Opening Keynote: The Golden Path to AI Value — Gartner IT Symposium/Xpo

[00:00:00] - We are at a crossroads. [00:00:06] This is the moment in history where people will look back [00:00:09] and say CIOs and AI leaders either made a mistake [00:00:14] or they helped put us on a path to greatness. [00:00:17] - Right now, there are two equal and opposite errors [00:00:20] that we can make with AI. [00:00:22] The first is to disbelieve in AI's transformative impact [00:00:27] because we haven't seen enough of it. [00:00:29] The other is to believe it too much, [00:00:32] to fall for the hype and visions of magic. [00:00:34] - Somewhere in between those two extremes lies greatness, [00:00:38] the golden path where value can be found. [00:00:42] Today, we will give you a roadmap to walk that path [00:00:46] and to find that value. [00:00:48] - Welcome to IT Symposium Expo. [00:00:58] I'm Alicia Mullery. [00:00:59] - And I'm Darryl Plummer. [00:01:00] - AI has shown us some amazing possibilities, [00:01:04] but when it comes to AI value, there is cause for concern. [00:01:07] In 2025, Gartner found the odds of an AI initiative achieving ROI [00:01:13] were one in five. [00:01:15] And the odds of an AI initiative achieving true transformation [00:01:19] were one in 50. [00:01:20] As AI slides from the peak of inflated expectations [00:01:25] towards the trough of disillusionment, [00:01:28] being trusted is critical. [00:01:30] Good news. [00:01:31] Gartner's latest C-suite survey finds that you, CIOs, [00:01:35] are the second most trusted C-suite member [00:01:38] in high-growth companies behind only the CFO. [00:01:42] But heroes are not made at the peak of hype. [00:01:50] Heroes are made in the trough. [00:01:53] CFOs have told us that 74% of you [00:01:57] are already getting productivity gains. [00:01:59] However, only 11% of organizations see clear ROI value. [00:02:04] Be a hero. [00:02:05] Take your AI use cases to the next level [00:02:09] because the road to value is not paved [00:02:12] with productivity wins alone. [00:02:15] And value? [00:02:16] Well, that's in the eye of the beholder. [00:02:18] In the private sector, value is growth. [00:02:20] In the public sector, it's mission success. [00:02:23] And all of you want cost reductions. [00:02:27] But is AI ready? [00:02:29] Are you ready? [00:02:30] Readiness is the key that unlocks that next level. [00:02:34] And here's the headline. [00:02:36] While not all AI is ready to deliver value, [00:02:39] humans are even less ready to capture value. [00:02:44] AI readiness doesn't mean, [00:02:46] can it pass the world's toughest test [00:02:48] or can it outthink a human? [00:02:50] It means, can it help you find value? [00:02:53] A lot of what's going on today isn't even about value. [00:02:56] It's just AI feeding on itself. [00:02:59] There's a word for that. [00:03:00] I'm going to let you say that. [00:03:02] In shitification. [00:03:09] Thank you. [00:03:11] Human readiness is about whether you have the right workforce [00:03:14] and organization to capture and sustain AI value. [00:03:18] 87% of employees are interested in using AI tools, [00:03:22] but only 32% are confident in leadership [00:03:26] to drive AI transformation. [00:03:28] So take a look at this. [00:03:29] In the beginning, AI and humans are not ready, [00:03:33] and this is where you should be highly skeptical of AI. [00:03:36] But when we cross over a threshold where both are ready, [00:03:40] we flip a switch into justified AI optimism. [00:03:45] The problem is, you are here, almost halfway up AI readiness, [00:03:50] but only one quarter of the way along human readiness. [00:03:54] Value is hard to achieve when you're so far out of balance. [00:03:58] AI readiness grows much faster than human readiness. [00:04:02] If all the vendors stopped innovating with AI today, [00:04:05] we'd still live years before we could catch up. [00:04:08] That's why it feels like we can't keep up. [00:04:10] For you to get value, you need to know where in this picture your use case falls. [00:04:16] Think of each use case as a step on the golden path. [00:04:20] This is your you are here GPS. [00:04:23] It helps chart your course on the path to value. [00:04:26] Wait a minute. [00:04:27] That's better. [00:04:29] The Gartner positioning system can only get it here. [00:04:36] Think of this as like navigating a map. [00:04:39] Following the GPS, you'll seek to find, capture and sustain AI value. [00:04:45] And if you're successful, you can transcend your limitations. [00:04:48] But before that, when AI and human readiness are both low, [00:04:53] value is in