95% of organizations are getting zero return on AI, while shadow AI thrives. But at what knowledge risk?

This article is part of an ongoing series looking at AI in KM, and KM in AI.

New research¹ by MIT NANDA reveals that despite billions of dollars in enterprise investment into generative AI (GenAI), 95% of organizations are getting zero return. The research used a multi-method design, including a systematic review of over 300 publicly disclosed AI initiatives, structured interviews with representatives from 52 organizations, and survey responses from 153 senior leaders collected across four major industry conferences. Systematic reviews² produce a more reliable knowledge base through accumulating findings from a range of studies.

In what the research report authors call the ‘GenAI Divide’, just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable impact on the bottom line. Organizations stuck on the wrong side continue investing in static tools that can’t adapt to their workflows, while those crossing the divide focus on learning-capable systems.

The GenAI Divide is starkest in deployment rates, with only 5% of custom enterprise AI tools reaching production. Chatbots succeed because they’re easy to try and flexible, but fail in critical workflows due to lack of memory and customization. This fundamental gap explains why most organizations remain on the wrong side of the divide.

Generic LLM chatbots appear to show high pilot-to-implementation rates (~83%). However, this masks a deeper split in perceived value and reveals why most organizations remain trapped on the wrong side of the divide.

In interviews, enterprise users reported consistently positive experiences with consumer-grade tools like ChatGPT and Copilot. These systems were praised for flexibility, familiarity, and immediate utility. Yet the same users were overwhelmingly skeptical of custom or vendor-pitched AI tools, describing them as brittle, overengineered, or misaligned with actual workflows.

The shadow AI economy

Behind the disappointing enterprise deployment numbers lies a surprising reality: AI is already transforming work, just not through official channels. The MIT NANDA research uncovered a thriving ‘shadow AI economy’ where employees use personal ChatGPT accounts, Claude subscriptions, and other consumer tools to automate significant portions of their jobs, often without IT knowledge or approval.

The scale is remarkable. While only 40% of companies say they purchased an official LLM subscription, workers from over 90% of the companies surveyed reported regular use of personal AI tools for work tasks. In fact, almost every single person used an LLM in some form for their work.

This shadow AI economy carries a high level of risk for enterprises. One risk is the widespread problem³ of workers entering confidential or private information into public AI platforms. Another is revealed in recent research⁴ by Semrush. As shown in the following graph, the most frequently cited domains across four LLMs – AI Mode, AI Overviews, ChatGPT, and Perplexity – are user-generated content sites rather than high-quality factual sources.

This list includes sites that are notorious for their disinformation, such as Facebook⁵, and others well-known for their significant biases such as YouTube⁶ and Google⁷.

Such sites are not going to be the sorts of information sources that a quality enterprise would use directly. But the ‘GIGO’ formula of ‘garbage in = garbage out’ applies to the citing of such sites by AI. So, with almost all of the workforce engaged in the shadow AI economy, many enterprises are now indirectly using knowledge garbage in their decision-making. This can be expected to result in seriously detrimental outcomes for enterprises and their customers and communities of interest.

Crossing the GenAI Divide

The The MIT NANDA research report authors advise that organizations on the right side of the GenAI Divide share a common approach: they build adaptive, embedded systems that learn from feedback. The best startups crossing the divide focus on narrow but high-value use cases, integrate deeply into workflows, and scale through continuous learning rather than broad feature sets. Domain fluency and workflow integration matter more than flashy user experience.

The organizations and vendors succeeding are those aggressively solving for learning, memory, and workflow adaptation, while those failing are either building generic tools or trying to develop capabilities internally. Winning startups build systems that learn from feedback (66% of executives want this), retain context (63% demand this), and customize deeply to specific workflows. They start at workflow edges with significant customization, then scale into core processes.

Organizations also need to establish clear policies and guidance in regard to confidentiality and privacy in the use of AI, and provide associated training to workers.

Header image source: Kathryn Conrad & Digit / Better Images of AI / CC BY 4.0.

References: