Why Organizations Struggle to Scale AI Beyond Testing

Testing AI vs. Embedding It in Daily Work

Trying AI once and putting it in charge of your daily work are two different decisions. When you’re testing it, you can just judge how plausible the answers sound. But in actual operations, you have to figure out who’s checking for errors and what it’s costing you.

There’s a body of research worth bringing into this comparison. One is a study from the 1940s that tracked how farmers in Iowa adopted new hybrid corn seed. The other is Chasm Theory, which describes how tech products spread beyond their earliest customers.

Let’s start with what the AI surveys actually asked. In McKinsey’s 2025 survey, 88% of respondents said their organization uses AI regularly for at least one business function. About a third said their organization had begun scaling AI use more broadly. McKinsey survey

In Deloitte’s 2026 survey, 25% of respondents said their organization had moved more than 40% of its AI experiments into production. That 40% isn’t the share of companies succeeding—it’s the share of projects within each organization that made it to production. In PwC’s 2026 CEO survey, 56% said they hadn’t seen both revenue growth and cost savings from AI in the past 12 months. That doesn’t mean they saw no effect at all—things like faster workflows or higher employee satisfaction aren’t captured in that number. Deloitte survey, PwC survey

These surveys asked different questions of different groups, so you can’t just subtract one number from another to arrive at an “AI failure rate.” What they do show is that adoption, breadth of use, and financial results need to be examined separately. I want to take the second of those—the challenge of scaling usage—and connect it to Geoffrey Moore’s Crossing the Chasm.

What Convinced Farmers to Switch Seeds

A chasm refers to the gap in which a technology product that has won over early customers struggles to spread to a wider customer base. Behind this concept lies diffusion of innovations theory, which studies how new technologies spread among people.

In 1941, Iowa State University sociologist Bryce Ryan and graduate student Neal C. Gross investigated how hybrid corn seed spread through two rural communities. At the time, farmers were being introduced to new seed varieties with improved yield and growing performance, but they adopted them at very different times.

According to the research process as summarized by Rogers, Gross interviewed 345 farmers and analyzed responses from 259 who met the study’s criteria. Of these, 257 had adopted the new seed between 1928 and 1941. Salesmen were most often cited as the channel through which farmers first learned about the seed, while neighboring farmers were most often cited as influencing the decision to actually adopt it. In other words, between hearing about the new seed’s advantages and deciding to plant it in one’s own field, there was the experience of someone else.

This is a conceptual diagram of chasm theory, not a graph of the actual survey figures from the Iowa farmers.

Published in 1943¹, this paper became an important starting point for later research on the diffusion of innovations. Everett Rogers synthesized related studies and published Diffusion of Innovations in 1962. Rogers divided adopters into five groups based on the relative timing of when they accepted something new.

• Innovators, 2.5%: The group that tries new technology first.
• Early Adopters, 13.5%: Those who adopt relatively early and also influence the judgment of those around them.
• Early Majority, 34%: Those who adopt earlier than average, but only after carefully examining other users’ experiences.
• Late Majority, 34%: Those who adopt only after the technology has become widely established and the uncertainty of adoption has diminished.
• Laggards, 16%: The group that adopts the technology last.

These percentages are a theoretical breakdown of adoption timing according to a normal distribution. Not every market produces the same ratios, and a given person doesn’t necessarily belong to the same group across every technology. When you plot the cumulative number of adopters over time, an S-curve typically emerges. Note that this differs on the vertical axis from the bell-shaped curve above, which divides adopters by period.

Moore, who applied this model to technology product marketing, paid particular attention to the difference between early adopters and the early majority. His point was that the reasons the first customers loved a product are not, by themselves, enough to convince the next wave of customers.

The Chasm Geoffrey Moore Discovered

Geoffrey A. Moore studied American literature at Stanford and earned his PhD in English literature at the University of Washington. He moved from teaching to sales and marketing at technology companies, and later did marketing consulting at Regis McKenna. His work involved connecting a product’s technical features with the reasons customers actually buy.

Product teams tend to emphasize new features. But buyers first check whether those features are actually useful for their own work, and whether they’ll get support after adopting them. Moore’s argument is that you have to understand this gap in purchasing criteria if you want to sell to the next wave of customers.

Moore published Crossing the Chasm in 1991. The book addresses the problems that arise when marketing that worked for early customers gets repeated unchanged for mainstream customers.

What he called the chasm is the gap that’s hard to cross between the early-adopter market and the pragmatic mainstream market. It’s less a law that every product hits at the exact same point, and more a framework tech companies can use to check their footing as they try to expand their market.

In Moore’s account, visionaries adopt technology even when it’s unfinished, as long as they see large future upside. They’re willing to solve problems themselves and adapt the product to fit their needs.

Mainstream customers, by contrast, put more weight on proven results in similar industries and similar workflows. A single case where a tech-savvy team succeeded through extraordinary effort isn’t enough to convince them it’ll work in their own organization. The installation, training, and support all need to be in place before they’ll commit to a purchase. Introduction to Moore’s book

Building a Product the Next Customer Can Actually Use

Building on Moore’s ideas of choosing a narrow customer segment and delivering a whole product, I’ve boiled down what to look at when expanding a market into four points. Tesla’s case is added here simply to illustrate the thinking.

