Many companies adopt AI chaotically - no strategy, no defined processes, everyone doing their own thing. It creates costs that could be avoided. We show what the research says about failed AI projects and how clear processes make the difference.

Everyone is adopting AI - but hardly anyone knows exactly what they are doing

AI runs in almost every company today. Rarely where someone made a deliberate decision about it. One colleague has ChatGPT write her quotes, the developer next door uses a different tool for coding, the marketing team has bought yet a third subscription. Nobody coordinates, nobody documents, nobody asks whether any of it is even secure.

This is not an exception, it is the rule. A study by Karlsruhe University of Applied Sciences of 517 mid-sized companies (20-500 employees) shows that only 21% have an AI strategy at all - even though 40% are already using AI. In nearly half of all companies (48%), the use of language models is not regulated at all. And according to a YouGov survey, 77% of employees in STEM professions use AI tools like ChatGPT, Gemini or Claude without IT approval - almost a quarter of them daily.

Microsoft coined a term for it: "Bring Your Own AI". 78% of AI users bring their own tools to work, and at smaller companies it is as high as 80%. At the same time, 60% of leaders say their own leadership lacks a plan and a vision for adopting AI.

The pattern is the same everywhere: lots of activity, no process. AI is not being adopted - it is seeping in.

What this chaos really costs

At first glance it looks harmless, even productive. After all, employees are getting on with it. The problem: activity is not the same as value creation. And the gap between the two is expensive.

The most important number comes from McKinsey: more than 80% of companies see no measurable effect from generative AI on their operating result (EBIT). Only 17% can demonstrate any meaningful contribution to the bottom line at all. The Boston Consulting Group reaches the same conclusion in its study of more than 1,250 companies: only 5% generate value from AI at scale, while 60% are laggards with no notable benefit.

This is not a technology problem. It is a follow-on-cost problem caused by missing structure:

Abandoned projects: According to S&P Global Market Intelligence, 42% of companies scrapped the majority of their AI initiatives in 2025 - up from 17% a year earlier. On average, 46% of all proofs of concept are discarded before they ever reach production. Gartner predicts that at least 30% of GenAI projects will be abandoned after the proof of concept - due to poor data quality, missing controls, escalating costs and unclear business value.
Duplicate and unused licenses: When every department buys its own tool, the same thing gets paid for several times over - and nobody has the overview.
Security and data protection risks: The IBM "Cost of a Data Breach" report 2025 puts a concrete figure on it: data breaches involving shadow AI cost $670,000 more on average. One in five companies has already had an incident caused by ungoverned AI - and only 37% have any policies at all to detect or manage shadow AI. German data protection authorities warn in no uncertain terms: without clear rules, employees use AI "on their own initiative and without any oversight".

So the costs of chaotic AI adoption never show up as a single line item on an invoice. They hide in discarded projects, in duplicate subscriptions, in data leaks - and above all in the time that flows into things that never deliver measurable value.

The real mistake: treating AI as a tool problem

This is the heart of the misunderstanding. Chaotic AI adoption treats AI as a question of tooling - as if it were enough to hand employees a few tools and hope for magic.

The numbers say the opposite. According to BCG, only around 10% of the value of AI comes from the algorithms themselves and about 20% from the technology. The remaining 70% come from people and processes - from the way work is organized. McKinsey examined 25 factors to find out what best explains the financial success of AI. The result: it is not the model, not the budget - it is the fundamental rethinking of the affected workflows. Yet so far only 21% of companies do exactly that.

And focus beats activism: successful companies concentrate on 3.5 use cases on average, according to BCG, while laggards spread themselves across 6.1 - and earn 2.1x less return in the process. More tools, more experiments, more parallel construction sites do not mean more value. Quite the opposite.

What actually works: hold on to what works, define processes

The good news: the companies where AI really makes money are not doing secret magic. They hold on to the things that work anyway - and they define clear processes. Five principles can be drawn from the research:

1. Start with the problem, not the tool

At the beginning stands a concrete, expensive, recurring business problem - not the question of what a particular tool can do. Which process costs unnecessary time and money every month? That is where AI belongs in a trial, not in a hype-driven side project.

2. Focus instead of spreading thin

Better to bring two or three use cases cleanly into production than to run a dozen experiments in parallel that all fizzle out at the pilot stage. Around 70% of AI's potential lies in the core functions anyway, such as sales, service, manufacturing and procurement.

3. Rethink processes, do not bolt AI on

The biggest lever is not the model, but the workflow around it. Putting AI on top of a broken process only breaks it faster. The process first, then the automation.

4. Steer by enabling - not by banning

Bans only create more shadow AI. A German specialist study (Wirtschaftsinformatik & Management, 2026) explicitly recommends alignment between IT and the business over a culture of prohibition: jointly supported pilot projects, role-based access, clear data rules and licensed, enterprise-grade tools. Where companies provide vetted tools, uncontrolled use drops by up to 89%.

5. Make it measurable

According to McKinsey, the single most effective measure is to track clearly defined KPIs for AI solutions - combined with an unambiguous roadmap. What is not measured cannot be steered. And what cannot be steered becomes expensive.

None of these principles is new or spectacular. They are the same fundamentals that have always defined solid software and process work: understand the problem, focus, clarify responsibilities, measure results. AI changes none of that - it just makes it more urgent.

How vensas supports companies with AI adoption

At vensas we experience the difference between AI activism and AI with a plan in practice. We help companies to

identify the use cases that are genuinely worthwhile - where an expensive process meets real savings potential,
turn scattered experiments into a clear, repeatable process,
build governance that enables people instead of slowing them down - with clear data rules and vetted tools,
and make success measurable instead of relying on gut feeling.

Our focus is not on the latest tool, but on business impact: less manual effort, shorter cycle times, lower risk.

Conclusion: AI is not the problem - the missing process is

In most companies, AI has long been adopting itself. The question is no longer whether your own employees use AI, but how - in an orderly, measurable way, or chaotically and expensively.

The research is clear: what makes AI fail is rarely the technology. It is the absence of exactly the things that work anyway - a clearly understood problem, focus, defined processes and measurable goals. Those who hold on to them turn AI from an invisible cost driver into a real competitive advantage.

Sources & further reading

All studies and figures as of 2024-2026. Findings from surveys and industry reports shift continually - verify against the original source before making strategic decisions.

AI Without a Plan: Why Companies Burn Money - and How Clear Processes Win It Back