As we head further into the generative AI era, many enterprises are finding themselves in a bind. On the one hand, they believe this technology is inevitably part of their future, because its abilities are simply so extraordinary. Generative AI can do things we once imagined would remain the preserve of humanity. These models can reason across unstructured data, understand natural language, handle open-ended tasks and create new content: everything from text and imagery to audio and video. They can write engineer-grade computer code, pass legal exams and serve as general-purpose thinking companions. Five years ago, the idea of brainstorming a pitch with AI would have seemed like science fiction. Now, it’s par for the course.
On the other hand, many organizations are struggling to realize value from generative AI—at least in terms of traditional metrics. Despite the trillions of investment dollars flowing into the space, a recent study shows that only 41 percent of leaders have seen positive ROI. Those that are struggling tend to fall into two groups. Some are simply in shiny object mode—enthralled by the possibilities but deploying non-strategically and, too often, helplessly stuck in pilot purgatory. Others are deploying at scale but unable to translate the benefits into clear revenue gains.
For many, the sheer pace of change in this space is a challenge in itself—with new models, market entrants, and use cases all developing at an unprecedented rate. “There is enormous excitement and a desire to adopt this technology quickly,” says Patrick Gormley, Senior Vice President and AI Practice Lead at IT services provider Kyndryl. “However, for generative AI to deliver the benefits it promises, it must be built on robust digital foundations and integrated into an enterprise’s holistic data strategy.”
The companies who are faring best with generative AI have realized that it requires new ways of thinking about enterprise tech adoption. “This is not just a technology like ‘cloud’,” says Dr. Shaun Barney, Global Data and AI Consulting Partner at Kyndryl. “Generative AI opens the door to create innovative use cases that can support organizational goals—from new ways to delight, engage, and retain customers, to opening up new revenue streams. However, this technology must also be coupled with curiosity, creativity and the necessary skills for it to reach its potential.”
In the report below, we explore four new ways to think about your AI strategy to help you get ahead—and stay ahead. But first...
Since the first wave of machine learning crashed over the economy a little over a decade ago, AI has mainly been seen as an efficiency tool. A way of improving the bottom line: automating tasks, optimizing processes, and saving money. When people talk about the value of AI—the benefits businesses can derive from it—this is traditionally what they have meant.
Generative AI can certainly help with those things. Coding assistants can help engineers complete tasks more than 50 percent faster, and agentic AI—which can complete multi-stage tasks independently—is effectively robotic process automation on steroids. “But focusing exclusively on cost savings is kind of narrow minded,” says Erik Brynjolfsson, a Stanford professor who runs the university’s Digital Economy Lab, and recently co-founded the company Workhelix to help companies identify and measure AI’s benefits. “Competing on price alone is rarely a sustainable competitive advantage, and it is short-sighted about the technology’s true potential.”
Generative AI has other forms of value to offer beyond simple efficiency. It is effectively a supply of cognitive capacity: it can help solve problems, enhance creativity, and reason across complex information. Smart companies are therefore recognizing that once the basics of a successful implementation are in place—the right data foundations, deployments that scale across the organization rather than existing at the fringes—it’s vital to reframe their understanding of AI-driven value and set their sights not just on efficiency but effectiveness, as it offers more potential value. If efficiency is about doing things faster or cheaper—less input for the same or greater output—effectiveness is about the degree to which objectives are achieved. It shifts the focus towards quality and impact, and delivering on strategic goals.
Brynjolfsson advises doing this on two different axes. The first is to consider how people can be made more effective at their jobs. “Every job is a bundle of tasks—these are the atomic units of work,” he says. “A radiologist might do 27 different things, for instance, and only one of them is actually reading medical images. Some tasks are perfect for AI to help with, others less so. Looking at it this way helps you really get your hands around what AI can do for you.” Generative AI may make it possible to speed up tasks that AI couldn’t previously help with, freeing up workers to be more effective by devoting more thinking time and creativity to other areas of their jobs. And in those other areas, AI may be able to function as a co-pilot to drive that effectiveness. Consider a property lawyer: less time having to draft routine documents is more time finding solutions to clients’ unique problems.
The second way to think about effectiveness is at a systemic level: how can the business as a whole be more effective?
In the near term, there is scope to improve established aspects of the business that—before generative AI—were unquantifiable. As the adage has it: if you can measure it, you can manage it. There are a range of things that generative AI is able to analyze and make tangible for the first time. “A concrete example is new approaches to customer sentiment analysis by looking at transcripts from your customer call centers,” says Brynjolfsson. “These tools can analyze language, intonation, and tone in ways—and at a speed and scale—that we just couldn’t do previously.” A more advanced version of that use case could entail delivering conversational analysis in real-time to nudge customer support people towards creating better outcomes.
