look by BU. You can look across the enterprise and it allows
[06:19] you to identify what areas are optimized through agentification. You don't have to guess. You'd actually know. Because in this module, you have the ability to
[06:31] see at the task level of the human roles that you have, what parts of those tasks are subject to agentification. So, let's assume you have 100 people who
[06:44] are doing a task that or doing a role. That role has 62 tasks associated to it and in the description that you have, you actually will be able to know that of those 62 tasks, 30 of them are
[06:56] actually subject to agentification and have a high fit and high probability. What that tells you is really where to look. If you will, it's where to hunt. I'm from Texas, we do a lot of hunting.
[07:08] The notion of what part of the forest you want to hunt in for the things that are agentifiable. The second thing that that allows you to then understand is what resource level you're going to free up both in terms of
[07:21] the human component because you understand what percentage of task you're taking away as well as understanding the OPEX cost associated to those roles predicated on those
[07:34] tasks. So, you actually understand the value side of the equation at the task level and the human side of the equation as well. That's game-changing because that's what actually allows you to get to
[07:48] impact analyzer, which allows you to see in a single view through a heat map version, what is identifiable by fit and what is identifiable by value.
[08:00] Because depending on your organization and your reason for prioritizing, the reality is you might choose high value targets that are hard, but ultimately you are driving a value
[08:12] proposition. You might determine that you're not going to take on the things that are highest value out of the gate, but you really want to understand the greatest fit because what you're trying to communicate to your organization is
[08:25] success rate and speed. And so high fit is going to give you speed, which creates buy-in, which creates trust. High value is going to be harder potentially, but ultimately
[08:38] reap a better return. The beauty is you get to see both of views in a heat map so that you can decide and understand when you decompose a capability or decompose the functions within a BU,
[08:51] where you want to spend your time and focus. That's how you get from what's probable to from what's possible to what's probable because you have the data to actually support and understand that nuance. So that's important. The
[09:04] third module really takes you to workflow designer. Now this is where it gets to be interesting and fun because before the first agent's coded or series of agents are coded, you
[09:16] actually understand both the agent's responsibility and the human's responsibility in the new process. Nobody has to guess. You ultimately actually with this tool
[09:28] get a actually the BRD documents that you would need to inform the building of the agent comes out of this module, but it's predicated on the future state, not the current state. Not what people think
[09:40] should be in the future state. You literally actually see, here is the agent's functionality, scope of responsibility, and here's the humans. You understand where the handoffs are. You understand where the engagement is.
[09:53] What that does for you is it allows people to move out of that place of I'm concerned about the job I'm responsible for and start to understand how the agent enhances the job they'll do next.
[10:06] It also allows you from an HR perspective to think through, how do I redefine the role now that I understand what's agentified, and therefore how do I then redefine the op model that comes
[10:19] along with it? Because the reality is if you're going to agentify at scale, you are going to ultimately change the role responsibilities and tasks of the individual, and you are by extension
[10:31] going to change your operating model because the operating model only matters in so far as it's in service to the work that you're trying to do. All of this matters in the context of, is this about cost?
[10:43] Is this about a new customer experience, a new service, or a new product? Or is this about giving your workforce a new way to work? All those things matter, and all those things set the stage for
[10:56] what's going to be more important at which part of your journey, depending on the arc of where you are. So, I recognize some of you right now are saying, we have a clear moral map.
[11:08] That's great. You can use this and validate the map that you have in a couple of weeks. For those of you who maybe are part of that 80% that don't yet have a clear uh strategy around what
[11:20] AI at enterprise really looks like, or you might have a clear vision, maybe leadership hasn't quite come along yet, or maybe there other stakeholders that you actually need to convince, it allows
[11:33] you to allow people to understand and see at a touch and feel level so that everybody understands what they're buying in the process of the decisions that are being made. And if we actually take on a little bit
[11:47] of data from your organization, which we have the ability to ingest, so this comes to you predicated on all the information that we have, public, proprietary, and purchased data sets.
[12:00] So, showing up, I can tell you where to hunt even before we have any other conversation. If you're willing to exchange a bit of information, opex financial information, HRS data information, not any of the
[12:14] uh uh privacy data, but just role task level data to understand how you sorted the roles that you have, I can tell you where you're going to actually have the greatest impact from the beginning.
[12:27] And then with a little bit of input, it becomes very customized to your organization. Because obviously you made data you made you made spend decisions, you made HR decisions, you made structural decisions
[12:40] that can have an impact. The beauty is when we're here, what you actually know is the future state you're driving to. And you know that with a level of clarity that allows you to get all of
[12:52] the sign-on that you're going to need in order to move forward productively. The last piece of the puzzle is, well, if I know these things, I understand where I'm going, I get the value I'm going to get, I understand my opex implication, I
[13:06] understand my HR implication, and I now know my new workflow, agents and agentic solutions do not spring forth from the ground, we all know that. They actually have to be built, leveraged, or bought. And so, the question is how do you do that?
