It has been anything but a quiet summer for AI. From OpenAI’s GPT-5 release to Gartner placing AI in its “trough of disillusionment,” the narrative has shifted again. Yet the headline that cut through the noise came from MIT: 95% of AI initiatives fail to deliver measurable returns. The immediate reaction was a sharp 10% drop in AI stock valuations.
Many rushed to declare the generative AI bubble burst, drawing comparisons with the .com bubble. But this is not the end of the story. Having navigated past technology cycles, a different pattern emerges: what looks like failure may be the necessary correction to reset expectations and force focus on value.
Focus On: From FOMO to Focus
The first wave of AI adoption was driven by fear of missing out. In 2023, the question every CEO faced was: “Where is our AI strategy?” The result was scattered pilots and unfocused spending. Harvard Business Review has since described this as the “experimentation trap.” Lots of activity, little scalability.
The real issue? Investment flowed disproportionately into front-office functions like sales and marketing, where creativity and nuance resist automation. Meanwhile, predictable, high-ROI back-office processes—finance, HR, compliance—saw far less attention.
Missing the Forest for the Trees
Businesses exist to solve problems for customers—not to prove they can build AI. History offers perspective:
Dot-com (1995–2001): hype ended in a crash, but the infrastructure built powered today’s digital economy.
Social media (2004–2012): hundreds of platforms, only a few endured.
SaaS (2008–2016): an explosion of CRM startups, but Salesforce and a handful of players defined the space.
The 95/5 split isn’t failure. It’s the signal-to-noise filter. Corrections separate sustainable value from temporary excitement.
Specialisation Over Superintelligence
Predictions of imminent AGI have given way to a more practical reality: model clustering and specialisation. Small language models (SLMs) now deliver 10x–100x efficiency gains in targeted use cases. A compliance checker doesn’t need storytelling skills. The future is orchestration of specialist agents—a coordinated ensemble, not a soloist.
The 5% of successful projects share a pattern: AI layered onto deterministic systems with human oversight. The human role is conductor, not competitor. AI thrives where:
Processes are codifiable
Feedback loops are tight
Human creativity adds little incremental value
Integration is seamless
This explains why back-office automation delivers impact, while front-office transformation lags (for now).
The IFD Framework: Choosing Where to Experiment
To escape the experimentation trap, apply three filters when selecting AI problems:
Intensity: How severe is the problem?
Frequency: How often does it occur?
Density: How many instances face this issue?
Start from customer value, not from technology and look at AI through a longer term lens:
AI Everywhere (2022–2023): visibility over value.
AI Where It Matters (2024–2026): focused deployment, ROI-driven.
AI as Infrastructure (2027+): embedded, invisible, indispensable.
Enterprises that adapt are shifting from internal builds (33% success rate) to partnerships (66% success rate). From AGI dreams to incremental improvements.
Ideas for Leaders
Back-office first: follow the money. Prioritise automation in functions with repeatable processes and measurable ROI, such as finance, HR, and compliance. These areas often deliver quicker payback and create credibility for further investment.
Connect experiments directly to strategy. Every AI pilot should map clearly to a business outcome. If you cannot show how an initiative supports customer value or operational efficiency, shut it down and redeploy resources.
Design teams for scale, not pilots. Successful AI adoption requires dedicated cross-functional teams with leadership sponsorship, not isolated “labs.” Invest in operating models that allow promising pilots to scale into enterprise platforms.
Orchestrate networks of agents, not monoliths. Specialist models integrated into a coordinated architecture will outperform large, general-purpose systems. Leaders should push for modularity to ensure flexibility, resilience, and cost efficiency.
Time entry: learn from early adopters, but act before lock-in. First movers have absorbed high costs and risks; fast followers can now adopt proven patterns. But wait too long, and standards, ecosystems, and vendor lock-ins will harden, reducing your strategic options.
The AI story isn’t ending—it is maturing. The correction is less a collapse than a clearing of noise. Leaders who frame AI as part of broader digital transformation—not as an end in itself—will move into the 5%.
How many of your AI initiatives tie directly to customer value?
What would change if you applied the IFD framework ruthlessly?
Follow me
That’s all for this week. To keep up with the latest in generative AI and its relevance to your digital transformation programs, follow me on LinkedIn or subscribe to this newsletter.
Disclaimer: The views and opinions expressed in Chronicles of Change and on my social media accounts are my own and do not necessarily reflect the official policy or position of S&P Global.