defending your position. [00:04:56] And that's primarily about augmenting employees. [00:04:59] Now this makes up nearly half of all AI investments. [00:05:04] And the problem here is misplaced expectations. [00:05:08] Your organization is expecting one type of value, but investing in another. [00:05:13] When AI readiness is low, it's difficult to see what's ahead. [00:05:17] It's like a fog has rolled in and you can no longer see the path. [00:05:21] But if you can navigate it, you'll find new sources of AI value. [00:05:25] Now, someone out there is thinking, come on, how can you keep saying [00:05:29] that AI is not ready? [00:05:30] People are getting lots of value from AI. [00:05:33] And that is true. [00:05:35] We've been using traditional AI techniques like neural networks and reinforcement learning for decades. [00:05:42] But Gen AI technical capabilities, costs and vendors are all of questionable readiness. [00:05:49] Let's start with technical capabilities. [00:05:51] Now, things such as search, collating data from multiple sources, content generation and summarization. [00:05:57] But we do those things with AI all day, every day. [00:06:03] But at this point in the path to find value, two points of interest should concern you: AI accuracy and AI agents. [00:06:10] Gen AI has an error rate of up to 25 percent, depending on the use case. [00:06:16] Now, yours might be a company that can live with that. [00:06:19] But what if you're the Federal Reserve Bank and require a 0.001 percent error rate in payments? [00:06:25] Gartner finds that 84 percent of CIOs and IT leaders don't have a formal process to track AI accuracy. [00:06:32] In fact, the top approach used today is human review. [00:06:37] But the human-in-the-loop equation is collapsing on itself. [00:06:41] AI can make mistakes faster than we humans can catch them. [00:06:46] Or worse yet, facts themselves can be distorted. [00:06:50] Gartner has just published a book on this called World Without Truth. [00:06:55] Sometimes you need a set of options, but sometimes you just need the right answer every time. [00:07:02] Like when I asked Bjorn, my partner, "How do I look?" [00:07:05] There is only one right answer. [00:07:10] You need to bring your own accuracy. [00:07:13] Let's call it your accuracy survival kit. [00:07:16] In it, you'll find formal metrics, which IT uses, like a comparison metric where you test AI output [00:07:23] against an established norm. [00:07:25] Two-factor error checking, which every employee should too, where you get one AI model to check another. [00:07:32] And the good enough ratio. [00:07:34] This is a measure of when AI accuracy is just good enough for your initiative. [00:07:38] And by the way, good enough may be harder to achieve than you think, because we hold AI to a higher standard. [00:07:46] Another must-see point of interest. [00:07:48] AI agents are at the top of the hype cycle. [00:07:51] Currently, 17% of CIOs report that their organization has already adopted AI agents. [00:07:58] And an additional 42% are planning to adopt within the next 12 months. [00:08:04] But not all agents are created equal. [00:08:08] Globally, 88% of IT leaders are focused on conversational agents. [00:08:13] And conversational agents are ready to have conversations. [00:08:17] But if you need them to make decisions, and you should, then conversational agents are not ready. [00:08:24] Decisions and reasoning are necessary for autonomous multi-agent systems. [00:08:28] So, remember the first time you called a customer support line and they kept going through a script? [00:08:35] No matter what you said, they just kept repeating the same thing. [00:08:39] Did you check to see if it's plugged in? [00:08:42] Did you restart? [00:08:44] Is your monitor on? [00:08:46] That's what it feels like when agents can't make decisions. [00:08:50] Using agents to handle conversations is missing the point. [00:08:54] For example, retailers are not looking to conversational agents as the next big AI win. [00:09:01] They need something more ambitious. [00:09:03] More ambitious than automatically ordering clothes online. [00:09:07] Or even doing a customer support call. [00:09:09] They need agents that can watch what customers are buying, [00:09:12] trigger multiple RFPs for replenishment, negotiate terms and conditions, [00:09:16] and decide the best offer for their company. [00:09:20] That is multiplex B2B negotiation. [00:09:24] And conversational agents don't do that. [00:09:27] Don't settle for conversations. [00:09:29] Demand reasoning and autonomous decision-making. [00:09:33] And AI agents should handle