[13:20] And that's the last piece of the puzzle, in the agent studio. This idea of this is where we start to understand what is in your tech stack. What is the goal set that you have? What are the decisions you make? What ERP do
[13:33] you have? What CRM do you have? Because all of that now, as we all know, everybody's bringing an agent to the party. The question is, once I understand what agents I need for the things I've chosen to agentify, do I
[13:45] build it? Do I buy it? Do I leverage it already? Given Given the technical decisions that I've already made in my infrastructure. It allows you to understand all of that and make those decisions as you move forward.
[13:59] The last piece to this that matters more than anything else from my perspective and what I found clients have been most excited about, this is where I come back to my cattle conversation at the start, which is this.
[14:12] So, when you run cattle, every spring and every fall, I have to get all the cattle through the chute. No matter what treatment those cows need to have, every cow has to come through the chute. So, whether I'm doing
[14:25] castration, giving shots, whatever the case may be, every cow comes through the chute. And the reason for that is the chute treats the cow the same way every time. It allows me to hold the cow in the
[14:37] right position each and every time to conduct whatever business I need to conduct. This is the chute. So, no matter nobody agentifies overnight. It is a multi-year journey for every organization there is, unless
[14:51] you happen to be native. Other than that, if you have any historical debt, if you have been around for more than 12 minutes, you have a lot of things you carry with. You are not going to be able to make these changes overnight because you have processes,
[15:04] you have structures, you have an infrastructure, you have applications. You may have 3,000, you may have 300. All of that has to be addressed as you're agentifying at components of the solution. The reality is you need every
[15:19] part of the business as you're making the decision to go through the same disciplined evaluation. That is what this is. It is the chute to my cattle, well, it is the AI chute. It
[15:32] allows you, whether it's year one or year two or year three, to come back and ask the same set of questions through every stakeholder every time to evaluate the value the same exact way. If you
[15:46] happen to be at a very heterogeneous organization right now, meaning your BUs are really self-contained, maybe you have multiple CIOs that roll up to a central CIO,
[15:59] maybe you have multiple CTOs, the reality is everybody's making a set of decisions predicated on their individual thought process. This standardizes the question, not the answer.
[16:11] That's important in a heterogeneous organization because everybody believes they are unique, and they are to some degree, but the questions you ask, the value you attach to them, the standard
[16:23] you create is all embedded. So, the evergreen nature of this asset allows you, no matter when you start, whether you say, "I'm going to start with a capability like finance or marketing." And we have clients who do that. Have
[16:36] clients who have said, "I'm going to start with a whole business unit." And then I'm going to go to the next business unit. Some have said, "Look, I want to solve for this one enterprise problem to start." Regardless, what it allows you to do is to to actually
[16:50] evaluate every time the same way consistently, so that whatever issues arise, whatever questions the stakeholders have, you're actually being able to answer them in a anchored, disciplined, data-driven, and
[17:02] informed way, so that if there are issues, they they should be issues that occur in a in a repeated way, so they can be solved for once. You can understand the value standard
[17:15] consistently, so that value in business unit A equals value in business unit B so that the organization as a whole is very clear on what its outcomes are. That's the beauty. It is evergreen in
[17:28] its nature and because we have the ability to ingest data, um, quite honestly, specifically when you get the workflow designer, we can ingest data in almost every form except
[17:40] video. So, if you've documented a process and any other thing, BPMN, it's just on in some Excel spreadsheet Excel sheets in a table, like we can ingest all of that so that it becomes super
[17:53] clear and customizable to you. We can ingest the financial data in a way that makes sense as well as the HRS data. All of that changes from what happens from an industry perspective cuz not every
[18:06] industry is impacted the same way to what is happening in your organization. So, all that said, it boils down to this. The journey is real. The struggle is real and we know that
[18:19] because those are the those are the conversations that we're having in every boardroom across America. I tell you that I sit on our US board. We're having and have had the same set of conversations. This kind of asset
[18:33] changes that conversation. It allows you to respond in a way that is clear and defensible and you can actually effectuate a plan and inform your strategy for it because we are no longer
[18:47] in a place where the technical architecture and strategy is separate from the corporate strategy. And every C-suite knows that or is becoming to understand that in ways they might not have expected before. And I
[18:59] will tell you a story in my last 1 minute just for context. We were serving a client and this is interesting and it is absolutely true and I will protect the I will protect the innocent by not
[19:11] naming the who, but here's the scenario. We showed up to have this conversation. We started having the conversation with the CIO who went No time, not interested. I got this. I'm good. Okay. Started having this
[19:25] conversation with the CHRO. CHRO said, "Huh. I got to solve for this problem. I got people I got to reskill. I got people who I need to move around given where we're headed from a from a strategy
[19:38] direction." Ultimately, the CHRO embraced this concept and this use. The CIO is no longer the CIO of the organization, and the new CIO coming in now reports to the
[19:50] CHRO. Because the reality is they were able to move fast. This takes 40% less time to do than any traditional approach we have ever done before.
[20:02] So, the reality is you can be complete, you can be comprehensive, and you can be quick. With that, my time is up. I appreciate your time, and with that, thank you.
[20:14] >>

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.