As generative AI matures, another category of systemic effectiveness will come into view: the ability to do new things entirely, which is yet another form of value. Take marketing—a number of technologists anticipate the emergence of what has been referred to as ‘liquid content’. This describes a single marketing asset that can shift in medium and scope depending on who is watching, thanks to generative AI. If one person likes short videos, they get it as a short video. If the next person tends to watch videos that are up to 30 seconds longer, they automatically get a cut that reflects that. If someone else would prefer to step into it in VR, then it snaps into a volumetric experience. The future could involve countless different experiences of the same piece of content—the dream of personalized marketing at vast scale becomes reality.
Ideas like liquid content could translate into hard sales figures, but often generative AI improves effectiveness simply by helping us think better. It’s about helping people be more creative, make better decisions, and be more innovative. So focusing on classical ROI metrics like cost and time may fail to capture its full impact. You need a greater range of KPIs that track those softer outcomes. And generative AI itself—with its ability to process unstructured, multimodal data—can, once again, help measure less concrete things like creativity. It can help, for instance, in assessing novelty or surprise. But be under no illusion that tracking outcomes is vital. “If you demand causal estimates of whether or not your solution is actually moving the dial on your key performance metrics, and adjust accordingly, this can be revolutionary,” says Brynjolfsson, “We could have a decade of explosive business value ahead.”
While creating a complex app would require the user to have the technical know-how to get under the hood and make tweaks, this kind of ‘citizen development’ can be adequate for creating simple tools or prototypes. “In general, it increases the speed of digital transformation quite broadly,” says Davenport. “If you only have an IT organization to build the helpful applications people might want to do their jobs better, it's going to take way too long. And this approach can be used for innovation. In pharmaceutical companies, for example, smart people can create new models, and statistical models are increasingly the key to identifying new drugs.”
Of course, anything involving sensitive data requires collaboration with other groups. Clearly, legal and compliance teams will need to be involved—as ever, innovation needs to be balanced with safety and responsibility. It will also likely involve IT, which will be keen to ensure security. But Davenport notes that IT professionals increasingly tend to welcome rather than resist citizen development. “People can actually show them their prototype and ask IT to make it more bulletproof and robust—that’s a far better dialogue than just throwing some written requirements over the fence to the IT organization and saying ‘build this for me’.”
Organizations that wish to foster citizen developers should aim to create a culture of experimentation in which people are open to failing, but which also has governance firmly in the process—not as an inhibitor, but to bring clarity and open up the lanes for innovation. It’s also vital to remember that citizen developers are volunteers. “Don't force anyone to do something that they don't care about or don't know enough about to want to learn,” says Davenport. “Instead, encourage the curiosity and initiative of those in your organization who are naturally enterprising.” Once you have identified and inspired that group, offer meaningful rewards to those who come up with tools that create real value. “Most organizations don't reward this kind of activity very well,” says Davenport. “But doing so is essential, not just to motivate them but to ensure they pass their discoveries back to the company.”
Davenport highlights a second new consideration for innovation cultures: looking beyond your own four walls. Davenport recently co-authored a paper revealing how, in the age of AI, companies are improving the speed and quality of corporate innovation by working with external partners. “We are seeing a degree of partnering among organizations, including competitors, as never before,” he wrote. Partly, this is because AI can enable secure sharing of mutually approved and beneficial data. It’s also because an organization may want to fine tune an AI model to perform an industry-specific task, but lacks sufficient data internally to do that. A new generation of purpose-built platforms for this very use case are enabling enterprises to collaborate on data ecosystems.
A note of caution: this willingness to support the sharing of approved data won’t come naturally to every organization. “Even groups within the same organization can encounter several roadblocks when it comes to data sharing,” says Kyndryl’s Patrick Gormley. “So, the first thing is to ensure your organization actually has an enterprise data strategy that removes data silos and supports the goal of securely sharing data.”
Yet it’s a new normal to which it may become important to adjust, not only because walled gardens can hinder innovation, but because the AI age may usher in new practices for the wider economy. One of the opportunities for AI, thanks to the ability of generative AI to integrate multiple platforms and protocols, is for agentic systems to handle tasks across multiple organizations for their mutual benefit. If this happens, seeing your business as part of an ecosystem—and being prepared to offer access to data accordingly—could become essential.
Fostering an ‘innovation culture’ has been a fixation of technology companies since the digital age took off. The tactics of the businesses that got it right have become canonical: design buildings so people with disparate specialisms mingle and cross-pollinate ideas; bring startups under the umbrella of your organization; give staff time to work on passion projects.
But what does innovation culture look like in the age of AI—are there new tactics that organizations need to embrace?
