
As AI adoption matures, some companies are reducing or rethinking projects because of costs and weak returns. The lesson is selective, measurable adoption—not abandoning AI.
From “AI everywhere” to “AI where it works”
The first wave of generative AI adoption encouraged companies to experiment broadly. By 2026, many teams were asking tougher questions about operating costs, reliability, data protection, and measurable return on investment. Some projects have been paused or scaled back. That is not proof AI has failed; it is a sign that businesses are beginning to judge it like any other investment.
The hidden cost of an impressive demo
A proof of concept may look inexpensive until it needs real customer data, integrations, security controls, monitoring, human review, and ongoing model usage. A workflow that seems fast in a demonstration can become costly when run thousands of times or when errors require extensive correction.
What the research shows
A 2026 McKinsey survey reported that about one in five respondents said AI operating costs had constrained use, even as a majority planned to increase AI investment. Gartner reported in October 2026 that only 13% of surveyed organizations achieved significant or transformational value from AI. These findings point to a gap between enthusiasm and dependable business results—not a universal retreat.
Why early projects run into trouble
Common problems include unclear goals, poor data quality, weak oversight, inadequate testing, unexpected usage bills, and applying generative AI to tasks that would be simpler with ordinary rules or software. A tool that works most of the time may still be unsuitable for a process that must work correctly every time.
AI is not the best tool for every job
A predictable form submission, a fixed pricing calculation, a basic image adjustment, or a small website correction often does not need a language model. Conventional software may be cheaper, easier to maintain, and more reliable. Conversely, AI may add real value when a task involves ambiguous language, large amounts of unstructured information, or complex exploratory work.
The smart response is not to abandon AI
Businesses should define the problem first, run a limited pilot, measure total costs and error rates, establish human oversight, and expand only when the results justify it. Sometimes the best outcome of a pilot is learning not to deploy the tool.
How LMM approaches the next phase
At Levy Media Marketing, we see AI as one option in a broader toolkit. Our job is to recommend the right solution—not the trendiest one. Whether that means custom software, a streamlined website, traditional automation, AI-assisted development, or a carefully designed AI feature, the standard remains the same: does it make the project better for the client?
Sources and further reading
McKinsey, “The State of AI in 2026: On the Road to ROI” (August 2026): Read source
Gartner, “More AI Investment Is Not Creating Enough AI ROI” (October 2026): Read source
S&P Global, “Generative AI Experiences Rapid Adoption, but with Mixed Outcomes” (May 2025): Read source