specific jobs to be done. [00:09:37] For me, I desperately want an agent that builds my presentation for me [00:09:42] while I talk to it. [00:09:43] That's an agent acting as an expert in graphic design. [00:09:47] And what if that agent was an expert that could do building approvals? [00:09:51] Or do my taxes? [00:09:52] Or the company's taxes? [00:09:54] As AI agents, as experts, can create real-world value. [00:10:00] They'll do it as long as you're clear on what you need from the technology. [00:10:04] And these technologies aren't cheap. [00:10:06] So we have to consider the true costs. [00:10:09] CFOs tell us they are losing track of what is being spent on AI projects. [00:10:14] They knew the cost on day one. [00:10:16] They don't know the cost on day 100. [00:10:19] 74% of organizations are barely breaking even or even losing money on their AI investments. [00:10:26] But if previous tech introductions had upfront transition costs, AI has a transition mortgage [00:10:32] that you just keep paying. [00:10:35] Why? [00:10:36] Well, when you implemented ERP, you licensed the software and trained your people on one kind of work. [00:10:42] But with AI, the work and the training keep switching contexts. [00:10:48] First, you learn to summarize emails. [00:10:50] Then it helps you do data analysis. [00:10:52] And after that, you're creating compelling videos. [00:10:54] You're using one model before lunch and a different one after. [00:10:59] You need people to have experiential knowledge. [00:11:02] Connect them with peers and show them what good looks like. [00:11:06] Because most people are just discovering what they can do, [00:11:09] not being trained on what they should do. [00:11:12] Here's the bill. [00:11:13] AI implementation costs average 1.9 million per company. [00:11:18] And that's just the bill on day one. [00:11:21] But then you need to invest heavily in AI training days and literacy programs. [00:11:26] Because for every 100 days of implementation, AI adds 25 more days to train staff. [00:11:33] And worse yet, change management costs add another 100 to 200 days. [00:11:39] That's up to 200% more effort. [00:11:41] Next on the bill is ancillary costs. [00:11:45] For every AI tool that you buy, there will be 10 ancillary costs you didn't anticipate. [00:11:50] Costs such as managing access credentials for autonomous agents, acquiring new data sets to ground AI, [00:11:58] or packing your accuracy survival kit. [00:12:01] There's a hidden transition cost to all of that. [00:12:04] So what do you do? [00:12:05] For every AI tool you buy, anticipate 10 hidden costs. [00:12:10] Conduct an analysis and decide which costs you'll fund. [00:12:13] The last thing you want is to be the unintended owner of a negative ROI business case. [00:12:19] And you still need to choose the right AI vendors for your needs. [00:12:23] Now, if choosing an ERP vendor was like getting married, [00:12:28] then choosing an AI vendor is more like getting married, having triplets, and moving to a different country. [00:12:35] It's complicated. [00:12:39] Just like kids who touch everything with their little hands, AI is touching everything. [00:12:45] And here's why. [00:12:46] When marrying an AI vendor, their models are like your children. [00:12:50] Children that need to be fed the right data to grow, constantly reeducated and grounded. [00:12:56] And those children, well, they take on your ethics and the ethics of the vendor you're married to. [00:13:03] And it's not just about the kids. [00:13:04] You have to pick the right country to move to. [00:13:08] In the AI race, those vendors are starting to resemble countries to live in. [00:13:13] We call them digital nation states. [00:13:16] A digital nation state is a vendor that controls land, power, water, talent, and capital to rival those [00:13:24] of actual nations. [00:13:26] Large vendors spend more on AI infrastructure per quarter than the annual GDPs of 47% of the world's [00:13:34] countries. [00:13:36] If you're planning a massive rollout of AI to your enterprise, bet on the major hyperscalers. [00:13:42] These are players like Microsoft, Google, Amazon, Alibaba, and Oracle. [00:13:47] They are superpowers, sometimes with government support and always with massive ecosystems and AI [00:13:54] infrastructure scale. [00:13:56] If you want industry specific use cases, partner with startups and industry leaders. [00:14:01] The combination of innovation and domain knowledge comes with access to new markets, customers, [00:14:07] and data. [00:14:08] And this is more like a developing country. [00:14:10] And this is more like a