Part of the challenge is that applied AI is not one uniform wave. According to an upcoming flagship report from ADL’s Innovation practice, adoption is clustering in specific industry and functional hotspots, while the supplier landscape remains highly fragmented. Generative AI (GenAI) is an important and highly visible part of that ecosystem, particularly where work is language-heavy, but it is not the full architecture of applied AI. Classical machine learning (ML) remains a major part of the market, and many emerging solutions are in fact hybrid solution stacks that combine GenAI with classical ML and advanced knowledge and data integration.

For executives, the implication is not simply to adopt more AI, but to navigate the ecosystem deliberately: scaling where the market is ready, redesigning work around the technology, building hybrid architectures, governing for scale, and retaining control where dependence would matter most.

Preliminary findings from the report suggest those hotspots are beginning to come into focus. Financial services, media and entertainment, consumer services, and transportation appear to be among the sectors moving fastest, while functions like operations, engineering, compliance, and marketing are emerging as early leaders in adoption.

But moving fast is not the same as getting value. The companies seeing measurable results aren’t necessarily the ones chasing the flashiest use cases. Instead, they’re getting a handful of fundamentals right: clean and accessible data, narrowly defined problems, redesigned workflows, and a relentless focus on business impact over experimentation for experimentation’s sake. Most important, the leaders who see ROI understand that AI transformation is not primarily a technology challenge — it’s an organizational one.

The move from experimentation to impact is happening fastest where companies apply AI to use cases with strong data foundations and clearly defined processes. Kolk points to one chemical company that used comprehensive data to rethink one of its most expensive and time-consuming activities: experimentation. Previously, determining how to create the right chemical for a scenario required repeatedly conducting real-world experiments to narrow down the possibilities.  Today, the company uses AI to analyze all relevant data (literature, patents, and experiments run internally and externally) to quickly identify the most promising routes to success. AI is then used to combine such predictive (synthetic) data with those coming from much more targeted physical experiments, which is far more efficient and effective than traditional approaches.

“Not only is physical experimentation costly, it is also time-consuming and often not effective. Having a superior new product in the market faster can easily generate tens or millions of dollars in value,” says Kolk.

Another early breakout area is software development, where AI is helping teams move faster on structured, well-scoped technical work, says Michael Papadopoulos, Partner in ADL’s Innovation practice and Chief Architect of Digital Problem Solving. His own work building software prototypes for the firm’s clients offers one example. Over the last 16 months, many of the prototypes Papadopoulos has developed — including applications for regulated industries — have been largely coded using AI, allowing teams to move from concept to working software much faster.

AI currently works best in workflows that are structured, narrow, and repeatable. This can include multistep workflows, as long as there are clear boundaries between steps. For example, Papadopoulos points to a code-review pipeline for resolving customer tickets, in which the workflow follows a consistent pattern and the steps are clearly defined. Organizations often find themselves disappointed when they push too far and expect AI to deliver value in workflows that require creative strategy, highly specific domain judgment, or complex reasoning chains.

“Let agents handle the well-defined, narrow tasks, but with strong human checkpoints,” says Papadopoulos. “You don’t want open-ended decisions being made by your AI agents yet.”

The same need for clear inputs and boundaries applies beyond software. René Bohnsack, Professor of Strategy and Innovation at Católica Lisbon School of Business & Economics, points to a credit-scoring bureau that reworked its processes using AI and can now generate credit scores faster and more accurately by integrating data from across the organization.

The limits become clearer when companies move from focused, well-bounded applications to broad access to generalized AI copilot apps. Despite major investment, these tools often have not delivered the productivity gains leaders expected, according to Bohnsack. Misuse can lead to documents plagued by hallucinations, misaligned solutions that are technically correct but don’t solve the problem, and results that require more time to fix than if a human had done it in the first place.

Employees naturally ask these tools to complete tasks they may not be suited for or lack the necessary context to deliver on, and that’s where they start failing spectacularly, says Papadopoulos. Context and data are king when it comes to AI, and for copilots and beyond, data remains one of the biggest challenges standing between organizations and successful AI transformation.

