At the 2025 World Economic Forum in Davos, Microsoft CEO Satya Nadella issued a stark warning: if AI benefits only accrue to tech firms, “it’s a bubble.” His concern reflects a growing paradox in AI adoption across industries. Despite massive investments in generative AI and widespread experimentation, genuine transformation remains conspicuously absent outside the technology sector.
Bridging the Discovery and Integration Gap in AI Adoption
The numbers tell an interesting story. According to MIT Technology Review’s survey of enterprise AI adoption, 95% of companies now use generative AI in some capacity. Yet most organizations have deployed AI in only very limited use cases. The gap between experimentation and scaled deployment is vast, and it raises an uncomfortable question: if the “AI” hammer is so powerful, why can’t most industries find the nails that can benefit from this hammer?
The answer lies in two interconnected challenges that organizations outside the tech world struggle to overcome: a discovery problem that prevents companies from identifying where AI can create genuine value, and an integration problem that keeps isolated AI experiments from scaling into organizational transformation.
Companies often experiment with AI in obvious, low-stakes areas like email drafting or chatbot customer service, while potentially transformative applications in procurement, quality control, maintenance optimization, and demand planning go relatively unexplored. The hammer exists and is powerful, but without the ability to identify where it should be applied, much of its potential remains largely theoretical.
Part 1: The Discovery Problem - When Capabilities Don't Translate to Applications
The first barrier is deceptively simple: most organizations cannot identify which of their problems are actually solvable with AI. This is not a laziness problem or a reluctance to innovate. It is fundamentally a discovery problem rooted in an awareness gap.
Consider a SME producing automotive components. The operations manager knows that quality inspection is labor-intensive, that supply chain coordination creates constant friction, and that maintenance schedules are inefficient. These are tangible, costly problems. What the manager may not know is that large language models can analyze unstructured maintenance logs to predict equipment failures, that computer vision can automate defect detection with greater consistency than human inspectors, or that AI can optimize production schedules by processing real-time data from suppliers, machines, and logistics partners.
AI capabilities and business problems: two languages, no translation
The problem is not that AI cannot address these challenges. The problem is that AI capabilities are communicated in technical language—natural language processing, computer vision, reinforcement learning—while business problems are expressed in operational terms like “reducing scrap rates,” “minimizing downtime,” or “improving on-time delivery.” The translation gap between these two vocabularies might leave countless high-value applications invisible to decision-makers.
Identifying AI use cases: blind spots from office to shop floor
This awareness gap manifests differently across the white-collar to blue-collar continuum. In professional services firms, executives understand that AI can draft emails and summarize documents, but they miss less obvious applications: analyzing contract portfolios to identify risk patterns, extracting key obligations from thousands of vendor agreements, or generating customized client reports by synthesizing data from multiple systems. In logistics and warehousing operations, managers see AI as futuristic automation but overlook immediate wins like route optimization based on real-time traffic and weather data, or demand forecasting that accounts for seasonal patterns, regional events, and competitor behavior.
Data quality as AI barrier – symptom, not cause
The MIT survey identifies data quality as the primary barrier to AI deployment, and executives are right to cite it. But the data quality problem is itself a symptom of the discovery gap. Organizations don’t know what data they need because they haven’t identified the right problems to solve. A chemical manufacturing plant sitting on years of production data might not realize that this information could train models to optimize reactor conditions, reduce energy consumption, or predict batch quality issues before they occur.
The discovery problem is compounded by how AI is sold and demonstrated. Vendor presentations typically showcase impressive generic capabilities—chatbots, image generation, sentiment analysis—but these demos don’t always bridge effectively to specific industry workflows. A hotel chain executive might see an impressive natural language interface but may not immediately connect it to automating guest service requests, analyzing review data to identify operational issues, or personalizing marketing based on booking patterns and guest preferences.
AI literacy meets domain expertise: the translators are missing
What makes discovery particularly challenging is that it often requires a rare combination of deep domain expertise and genuine AI literacy. The operations manager typically understands their processes intimately but may lack the technical knowledge to recognize which steps are amenable to AI augmentation. The data scientist understands what models can do but may lack the contextual knowledge to identify which business problems are most critical or feasible to address. Organizations need people who can operate fluently in both languages, and such individuals tend to be scarce outside technology companies.
The result is that many high-value “nails”—opportunities where AI could create substantial impact—may remain invisible. Companies often experiment with AI in obvious, low-stakes areas like email drafting or chatbot customer service, while potentially transformative applications in procurement, quality control, maintenance optimization, and demand planning go relatively unexplored. The hammer exists and is powerful, but without the ability to identify where it should be applied, much of its potential remains largely theoretical.
From Hammer to the Right Nails
Why 95% of companies use AI but very few achieve genuine transformation – and how to bridge the gap
The Hammer Without Nails
Isolated experiments, individual productivity gains, ChatGPT for emails and summaries. AI usage is high, but transformation remains elusive.
Finding the Invisible Nails
Translating between AI capabilities and business problems. Identifying high-value applications that remain invisible because technical and operational languages don't connect.
Workflow Transformation
End-to-end process redesign, not point solutions. Strategic top-down direction combined with bottom-up innovation. AI embedded systematically across the organisation.