Tom Davenport, the President’s Distinguished Professor of IT and Management at Babson College in Massachusetts, and the co-author of All-In On AI: How Smart Companies Win Big With Artificial Intelligence, says that the pace of AI development—coupled with the fact that there is no manual for what to use it for—means that the time-tested tactic of crowdsourcing insights and ideas from your workers will be more important than ever. But he highlights two specific new considerations that generative AI introduces.
First, there is the ability for your team to make their own software. Generative AI allows non-developers to create programs, because they can effectively speak things into reality. This is new—and it’s a game changer. “If you can say what you want, you can basically generate an application or a website,” says Davenport. A term has recently emerged for this practice: vibe coding. You provide the natural language prompt for the software, see what the AI comes back with, tell it how to tweak it, and go through a process of iteration until you have something you’re happy with. If there’s a little friction in your job that a micro-app could help solve, or a tool that would benefit the company at large, but you can’t find something out there that offers the functionality, no problem: vibe code it.
It might do so in a way that solves a longstanding bottleneck when it comes to technology’s capacity to improve workplace performance. Previous waves of digital transformation, such as the introduction of desktop computing, did not lead to big, macro productivity gains. This may be down to ‘Polanyi’s Paradox’. Named for the information theorist Michael Polanyi, it posits that much knowledge can’t simply be written down and disseminated because it is tacit: it is about know-how and judgement born of experience. Therefore a lawyer could have the best textbooks and technology in the world, but still not perform as well as a lawyer with an extra decade under their belt.
“There are many things you might know, particularly if you've been in a profession for a long time, that you’re not actually able to articulate to someone,” says Jeremy Kahn, the award-winning author of Mastering AI: A Survival Guide To Our Superpowered Future. “A good example is great salespeople. They pick up on subtle signals and improvise on the fly, but it's very hard for them to pinpoint exactly how they’re doing that.”
Generative AI could enable us to thwart Polanyi’s Paradox. It is distinguished by its ability to learn from unstructured data in an unsupervised fashion. “That means it has the potential to capture lots of things that are correlated together, including natural language in documents, but also potentially video and audio too, and to correlate that with good outcomes.” To return to the sales example, it could listen to successful calls by top salespeople in order to understand the techniques that led to closing deals.
It may be possible to articulate some of those techniques as rules or principles, capturing and making visible strategies that were previously hidden. However, it’s possible that much tacit knowledge will elude codification. Here again, generative AI can help, coaching others on how to mimic what it finds is effective—whether that’s a way of approaching a certain problem or navigating a particular conversation—either by providing lessons or doing it on the fly.
So what does this mean for organizational design? “You could get much flatter organizational structures,” says Kahn. Hierarchies today are largely determined by experience and ability. If more people can perform at a high level, we may see a less stratified org chart. We may also see new business models where organizations are able to scale in their complexity while remaining lean—a trend intimated by the emergence of AI-driven ‘tiny team’ businesses, with one company achieving $100m in revenue with just 20 employees.
The ability to capture tacit knowledge could change the dynamics within the organization. The idea of employing graduates who spend years performing ‘grunt work’ for which they are over-qualified—but which is seen as a necessary part of training in sectors such as law and banking—could become outmoded, and those graduates may be happier as a result. It may also change the value proposition of labor. If it’s possible to ‘save’ the know-how of star performers for posterity, those people may have less leverage in salary negotiations. “On the other hand, though, the star performers know that the firm is training AI on their work, and that’s a reason to demand more compensation up front,” says Kahn. “If you're going to give some of your personal IP to the firm, you're going to want to be compensated.”
General-purpose technologies tend to usher in new ways of organizing businesses. The steam age gave us the org chart when the scale and complexity of 19th Century railways necessitated visualizing reporting lines and roles. The internet gave us methodologies such as agile, as companies realized that networking and data enabled more dynamic ways of working. What will the age of AI give us? Could this technology allow us to rethink the architecture of business again?
“Many companies are currently built around a hierarchy,” says Kyndryl’s Dr. Shaun Barney. “This can sometimes lead to unnecessarily lengthy, convoluted approval chains—where centralized decision-making can only operate at a certain pace, and this actually slows down innovation. For a company to be AI-ready, they must also be people-ready. This may require restructuring your organization to align all business and technology leaders to the goals you want generative AI to help you achieve, while continuing to prioritize its responsible application with an established governance model.”
Futurists’ presentations are already full of speculation for how this will be addressed in the years ahead. Will companies incorporate a pervasive AI layer that supervises all activities, distributing tasks, providing feedback, and shaping overall strategy? Might traditional department delineations vanish, replaced by flexible teams that can form and disband as needed, supported by a pool of highly capable AI assistants? Your view on such matters will typically be a function of how far you believe AI will advance: the extent to which it becomes general-purpose, autonomous, and capable of intuiting our intentions and producing outcomes that align with them. Opinions, therefore, differ.
