Editorial
The Gist
- AI bubble correction was inevitable. Hype around rapid breakthroughs created unrealistic expectations that needed a reset.
- Enterprises still chase real ROI. Most organizations struggle to scale AI beyond pilots and must focus on practical, high-value use cases.
- Model progress is slowing. Advances are shifting from giant frontier models to smaller, specialized ones tailored to industries.
- Humans plus AI drive the near term. The strongest results come from pairing AI tools with human judgment, not replacing it.
Earlier in July, Apollo Academy published a report that the AI “bubble” is bigger than the dotcom bubble of the late 1990’s. They observed that the difference between the dotcom bubble in the 1990s and the AI bubble today is that the top 10 companies in the S&P 500 today are more overvalued than they were in the 1990s.
As if that wasn’t enough, then “The Report” landed.
That report, published by the MIT NANDA initiative, sent shockwaves through the AI market and had exuberant investors scrambling to refute what many have observed on the ground.
The Productivity Gap in Enterprise AI
The report concluded that 95% of AI projects are not delivering a return on investment. It goes on to highlight a “Pilot to Production Chasm” that reveals a steep drop-off between investigations of AI, to the pilots and ultimately product adoption of AI solutions. The report defines successfully implemented AI solutions as ones where users or executives have experienced a marked and sustained productivity and/or P&L impact.
The key highlight in the report was that the core barrier to scaling AI solutions is not infrastructure, regulation, or talent. It is learning. Most of the AI systems do not retain feedback, adapt to context, or improve over time. The primary factor keeping organizations on the wrong side of the GenAI Divide is the learning gap, tools that don't learn, integrate poorly or match workflows.
Users prefer AI tools for simple tasks but abandon it for mission-critical work due to its lack of memory. What's missing is systems that adapt, remember and evolve, capabilities that define the difference between the two sides of the divide.
Pilot-to-Production Barriers
Key reasons why most enterprise AI projects fail to scale successfully.
| Barrier | Description |
|---|---|
| Lack of learning | AI systems fail to retain feedback or adapt over time. |
| Poor workflow integration | Tools do not align with existing enterprise processes. |
| Hype-driven initiatives | Projects prioritize flashy narratives over practical ROI. |
| Resource bottlenecks | Scaling AI requires computing, budget and organizational buy-in. |
Ask anyone in the trenches of a large enterprise, and you will hear a lot of excitement and usage of AI tools in their daily workflow or in discrete tasks. But ask them about organizational benefit and ROI and the discussion falls flat.
Related Article: Two Years of Generative AI: How Has Customer Experience Delivery Changed?
Grassroots Adoption vs. Corporate Stalemate
Ultimately, the MIT NANDA report surprised people because it exposed a hidden truth. It’s not that AI is a failing technology. It’s that the AI revolution is succeeding quietly at the grassroots level, driven by individual employee adoption. It exposed that formal, top-down corporate AI initiatives often stalled due to misalignment, poor integration and a focus on hype over practical results.
It's clear there is a fantastic opportunity to activate AI at enterprise level across organizations.
Table of Contents
- Why Enterprises Needed a Reset on AI Expectations
- Why GPT-5 Fell Short of CX and Marketing Hype
- AI Specialization: The Future of Industry-Specific CX Tools
- Probabilistic AI Limits: What Marketers and CX Teams Must Know
- Human-AI Collaboration: The Key to Customer Experience Success
- The Path Forward: Practical AI for Real ROI in CX
- Core Questions About the AI Bubble Correction
Why Enterprises Needed a Reset on AI Expectations
The AI narrative has been so explosive over the past several years it’s not unrealistic to expect some correction in AI sentiment was needed. It’s not dissimilar to financial market correction, which correlates to the ~10% market correction we just experienced in public AI stocks after the MIT report was released.
It’s not hard to see why this happened.
After the release of ChatGPT in November 2022, several AI “super-narratives” emerged.
The dominant AI narrative became that AGI (artificial general intelligence) was only two to three years away. That’s a very bold proclamation for a technology that has been searching for this for over eight decades. AGI then started to become bifurcated where some began defining it as smarter than human super-intelligence or artificial superintelligence (ASI). ASI is AI that has gone beyond human intellect and would be superior in many, if not all, areas.
From AGI Dreams to Operational Reality
We have not yet attained the AGI and yet the next category was already being envisioned. It is unknown whether we will reach AGI soon or if AGI will take decades to fully achieve.
Due to the rapid progression of models, the belief was that leading models were on a path to self-learn. The idea was that models would develop a recursive self-improvement (RSI) loop making models rapidly improve autonomously getting closer to human super intelligence or AGI.
A second major “super narrative” was the expected loss of jobs due to AI.