Sometimes, the data exists but isn’t maintained, clean, or accessible enough for AI models to use effectively. According to Kolk, many organizations mistakenly think they can jump into AI without taking care of the data side of the equation, a misstep that hinders their ability to drive successful AI transformation at scale.

“Many companies have not paid enough attention to data — the way they manage it, data hygiene, governance, integration, and so on. Now with AI, they suddenly expect to be able to run, but they cannot because that underinvestment is coming back to bite them,” he says.

Workflows are another significant challenge. To achieve real transformation, they must be fully redesigned around AI, not augmented with it. According to Papadopoulos, that means rethinking roles, decision rights, and core processes. Leaders need to drill into how AI impacts their business at the most fundamental levels, not bolt-on AI and hope for the best, he says.

Companies are also struggling with how to effectively bring the workforce along. AI transformation requires people to completely change how they work, and most people are not too eager to do that, says Kolk. For many workers, performing certain tasks — or performing them in a particular way — has long been part of their professional identity. This is a key adoption challenge, says Kolk, yet he’s seen many leaders ignore it. Instead, he suggests they address these resistances head-on with clear communication and sincerity.

There are steps leaders can take to overcome these challenges, starting with ensuring all the essential ingredients for AI transformation are in place.

The first step is getting your data in order and recognizing that this isn’t just a technical problem. In fact, to a large extent, it’s about instilling the right governance, behavior, and accountability, says Kolk.

Next, leaders need to connect AI execution to a clear business strategy, with the governance, skills, and organizational support required to move promising use cases into day-to-day operations. Without that foundation, even technically successful pilots can struggle to scale or deliver measurable impact.

To effectively bring together skills and strategy, leaders must strike the right balance between quick wins they can execute on now and longer-term AI initiatives — and be clear about which use case falls into which category. Recognizing that you may not yet have the data for a specific use case, or that it will require significant workflow changes employees aren’t ready for, and either holding off or taking time to prepare, is vital to avoiding pilot purgatory. In the meantime, companies should focus on what they can realistically achieve with AI today, helping them build momentum along the way.

“Think about time to value. How quickly can you move an AI initiative from pilot to deployment, then into something that will have measurable business impact?” says Papadopoulos.

This includes considering AI uses that may be technically narrow but have broad impact. For example, Kolk worked with a company to map the needs of its various business units. Company leaders discovered that despite differences across those units, many needs could be addressed by the same core AI solution with a few tweaks to the front-end user experience. It became an opportunity to make a significant impact while limiting complexity, making it an achievable early win.

“You need to move in steps that are individually valuable, but collectively become transformational,” says Kolk.

Leaders are responsible for understanding where true value lies and focusing on substantial business impact, including resisting the temptation to chase shiny use cases that look impressive on paper but fail to move the needle in practice. That means getting smarter about how you track and measure AI impact. According to Papadopoulos, metrics like tokens used or prompts submitted are pointless. He suggests measuring impact (not activity) through metrics tied to efficiency, quality improvements, and economic value.

AI may represent a technical leap unlike anything we’ve ever seen, but the fundamentals of business still hold. Organizations that tie their AI projects to a clear strategy and recognize that this is a business transformation rather than a technical upgrade will be best positioned to reap the benefits.

“The best thing leaders can do right now is focus on strong fundamentals, and those haven’t changed in decades,” Papadopoulos says. “Clean data, clear use cases, adaptable teams, and strong change management are still essential.”

• Data fundamentals come first. Companies can’t succeed with AI if their data isn’t in order.
• Balance quick wins with long-term AI initiatives. This helps leaders avoid getting stuck in pilot purgatory.
• Focus on solving urgent business problems, not shiny use cases.
• Measure AI impact rather than activity.
• Remember that AI transformation isn’t a technology problem. It’s a people, process, and business challenge.