There is, however, growing consensus over a nearer term shift that may happen as a result of one of AI’s key attributes: it not only has the capacity to amplify the skills of talented workers, but the capacity to bring junior employees up to a good standard of performance much faster. It can serve as a co-pilot in performing work tasks, and also act as a powerful personal tutor and coach.
For decades, the search engine has been the internet’s front page. That could be about to change—many believe that the web of the future will be mediated not by search engines but by chatbots.
The frontier labs have been adding web search functionality to their generative AI chatbots. At the same time, dedicated AI search startups have emerged. For all of them, the proposition is the same: if a traditional search engine merely gives you ‘10 blue links’, AI search gives you a direct, relevant answer. Straightforwardly, it’s a time saver.
There is increasingly a sense that this will go further, thanks to AI agents. There are countless different definitions of ‘agent’, but the prevailing vision is an automated assistant that can complete multi-stage tasks using digital tools to achieve a given goal. This could turbocharge search, researching queries and triangulating data points to a degree you would never do yourself.
There is growing interest in how these forms of AI search could change purchasing journeys. For purchases that normally begin with research—trying to find a new car, say, or a new washing machine—companies will need to ensure the chatbot is likely to find and recommend their products. “The near term risk of this for businesses is being accidentally invisible,” says Matt Webb, founder of product design studio Acts Not Facts, and an influential thinker on the future of technology, design, and society. So what does SEO look like in an age of AI?
In the near term, Webb believes that it will be about ensuring your website is easy for a large language model (LLM) to read. “To give you a small example, at the moment a lot of people spread their content over maybe five pages, because you want people to click through and engage. That makes it hard for LLMs to read the content. So you probably just want to have really long pages instead.” You probably also want to quantify any claims about the product. If you sell high-end bedding, for instance, it would be better not to simply say that your sheets are “luxurious” but to offer data, such as thread counts, which prove this to be the case. This will make it more likely the AI will include it in data-driven product assessments. “It may also be possible to lead the AI by the nose a little,” says Webb. “Perhaps explicitly write things on your site that tell the AI who your product is for—and reflect this in the images too—in a way that anticipates the kinds of questions your target audience may be asking their chatbots.”
In the longer term, however, if consumers come to use chatbots to handle entire purchasing journeys, a new paradigm could emerge. “I expect that to avoid hallucinations, chatbots will want to use external data sources to do the searching,” says Webb. He imagines this will involve the user connecting plug-ins to connect the chat app to the specific platforms that they want to act as the search tool for specific needs. So: if you want to order a takeaway, the chatbot will use the plug-in to the takeaway app that you like best. This will have the further benefit of producing results that you trust. “Anyone who you might have a shopping account with could offer such a plug-in. I expect the chatbot will be searching via these, rather than just searching the whole web.”
What’s the most significant upshot of that for businesses? “You're going to want people to discover and connect your plug-in, in the same way as you want to get them to download your app, or a particular platform’s app, today,” says Webb. “That means your customer relationships are going to be more important than ever.”
Are you ready for the new age of business?
For more original thought about the next frontier of technology, and how to harness the power of AI for advantage in your organization, go to:
KYNDRYL
Stop fixating on bottom-line efficiencies. Open the aperture of imagination
“Competing on price alone is rarely a sustainable competitive advantage, and it is short-sighted about the technology’s true potential.”
Erik Brynjolfsson, Jerry Yang and Akiko Professor, Senior Fellow at Stanford University, and Director of The Digital Economy Lab
What are the three ways to think about ROI for generative AI?
“Encourage the curiosity and initiative of those in your organization who are naturally enterprising”
Tom Davenport, the President’s Distinguished Professor of IT and Management at Babson College in Massachusetts
What is vibe coding, and why should you care about it?
What could cause online businesses to become invisible to customers?
What is Polanyi's Paradox, and how might generative AI solve it?
“For a company to be AI-ready, they must also be people-ready. This may require restructuring your organization to align all business and technology leaders to the goals you want generative AI to help you achieve”
Dr. Shaun Barney, Global Data and AI Consulting Partner at Kyndryl
1
Embrace the next generation of innovation culture
2
You will need to rethink organizational design—and possibly your business model
3
AI is going to influence what consumers buy. Get used to it
4
"The near term risk of this for businesses is being accidentally invisible”
Matt Webb, founder of product design studio Acts Not Facts
“That means your customer relationships are going to be more important than ever”
Matt Webb, founder of product design studio Acts Not Facts
Efficiency gains, improved effectiveness, and new capabilities
Natural language-based coding, allowing non-developers to create programs and accelerate innovation
The theory that much human knowledge is tacit and cannot be captured and disseminated - generative AI may solve this due to its ability to reason across disparate, unstructured data sets and identify correlations
Failure to optimize for AI-mediated purchasing
DOWNLOAD THE PDF