In May 2025, Anthropic CEO Dario Amodei issued a blunt warning that AI could wipe out half of entry-level knowledge workers jobs over the next three to five years. This prediction has echoed among other prominent technologists and executives, including Ford CEO Jim Farley, who similarly forecasted that AI could reduce the number of white-collar jobs in the United States by half.
And you've got Sam Altman this week warning AI will replace all customer service jobs.
Jobs Lost or Jobs Rewired?
However, research has found limited support for large-scale layoffs from AI. Wired, Forbes and the New York Times have featured analysis noting that although AI is changing the workforce and the way work gets done in enterprises, it is not causing irreversible job loss. In fact, many new careers and creative opportunities are arising due to AI adoption.
This correction in sentiment was clearly needed to offset.
Why GPT-5 Fell Short of CX and Marketing Hype
It didn’t help that ChatGPT 5 (GPT-5) was overhyped and underwhelming.
This really crystalized around the launch of GPT-5 where many thoughts this would be a breakthrough model. Sam Altman did his usual pre-release hype by posting a picture of the Death Star (from the Star Wars franchise) with no accompanying text, just a day before the GPT-5 release. It was an odd reference likely meant to tease the massive capabilities of GPT-5 but the connotations around “evil” was not Altman’s best look.
Because expectations were through the roof, a huge number of people viewed GPT-5 as a major letdown. It’s not that the model didn’t reflect progress, it just fell short of these expectations. Many expected GPT-5 to be a next generation, “frontier model” that dramatically transformed AI capabilities.
Webinar
Oct
16
Agentic AI Playbook: Real-World Customer Service Use Cases You Can Deploy Now
Boost self-service by 30% and slash call volume by 63% with agentic AI.
Register
Webinar
Oct
22
Beyond Storage: Smarter Content, Bigger Impact with DAM + AI
Discover how the DAM + AI duo makes content smarter, stronger and more accessible.
Register
Webinar
On demand
Ready or Not: How Data-First Organizations Are Unlocking Agentforce Potential
Learn how to cut through the noise, activate Agentforce and build a Salesforce AI strategy that actually delivers.
Watch Now
Webinar
On demand
AI in Customer Service: Faster Resolutions, Happier Customers
Don’t let rising demand burn out your team. See how to build a smarter, more resilient support org.
Watch Now
Webinar
On demand
From Hype to High-Impact CX Strategies That Actually Scale
Turn buzzworthy AI and outsourcing trends into measurable CX wins with fresh 2025 data.
Watch Now
Webinar
On demand
Insights to Action Rethinking the Contact Center for Real Business Impact
Join our exclusive webinar to hear CX executives share their innovative strategies for transforming service delivery.
Watch Now
Expectations Versus Benchmarks
Historically we saw 20-30% increase in performance from GPT-3 to GPT-4 with major leaps in reasoning and coding. GPT-5 performance increases were closer to ~10% (although it does vary by benchmark) with focused gains in multimodal reasoning, coding and tool use.
There were also expectations around capabilities to handle a million-token context window and deep multimodal reasoning, which GPT-5 moves toward but does not yet fully deliver.
The development of models is on a more traditional technology evolution curve. There is unlikely to be a standout model that can achieve AGI quickly. Model performance is now likely to be much more incremental. We are now seeing a clustering of model performance around the same benchmarks.
As a result of this we can apply more normal technology curve and re-align expectations around future model releases.
Related Article: OpenAI Forgot the Golden Rule of CX: Don't Yank Away What Customers Love
AI Specialization: The Future of Industry-Specific CX Tools
Models are beginning to become increasingly specialized. They are developing distinct personalities. Google excels at video, Anthropic is really good at coding. Claude excels at chat. Llama has strong domain adaption.
One model is unlikely to become all knowing and powerful. Model performance will continue to be leapfrogged. But the real evolution will now come from specialization and smaller models.
Large, general models (like LLMs) are powerful but can be costly, computationally intensive and sometimes generate irrelevant outputs for industry-specific tasks. Businesses require AI that is tailored, flexible and efficient, rather than one-size-fits-all solutions. The success of specialized models relies on high-quality, relevant data curated for specific domains, leading to more accurate and actionable results.
Smaller Models, Bigger Impact
Small language models (SLM’s) offer compact, efficient performance making them ideal for use in resource-constrained environments like mobile devices, edge hardware, or applications requiring low latency and privacy.
For example, a law firm could use a specialized AI for analyzing critical legal documents, a healthcare provider for assessing medical scans or a retailer for AI-powered customer service.
Probabilistic AI Limits: What Marketers and CX Teams Must Know
It is possible that the challenges resulting from probabilistic systems are beginning to become obstacles for AI applications to get into that last mile of production.