developing country. [00:14:11] Bet on wildcard vendors such as OpenAI, Anthropic, Meta, DeepSeq, and Misral AI. [00:14:18] That's if you want leading edge capabilities. [00:14:21] They don't have the scale and reach of a superpower yet, but their influence is rapidly growing. [00:14:27] You should note, these are newly minted nation states, and their behavior can shake the ground under [00:14:33] everyone else. [00:14:34] They're fully innovation ready, but not fully enterprise ready. [00:14:39] As a potential AI bubble builds, there are a lot of circular deals being made with digital nation states. [00:14:47] For vendors, these deals can be great because they create money seemingly out of thin air. [00:14:52] But with all the heavy investments, some will never recover that money in real revenue. [00:14:59] And if the attention that they have is so focused on AI, how much attention is left to focus on the [00:15:05] products that you already own? [00:15:08] So don't select a vendor mostly based on its high valuation or just its AI technology. [00:15:13] Select it based on its ecosystem of partners and their shared ability to deliver game-changing [00:15:19] solutions or outcomes to you. [00:15:22] That finds value regardless of whether there is a bubble or not. [00:15:27] Picking a vendor to marry is hard. [00:15:29] So here's an easy button for you. [00:15:31] If you're going to select an AI vendor today, you had better select one that is good with AI agents. [00:15:37] Gartner created an agentic compass that matches vendors with your requirements. [00:15:43] It learns your use case characteristics, maps them to AI capabilities, [00:15:47] and then points you to the AI vendors who provide them. [00:15:52] We'll put vendor selection on the map here. [00:15:55] But bear with me a moment because there's one more thing about vendor selection [00:15:59] that should be important to everyone here, and that is AI sovereignty. [00:16:03] Now, most of you are used to dealing with data sovereignty, [00:16:07] but with AI, a lot of data is hidden behind the models. [00:16:11] To ensure sovereignty, you can't just protect the data. [00:16:14] How do you even know what sovereign data has been trained into a model? [00:16:18] You have to protect the model, and the results the model generates, [00:16:22] and who has access to those results. [00:16:25] And guess what? [00:16:25] You can get locked into a model or locked out of a relationship. [00:16:31] The US, China, and the European Union are building sovereign AI solutions with proprietary models [00:16:37] that they control. [00:16:39] By 2027, 35% of countries will be locked into region-specific AI platforms using proprietary, [00:16:46] contextual data. [00:16:49] Now, if you're a global enterprise, that could be very disruptive. [00:16:52] And even if you're not a global enterprise, you might get locked out of a long-term partner relationship. [00:16:59] So here's what you do. [00:17:00] Build, buy, or steal. [00:17:03] Don't steal anything. [00:17:06] Acquire a digital tokenization solution to anonymize data so the real data doesn't leave your shores, [00:17:13] even inside a model. [00:17:14] So here's what we're going to do. [00:17:16] We're going to do a lot of data. [00:17:17] We're going to do a lot of data. [00:17:18] We're going to do a lot of data. [00:17:19] We're going to do a lot of data. [00:17:20] We're going to do a lot of data. [00:17:21] We're going to do a lot of data. [00:17:22] We're going to do a lot of data. [00:17:23] We're going to do a lot of data. [00:17:24] We're going to do a lot of data. [00:17:25] We're going to do a lot of data. [00:17:26] We're going to do a lot of data. [00:17:27] We're going to do a lot of data. [00:17:28] We're going to do a lot of data. [00:17:29] We're going to do a lot of data. [00:17:30] We're going to do a lot of data. [00:17:32] We're going to do a lot of data. [00:17:34] We're going to do a lot of data. [00:17:36] We're going to do a lot of data. [00:17:38] We're going to do a lot of data. [00:17:39] We're going to do a lot of data. [00:17:41] We're going to do a lot of data. [00:17:43] We're going to do a lot of data. [00:17:45] We're going to do a lot of data. [00:17:46] We're going to do a lot of data. [00:17:47] We're going to do a lot of data. [00:17:48] We're going to do a lot of data. [00:17:49] We're going to do a lot of data. [00:17:50] We're going to do a lot of data. [00:17:51] We're going to do a lot of data. [00:17:52] We're going to do a lot of data. [00:17:53] We're going to do a lot of data. [00:17:55] We're going to do a lot of data. [00:17:55] We're going