Why AI Adoption Produces ROI for Some and a Headache for Others

McKinsey reported in 2024 that AI adoption had nearly doubled, but something stuck with me more than the headline number. One buried finding kept me thinking about it for days: only a small fraction of organizations, around fifteen percent, report significant revenue impact from AI. The rest are spending, integrating, rolling out tools, and still not getting the returns they expected. I kept asking myself why the distribution was so skewed, and I kept coming back to a framework I had been studying for my comps.

Tornatzky and Fleischer (1990) built the TOE framework around a simple observation. Technology adoption depends on three contexts working together. The technological context covers what AI tools are available and what infrastructure the organization already has in place. The organizational context covers firm size, management support, human resources, and slack resources available for innovation. The environmental context covers industry competition, regulatory conditions, and market uncertainty. Each context is necessary. None is sufficient alone. And when I looked at the McKinsey data through this lens, the pattern snapped into focus.

The fifteen percent that see significant ROI are organizations where all three contexts are aligned. They already had AI-relevant infrastructure before the latest wave hit, data pipelines, cloud architecture, integration-ready legacy systems. Their leadership actively pushed AI integration rather than waiting for bottom-up experimentation to prove something. And their competitive or regulatory environment either incentivized adoption or at least did not block it. The eighty-five percent that see no significant impact are missing at least one of these three. Maybe the technology works but the organization has no management champion and no slack to experiment. Maybe the organization is ready but the regulatory environment punishes the kind of data processing AI requires. Maybe the technology and environment are fine but the firm has no prior AI capability to build on.

This is not a failure of the tools. It is a failure of context alignment, and the TOE framework predicts it.

I thought about the BCG experiment that Harvard researchers ran with 758 consultants, the one that showed AI users completed 12.2 percent more tasks, 25 percent faster, and with higher quality within the AI capability range. Why did those consultants see benefits that a small business or a government agency would not? The BCG consultants sat inside an organizational context with strong management support, clear incentives to adopt, and a culture that rewards tool experimentation. They had the technological context of enterprise-grade AI tools with IT support, training, and integration into existing workflows. And they operated in an environmental context where consulting competitors were also adopting AI, so the pressure to use it was both internal and competitive. The small business trying to adopt a chatbot or a content-generation tool faces none of these conditions. The owner is the manager, the IT infrastructure is a few SaaS subscriptions, and the environmental pressure is survival, not competitive AI deployment. TOE says the different outcomes are not surprising. They are expected.

Regulated industries tell the other side of the story. Healthcare organizations and financial services firms have the organizational resources, large IT departments, compliance teams, and management structures that can absorb new technology. They often have strong technological contexts too, enterprise data systems that have been in place for years. But the environmental context in a regulated industry constrains what AI can touch. A hospital cannot deploy a diagnostic AI without FDA clearance. A bank cannot route loan decisions through a model that regulators have not validated. The environmental context in TOE terms can be a blocker regardless of how good the other two contexts are. This is why AI adoption in healthcare progresses differently from AI in startups. The startup has environmental freedom and no organizational resources. The hospital has organizational resources and no environmental freedom. TOE says neither has all three contexts, so neither gets transformative results, though for opposite reasons.

The result is the bimodal distribution I keep seeing in the data. Organizations with all three contexts aligned get transformative results. Organizations missing any context get marginal or no returns. And I think this gap will widen before it narrows. The organizations that already have all three contexts will invest further, build more AI capability, improve their infrastructure, deepen management support, and shape their environment through lobbying and standard-setting. The organizations missing one or two contexts will fall further behind because AI capability development is path-dependent and cumulative, exactly as Cohen and Levinthal (1990) described absorptive capacity. Prior AI knowledge enables faster AI learning, and those who have it will pull away. I wrote about this dynamic before in my post on absorptive capacity and why tools alone fail.

The organizational context is the hardest to fix because it is the least visible. A management team that does not push AI integration cannot be fixed by buying better tools. Slack resources for experimentation cannot be created by a memo. And the environmental context is the hardest to predict because regulation moves independently of organizational readiness. An industry-wide data privacy rule can shut down an AI initiative that took two years to build. TOE gives no mechanism for predicting which context will break, and my comps professor was explicit that TOE is a bucket, not a theory, but that is exactly the point. It tells you where to look. The mechanism comes from whatever theory you pair it with, institutional theory for environmental pressure, absorptive capacity for organizational readiness, diffusion of innovation for technology attributes. I linked all of this in my earlier post on the TOE framework more generally, and the pattern holds for AI specifically.

The practical takeaway for anyone watching an AI investment underperform is not to blame the tool. It is to ask which of the three contexts is missing. If the technology works but nobody in the organization knows how to integrate it, the gap is organizational. If the organization is capable but the market or regulation blocks deployment, the gap is environmental. If the environment is open and the organization is ready but the AI tool does not fit existing infrastructure, the gap is technological. The answer is never to try harder. The answer is to figure out which bucket is empty.

That is the real value of TOE for AI adoption in 2026. Not that it predicts anything. It does not. But it stops you from blaming the wrong thing.

```json
{
"@context": "https://schema.org",
"@type": "Article",
"headline": "Why AI Adoption Produces ROI for Some and a Headache for Others",
"author": {
"@type": "Person",
"name": "Ali Safari",
"url": "https://alisafari.space"
},
"datePublished": "2026-05-14",
"description": "TOE from Tornatzky and Fleischer explains why most AI investments fail while a few transform organizations: the three contexts must align.",
"keywords": ["AI adoption", "TOE framework", "AI ROI", "organizational context", "digital transformation", "information systems"]
}
```

```json
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "Why do most organizations fail to get ROI from AI adoption?",
"acceptedAnswer": {
"@type": "Answer",
"text": "The TOE framework attributes AI adoption failure to misaligned contexts. Most organizations lack one of three necessary conditions: technological infrastructure, organizational readiness including management support and slack resources, or an environmental context that permits AI deployment. Missing any one context blocks transformative ROI."
}
},
{
"@type": "Question",
"name": "What is the TOE framework in information systems?",
"acceptedAnswer": {
"@type": "Answer",
"text": "The Technology-Organization-Environment framework, developed by Tornatzky and Fleischer in 1990, identifies three contextual dimensions that influence technology adoption. It is a classification framework rather than a causal theory. It organizes adoption factors into technological, organizational, and environmental buckets without specifying mechanisms."
}
},
{
"@type": "Question",
"name": "Why is AI adoption in regulated industries slower than in startups?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Regulated industries have strong organizational resources and technological infrastructure but face environmental constraints from compliance requirements. Startups have environmental freedom but lack organizational slack and infrastructure. The TOE framework predicts different adoption bottlenecks depending on which context is weakest."
}
}
]
}
```