While the world is not inherently deterministic, the current AI systems leverage probabilistic AI, grounded in Bayesian modeling. These models are fundamentally about reasoning under uncertainty providing a range of possible outcomes along with a measure of uncertainty. Which means they can hallucinate and are not entirely reliable for all classes of applications.
When Uncertainty Becomes a Risk
Probabilistic AI is not a fit for all applications because its core nature relies on uncertainty and statistical likelihoods, which can be unacceptable or even dangerous in high-stakes, safety-critical or rules-based tasks. Unlike a deterministic system, which produces the same exact output for a given input, a probabilistic system's output vary.
AI is increasingly embedded in everyday life and critical business processes we will need AI models that provide deterministic outcomes.
Transitioning to a purely deterministic approach would require significant advancements in AI research and development. Therefore, today, the reality is that most AI applications will need to incorporate human oversight and intervention where necessary, especially for critical decisions or sensitive tasks.
Some applications are perfect for probabilistic AI. Take marketing mix modeling (MMM) where companies look to understand the effectiveness of their spend across different channels — Google Ads, TV, social media, etc. MMM models built with probabilistic methods don’t just output ROI estimates, they also quantify the uncertainty around those estimates.
Probabilistic AI mimics human reasoning processes, which often rely on assessing probabilities rather than certainties, thus providing responses that feel more natural and intuitive.
Designing for Transparent Uncertainty
Future AI systems will integrate generative models with structured probabilistic reasoning to create more transparency. The goal is to create systems that can communicate their level of uncertainty, allowing for smarter and more trustworthy human-AI collaboration.
Human-AI Collaboration: The Key to Customer Experience Success
The other hidden truth that surfaced is the reinforcement that AI is best when paired with humans. The reality is you can’t just spin up AI and expect it will run organizational functions like HR, Finance, IT, marketing, etc. on its own. The practicality is that systems and functions that embed AI in processes and augment human workers are most likely to deliver real ROI.
The partnership between human and artificial intelligence represents one of the most promising developments in our technological evolution. Rather than replacing human workers, AI enhances human capabilities, creating a synergy that drives innovation and productivity to new heights.
Human-AI Synergy in Practice
Examples of how AI augments, rather than replaces, human expertise.
| Industry | AI Role | Human Role |
|---|---|---|
| Healthcare | AI analyzes scans for subtle patterns | Doctors interpret results and make final decisions |
| Financial services | AI runs scenario modeling and risk analysis | Advisors apply judgment and client context |
| Marketing | AI predicts consumer behavior and optimizes ads | Marketers craft narratives and emotional appeal |
| Customer service | AI automates simple inquiries | Agents resolve complex or high-stakes issues |
Best-of-Both-Worlds Use Cases
There are many examples of this partnership between human workers and AI in practice today.
AI in healthcare has improved the accuracy of medical diagnoses. A Stanford study reported that AI powered diagnostic systems results in 92% accuracy, assisting doctors to find subtle patterns that might go unnoticed. AI acts like a built-in, highly trained second “pair of eyes” that can rapidly analyze complex medical images like X-rays, MRIs and CT scans to help detect early signs of conditions like cancer or heart disease.
Financial services professionals now have tremendous analytical power through AI. AI is able to provide deeper risk assessment and quickly evaluate multiple scenarios to help protect investments. AI reduces emotional bias in investment decisions. While human investors might panic during market downturns or get caught up in investment hype, AI systems maintain objectivity, adhering strictly to data-driven strategies.
Marketing’s Edge: Data Plus Story
Marketing was one of the early applications of AI. AI has become an indispensable tool for marketers offering unprecedented efficiency and data analysis capabilities. Marketers have used AI to predict consumer behavior, segmenting audiences, and optimizing Ad performance and bidding strategies. Yet, despite AI’s growing capabilities in marketing, AI can’t craft stories that truly move people to action. That emotional connection is something that AI still struggles to replicate. The most powerful marketing strategies combine AI’s computational power with human creativity.
The Path Forward: Practical AI for Real ROI in CX
AI is a power tool and technology and will absolutely unlock tremendous value in the economy. But it will take us a while to get there. In the interim, it’s critical that enterprises create practical AI solutions that have demonstratable AI, starting with small use cases and quick wins.
The question or call to action for all enterprises is how you become part of the 5%!
Core Questions About the AI Bubble Correction
Editor's note: Key questions surrounding the AI “bubble,” enterprise ROI and the evolving balance between hype and practical adoption.
Growth will come from smaller, specialized models paired with human oversight. Practical use cases that augment employees will deliver the most immediate impact.
Hype around rapid breakthroughs inflated unrealistic expectations. The MIT NANDA report and market performance highlighted the gap between investor optimism and enterprise reality.
The majority of projects stall in the “Pilot to Production Chasm.” Tools often fail to learn from feedback, integrate poorly into workflows and lack long-term adaptability.
Learn how you can join our contributor community.