First, you have to narrow down the customer problem you’re solving first. Decide which task, in which industry, you’re going to solve, and build a product that customer can actually use. From there, you can expand to customers with similar needs. Tesla didn’t launch every vehicle class at once either. Its 2006 master plan started with the Roadster, a high-priced electric sports car, and laid out a plan to expand toward cheaper models from there.

Second, you need to provide all the support required to actually accomplish the purchase. Moore calls this the whole product². Someone buying an electric car needs not just the car, but a place to charge it and a place to service it. Tesla’s decision to build out the Supercharger network alongside selling vehicles can be understood from this angle.

Third, you have to give customers confidence that they can keep using the product. Whether it’s compatible with other products, whether the supplier will keep supporting it, and whether other companies in the ecosystem use the same technology all affect the purchase decision. Tesla’s 2014 patent announcement was also an attempt to broaden the use of electric vehicle-related technology. It was a pledge not to bring patent lawsuits against anyone who met certain good-faith conditions of use — not a wholesale abandonment of its patent rights.

Fourth, you have to lower the burden of initial adoption. Beyond price, you need to account for the time it takes to learn the product and the cost of changing existing habits. Tesla expanding its lineup to target lower price points is one example of easing that price burden. For business software, this might mean letting it connect to existing systems, or letting teams pilot it on just a portion of their work first.

Oswarld’s Lens

The question I find most important in chasm theory is whether the first successful customer and the next customer need the same conditions. It’s worth checking whether you’re demanding the same self-sufficiency from your next user that your early, skilled users provided for themselves.

I think this question is useful for AI as well. Menlo Ventures estimated enterprise generative AI spending at $37 billion for 2025. That’s roughly 3.2 times larger than the $11.5 billion figure for 2024, recalculated using the same scope for comparison. The estimate combines a survey of 495 AI leads at U.S. companies with market analysis, and excludes spending on semiconductors or inference services. It’s evidence that investment is growing, but as with the PwC survey mentioned earlier, whether individual companies are actually seeing financial returns needs to be verified separately.

Diffusion across customer segments in a market and expanding AI use within a single company aren’t the same thing. Still, both cases share something in common: you need to find out what your next user is worried about. A task that’s easy for the team that built the AI can be hard for a frontline employee, and whoever bears responsibility when something goes wrong may demand a stricter standard of proof.

If I were expanding a company’s AI use, here are the four things I’d check first.

• Scope of work: Pick a concrete task—like classifying customer inquiries or reviewing code—and confirm that processing time and error rates actually improve.
• Operational readiness: Connect the necessary data, and prepare access permissions, employee training, and a process for handling problems when they arise, all together.
• Evaluation criteria: Decide in advance what counts as a passing result, and compare before-and-after outcomes against the same standard.
• Adoption burden: Consider whether you can start with human-reviewed assistive tasks. As you scale up automated execution, you need to build in commensurate controls.

Just as neighbors’ experience mattered in the study of Iowa farmers, results from similar tasks elsewhere can help inform AI adoption decisions. But rather than importing another organization’s success wholesale, you should check how similar the data, personnel, and operating conditions really are.

If model performance falls short, you need to improve or replace the model; if the data connections or review procedures are lacking, you need to fill those gaps. I think persuading the skeptics and actually putting the conditions for adoption in place have to happen together.

Closing

To extend a new technology’s early success, you need to understand the next user’s job—specifically, you need to ask what they can solve on their own, where they’ll need support, and what results would make them keep using it.

Rather than rolling AI out across every department at once, I’d rather confirm both the performance gains and the operational burden in a single function first. Basing the decision of how far to expand on that outcome gives me solid grounds for recommending it to other departments.

References & Further Reading

• Geoffrey A. Moore, Crossing the Chasm. Covers how to expand from early adopters to the mainstream market.
• Bryce Ryan & Neal C. Gross, The Diffusion of Hybrid Seed Corn in Two Iowa Communities, Rural Sociology 8, 15–24, 1943. The bibliographic source for the Iowa seed-diffusion study.
• Everett M. Rogers, Diffusion of Innovations, 5th Edition, 2003. The publisher’s public excerpt confirms the corn study’s survey and analysis subjects and results.
• McKinsey, The State of AI in 2025. Surveys the scope and scaling stages of AI adoption.
• Deloitte, State of AI Report 2026. Covers the share of AI experiments that have moved into real operations.
• PwC, 29th Global CEO Survey, 2026. Asked about revenue and cost changes driven by AI.
• Menlo Ventures, 2025: The State of Generative AI in the Enterprise. Explains the estimates and coverage scope of the enterprise generative-AI market.
• Tesla, 2006 Master Plan, Patent Pledge. Confirms the terms of the product-expansion plan and the patent policy.

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

Footnotes

1. Ryan & Gross (1943) studied the diffusion of hybrid corn seed in two rural communities. Per Rogers’s account, 345 people were interviewed, 259 were included in the analysis, and 257 had adopted the seed by 1941. ↩

2. The “Whole Product” refers to a complete bundle of product, service, and support needed for a customer to achieve their purchase goal. For software, this can include installation, integration with existing systems, training, and maintenance. ↩