to do a lot of data. [00:17:57] We're going to do a lot of data. [00:17:59] We're going to do a lot of data. [00:18:01] We're going to do a lot of data. [00:18:02] We're going to do a lot of data. [00:18:03] We're going to do a lot of data. [00:18:04] We're going to do a lot of data. [00:18:05] We're going to do a lot of data. [00:18:06] We're going to do a lot of data. [00:18:07] We're going to do a lot of data. [00:18:08] We're going to do a lot of data. [00:18:09] We're going to do a lot of data. [00:18:10] We're going to do a lot of data. [00:18:11] We're going to do a lot of data. [00:18:12] We're going to do a lot of data. [00:18:13] We're going to do a lot of data. [00:18:14] We're going to do a lot of data. [00:18:15] We're going to do a lot of data. [00:18:16] We're going to do a lot of data. [00:18:17] We're going to do a lot of data. [00:18:18] We're going to do a lot of data. [00:18:19] data is a sovereignty decision. [00:18:21] So track the rise of digital nation states. [00:18:24] We've just gone through the fog of finding value, which answers the question, can this be done? [00:18:30] When human readiness is low, we must ask ourselves if we can capture and sustain value. [00:18:36] Not just can we do it, but can we keep doing it? [00:18:39] The golden path here is littered with obstacles, which will slow you down. [00:18:44] Our people alone can be a big obstacle. [00:18:47] I live in London and every Sunday, Bjorn and I love to walk to the farmer's market. [00:18:53] I just want a nice stroll, but he is always in a rush to get his favorite pastry. [00:18:57] And this is what low human readiness feels like. [00:19:00] It's like, come on, Alicia, hurry up already. [00:19:03] If human readiness is low, it will slow you down. [00:19:06] In fact, 71% of CIOs and IT leaders report that their workforce is not ready for AI. [00:19:13] Why? [00:19:14] Because AI unleashes a toxic mix of a steep learning curve and the primal fear that AI is going to replace us. [00:19:22] Headlines everywhere are claiming that AI is taking jobs and replacing staff. [00:19:27] The Anthropic CEO says AI could wipe out half of all entry-level white-collar jobs in the next five years. [00:19:35] And 60% of CFOs are expecting they can reduce headcount due to AI. [00:19:42] Those kinds of statements make it seem like there's an AI jobs bloodbath going on. [00:19:48] But is that true? [00:19:49] Nope. [00:19:50] We find that only 1% of headcount reductions are directly due to AI today. [00:19:55] Let me say that again. [00:19:57] Only 1%. [00:19:58] It's not about job loss. [00:20:01] It's about job chaos. [00:20:03] And it's your job to settle the chaos. [00:20:06] In fact, job and role redesign is a 20 times bigger effort than layoffs or hiring. [00:20:13] Now, yes, people will lose jobs, and that matters to us all. [00:20:18] But it's not a bloodbath. [00:20:20] What we do see is hiring restraint for junior-level jobs. [00:20:25] Senior staff can now delegate to AI much of the work that juniors used to do. [00:20:30] That captures value. [00:20:32] But the problem is, we keep getting distracted by headlines. [00:20:36] You may have read news recently about Salesforce eliminating 4,000 jobs. [00:20:41] What didn't make the headline was that while they are cutting customer support staff, they are also hiring staff to support net new AI revenue streams. [00:20:52] This is a talent remix. [00:20:54] And it seems like all the big tech providers are doing it. [00:20:57] Shifting talent from lower performing business units to higher potential AI-driven business lines. [00:21:03] Unless you're a tech company, this strategy isn't your strategy. [00:21:08] Your ability to set up whole new lines of business is limited, so your returns from copying this strategy will be limited. [00:21:16] Instead, your strategy needs to be a value remix. [00:21:21] Look at how AI can cut your backlog, how it can reduce fraud, or how it can grow revenue through human empathy. [00:21:28] Like Sanlam, the South African finance company whose customers ask for loans to pay off a rising debt. [00:21:35] Sanlam uses AI to empathize with customers and help them accept the idea of a debt reduction plan instead of a new loan. [00:21:45] The value captured here is not from headcount reduction. [00:21:48] It's from less bad debt and healthier customers. [00:21:52] And guess what? [00:21:53] All of you have a human empathy opportunity somewhere in your organizations. [00:21:58] But almost none of us are thinking about it as our next big AI breakthrough. [00:22:03] Tell your CFO that you need a value remix, not a talent remix.