Part 2: The Integration Problem – From Individual Experiments to Organizational Transformation
Even when organizations successfully identify valuable AI applications, a second and equally formidable barrier emerges: translating individual experiments into scaled, organizational transformation. This is where the gap between 95% usage and very limited deployed use cases becomes most telling.
Andrew Ng, founder of DeepLearning.AI and co-leader of AI advisory firm AI Aspire, captured this challenge precisely in his reflections from Davos 2025. Speaking with CEOs about AI adoption, Ng observed that “running many experimental, bottom-up AI projects—letting a thousand flowers bloom—has failed to lead to significant payoffs.”
The problem, Ng argues, is that real transformation requires workflow redesign: “taking a broader, perhaps top-down view of the multiple steps in a process and changing how they work together from end to end.”
AI implementation requires workflow redesign, not point solutions
Consider his example of a bank issuing loans. The workflow consists of discrete stages: Marketing → Application → Preliminary Approval → Final Review → Execution. Suppose preliminary approval previously required an hour-long human review, but a new AI system can complete this automatically in 10 minutes. Simply swapping AI review for human review—while keeping everything else unchanged—provides a minor efficiency gain but isn’t transformative.
What would be transformative, Ng explains, is recognizing that “instead of applicants waiting a week for a human to review their application, they can get a decision in 10 minutes. When that happens, the loan becomes a more compelling product, and that better customer experience allows lenders to attract more applications and ultimately issue more loans.”
But achieving this requires more than just implementing AI in one step. It demands redesigning the entire workflow. Marketing must change to promote a “10-minute loan” product. Applications need to be fully digitized and routed more efficiently. Final review and execution must be redesigned to handle dramatically higher volume. Even though AI is applied to only one step, the organization must implement not just a point solution but a comprehensive workflow transformation.
Individual AI adoption is not enterprise transformation
This insight illuminates why most organizations are likely to remain stuck at limited AI use cases. Individual employees discover that ChatGPT helps them draft reports faster or that AI tools can summarize meeting transcripts. These personal productivity gains are real but isolated. They don’t require coordination, don’t challenge existing processes, and don’t demand organizational change management.
Scaling these individual experiments to create enterprise-wide impact is an entirely different challenge. It requires answering difficult questions: How do teams collaborate when some members use AI and others don’t? How do we standardize outputs when everyone prompts AI differently? Who owns the responsibility for redesigning workflows—IT, operations, or department heads? How do we overcome institutional inertia when existing processes have momentum and established stakeholders?
Change management for AI: who owns the transformation?
The coordination barriers are substantial. When AI usage remains ad hoc and individual, outputs are inconsistent, quality control becomes difficult, and the cumulative impact is limited to the sum of individual productivity gains. Moving from “AI-literate individuals” to “AI-enabled organizations” requires systematic process redesign, shared standards, clear ownership, and change management capabilities that most non-tech companies lack.
AI strategy needs leadership from the top
As Ng notes, “Bottom-up innovation matters because the people closest to problems often see solutions first. But scaling such ideas to create transformative impact often requires seeing how AI can transform entire workflows end to end, not just individual steps, and this is where top-down strategic direction and innovation can help.”
The integration problem explains to a large extent why the MIT survey shows high AI usage but minimal deployment at scale. Organizations have successfully completed the first step—individuals are experimenting with AI tools. But they have not completed the second step—redesigning business processes and workflows to embed AI systematically. The gap between these two stages is where most AI initiatives stall, and where the difference between incremental efficiency and genuine transformation lies.
Bridging Both Gaps
The challenge facing non-tech industries is not a lack of powerful AI capabilities or a shortage of problems worth solving. The challenge is twofold: discovering which problems are genuinely amenable to AI solutions, and then integrating those solutions into redesigned workflows that unlock transformative rather than incremental value.
Addressing the discovery problem requires bridging the vocabulary gap between technical AI capabilities and operational business challenges. It demands professionals who can translate between these domains, identifying “invisible nails” that represent high-value opportunities currently missed by organizations.
Addressing the integration problem requires moving beyond bottom-up experimentation to strategic, top-down workflow redesign. It demands organizational capabilities in change management, process engineering, and systems thinking—skills that enable companies to scale isolated AI experiments into enterprise-wide transformation.
Both gaps represent not just technical challenges but coordination and capability challenges. Organizations that can bridge these gaps—through internal development or specialized external expertise—will be positioned to move AI from experimental novelty to genuine competitive advantage. Those that cannot will likely remain in the 95% that use AI but the vast majority that fail to transform with it.
If AI’s benefits are to extend beyond technology firms, the focus must shift from acquiring more hammers to becoming far better at finding—and properly driving—the nails.
What’s next?
Want to find out where the “invisible nails” are in your organisation – and how to move from isolated AI experiments to genuine workflow transformation?
AI Decisions in Business: Why Humans Must Remain Final Authority
[…] if the only tool you have is a hammer, every problem looks like a nail. Or, applied to this topic: the best hammer is useless when you actually need a screwdriver. And not every nail can be driven in with the same […]
AI Tools for Business: Which Tool for Which Task? | NordAGI
[…] if the only tool you have is a hammer, every problem looks like a nail. Or, applied to this topic: the best hammer is useless when you actually need a screwdriver. And not every nail can be driven in with the same […]