The GenAI Divide: Why We Should Separate Hype from Reality (for now)

Artificial intelligence dominates today’s business conversations. Many board and management meetings have AI on the agenda. Every CEO is asked, “What is our AI strategy?” Yet beneath the noise, a sharp divide is emerging between hype and actual transformation.

The State of AI in Business 2025 report from MIT’s Project NANDA shows that despite $30 to $40 billion in enterprise investments, 95% of organisations report no measurable return. Only 5% of pilots move into production and deliver value. This is what the researchers call the GenAI Divide.

As board members and executives, understanding this divide is essential. It is not about model quality, regulation, or even talent. It is about whether organisations approach AI as a tool to experiment with, or as a system that must learn, adapt, and embed deeply into workflows.

Adoption Is High, Transformation Is Low

The numbers are striking. Over 80% of organisations have tried tools like ChatGPT or Copilot. Almost 40% report some level of deployment. But these tools primarily improve individual productivity, not company performance. They help employees draft emails, summarise notes, and translate documents, but they rarely move the P&L.

When it comes to enterprise grade systems, those designed to automate workflows or drive operational efficiency, the picture changes dramatically. 60% of organisations evaluated such tools. Only 20% reached pilot stage. Just 5% reached production.

The result is clear: trying AI is not the same as transforming with AI.

Where Real Disruption Is, and Where It Is Not

Seven out of nine major sectors show little structural change from GenAI adoption. Despite the headlines, only technology and media display clear disruption.

In technology, AI-native start-ups are reshaping workflows and challenging incumbents. In media, AI-generated content is shifting advertising and engagement models.

By contrast, sectors like healthcare, energy, financial services and advanced industries show minimal disruption. Pilots are common, but business models remain intact.

This finding should give us pause. Despite heavy investment, most industries are still “business as usual”. The hype says transformation is everywhere. The data we have today says otherwise.

The Myths Boards and Executives Should Challenge

The report highlights five myths about enterprise AI. These are worth keeping in mind at the next strategy discussion:

1. “AI will replace most jobs soon.” Reality right now: job impact is limited and concentrated in outsourced functions such as customer support or document processing. (If you sit on the side of the outsourcer, the impact may already be visible.)
2. “AI is transforming business.” Reality: adoption is high, but structural transformation is rare.
3. “Enterprises are slow adopters.” Reality: enterprises lead in pilots but fail to scale.
4. “The main barriers are legal or model quality.” Reality: the true barrier is that most tools do not learn or adapt to workflows.
5. “The best enterprises build their own tools.” Reality: internal builds fail twice as often as external partnerships.

This means asking management to separate signal from noise. Investments should be scrutinised not for novelty, but for measurable outcomes.

The Shadow AI Economy

Perhaps the most surprising insight is the rise of shadow AI. Employees are already using consumer tools like ChatGPT or Claude on their own accounts, often without IT approval.

Only 40% of companies officially purchased AI subscriptions. Yet over 90% of employees in the study admitted to using AI tools personally for work. In some organisations, this shadow usage delivers more ROI than official initiatives. However, the security dimension of this AI shadow IT must not be underestimated.

Boards and executives should take note. Innovation is already happening inside organisations, but it may not be captured by formal strategy. Rather than resist it, leading firms are studying shadow usage to identify where value emerges, then investing in enterprise grade alternatives.

Why Pilots Stall

The central barrier is the learning gap. Today’s AI tools are good at generating output but poor at learning from feedback, retaining memory, and adapting to specific workflows.

Executives repeatedly emphasised that ChatGPT is better for personal use than most enterprise AI tools. Why? Because it feels flexible, fast and familiar. Yet when applied to mission-critical tasks, its lack of memory and customisation makes it unsuitable.

This gap explains why 95% of enterprise AI pilots fail to scale. For boards and executives, the key question is not “Do we have AI pilots?” but “Do these systems adapt and improve over time?”

How the Best Cross the Divide

Despite the grim averages, some organisations are succeeding. Their approach offers lessons for every board:

1. Buy, don’t build. Internal builds fail twice as often. External partnerships achieve 66% success rates versus 33% for in-house projects.
2. Start small, then scale. Winning vendors begin with narrow use cases like call summarisation or document tagging, then expand into core processes.
3. Demand learning systems. Executives want AI that adapts, remembers and evolves with use.
4. Benchmark outcomes, not models. ROI is measured in cost savings, customer retention, or BPO elimination, not in technical accuracy scores.

Where the Real ROI Lives

While most AI budgets go to sales and marketing, the highest returns are in the back office. Organisations that cross the divide reduce outsourcing, eliminate business process outsourcing (BPO) contracts, and cut agency costs.

Examples include:

• Eliminating $2 to $10 million annually in BPO contracts.
• Cutting external creative agency spend by 30%.
• Saving $1 million annually in outsourced risk management.

Importantly, these savings come without mass layoffs. The shift is from external costs to internal capabilities.

The Next Horizon: Agentic AI

Beyond today’s AI tools lies agentic AI. Instead of isolated systems, autonomous agents will coordinate across networks, negotiate contracts, and manage workflows without human intervention.

Protocols like the Model Context Protocol (MCP) and Agent to Agent (A2A) are already laying this foundation. In the coming years, enterprises will move from experimenting with pilots to operating in a connected ecosystem of learning, adaptive agents.

For organisations, this is not science fiction. It is the next wave of digital transformation.

What Boards and Executives Should Do Now

1. Ask the right questions. How many pilots have moved to production? What business outcomes are measured? Do tools learn and adapt over time?
2. Redirect investment. Challenge management to look beyond sales and marketing and explore high-ROI back-office opportunities.
3. Embrace partnerships. Favour vendors who integrate deeply and evolve with your processes over internal builds that stall.
4. Leverage shadow usage. Study where employees already use AI informally and build enterprise grade solutions around those workflows.

Summary

The GenAI Divide is not about technology. It is about approach. Organisations that treat AI as just another software licence will stay stuck in pilots. Those that treat it as a learning system, embedded in workflows, will see measurable impact.

For boards and executives, the task is clear: move beyond the hype, cross the divide, and ensure AI investments translate into real business transformation.

Source: The GenAI Divide: State of AI in Business 2025 (MIT NANDA)