Many transformations begin with a structural asymmetry. A small group defines the problem, target state, technology and timeline; a much larger group is then expected to “come on the journey”. When adoption disappoints, a design problem becomes an attitude problem. The organization cites mindset, fear of the new or change fatigue.

This label is attractive because it locates the cause where the programme has to change least: in the people affected. The harder question is: **What if employees’ response is an accurate evaluation of the design?

That question matters whenever technology becomes part of the organization’s decision and leadership architecture. An AI system that recommends actions, allocates work, evaluates performance or mediates access to knowledge does more than change a process. It changes who judges, who explains and who carries the risk of error.

1. People respond not only to technology, but to unclear accountability

A 2025 OECD study examined algorithmic management using a survey covering 6,047 firms with at least 20 employees in six countries. Nearly two-thirds of surveyed managers reported at least one concern about the trustworthiness of the systems they used. The most common were unclear accountability for wrong decisions (28%), an inability to follow the logic of algorithmic decisions or recommendations (27%), and inadequate protection of workers’ physical or mental health (27%).[1]

These are not peripheral adoption concerns. If a system influences behaviour while accountability remains diffuse, hesitation is professional. It signals that the organization has made a technical decision without completing the associated governance.

At the same time, 63% of managers reported worker consultation as a governance measure. The OECD therefore treats participation not as a courtesy around implementation, but as part of trustworthy algorithmic management.

2. Beteiligung reduziert Angst, weil sie Einfluss und Information verändert

The OECD’s 2023 research, opens an external URL in a new window drew on 96 workplace AI case studies in finance and manufacturing across eight countries. It found job reorganization more often than displacement. Direct worker involvement in AI development and implementation could reduce job-loss anxiety and increase workers’ willingness to engage with the technology.

Participation does not work merely because people prefer being asked. It works when operational knowledge changes tasks, interfaces, safeguards and implementation. A feedback session with no decision space is communication, not co-design.

The study is qualitative and should not be read as a population estimate for all firms. Its practical value lies elsewhere: it reveals mechanisms that aggregate adoption figures can hide.

3. “Why do we have to change again?” is not the same as rejecting AI

A 2026 case study, opens an external URL in a new window examined the introduction of GenAI-supported HR search in a multinational technology organization with more than 300,000 employees. The authors combined system logs, an exploratory survey of 25 employees and ten interviews. The survey is not representative; the value comes from triangulating evidence during a live transition.

One employee’s response amounted to: why use yet another HR system when the old one is familiar? The researchers explicitly argued that this should not be treated as general resistance to AI. The new route had to offer enough value to justify additional navigation and learning effort.

Adoption depended on situational fit: role, device access, work setting, language, tenure and information needs affected practical usefulness. Search literacy, the quality and localization of the knowledge base, training, source checking and accessible human escalation routes also shaped adoption.

The leadership lesson is straightforward: formal access is not practical usability. High login or query volumes may still conceal the groups who cannot participate meaningfully.

It is: What rational information is contained in non-use, workarounds or criticism?

That reframes the leadership task. Before trying to communicate a predetermined solution into adoption, examine six potential design failures:

1. Problem fit: Does the change solve a relevant problem in daily work—or mainly a problem perceived by the project team?
2. Situational fit: Does it work across roles, languages, devices, shifts and local processes?
3. Accountability design: Is it clear who decides, verifies, corrects and answers for harm?
4. Knowledge infrastructure: Are sources, content, terminology, metadata and ownership current and findable?
5. Participation design: Can affected employees alter the solution—or merely react to it?
6. Learning architecture: Are there bounded experiments, feedback loops, training at the moment of use and functioning escalation routes?

This matrix is not an engagement survey. It forces transformation teams to separate observed behaviour from premature interpretation.

A 45-minute test before the next change intervention

• Step 1—Describe behaviour (10 minutes). 
  Do not infer motives. Replace “employees are anxious” with “in 42% of cases, employees contact HR after receiving the AI answer.
• Step 2—Develop competing hypotheses (10 minutes). 
  Include at least one hypothesis about value, work context, accountability and knowledge.
• Step 3—Choose the cheapest evidence (10 minutes). 
  Five task observations, failed queries, process times or ten short interviews will often reveal more than another awareness campaign.
• Step 4—Define a reversible experiment (10 minutes). 
  One role, one location, one task, one measurable change. Do not test an enterprise-wide solution when its critical assumption can be examined locally.
• Step 5—Set a stopping rule (5 minutes). 
  Decide in advance which result triggers adaptation, pause or termination. A pilot without a stopping rule is often just a rollout with a smaller audience.

Good transformation design will not remove every conflict. Some changes redistribute power, resources or risk. Resistance may then be legitimate. The alternative to labeling people is not unlimited participation or consensus at any price.

The alternative is more precise leadership: make the problem visible, state the non-negotiable constraints, clarify decision rights, incorporate relevant knowledge and test solutions reversibly where cause and effect remain uncertain.

This is business competence for a technological world. Once technology prioritizes work, prepares decisions or authorizes knowledge, implementation becomes a leadership responsibility. A functioning system is not enough. Leaders must ensure that people can use it meaningfully in context, recognize its limits and remain able to act when uncertainty arises.

The most productive response to “change resistance” may therefore be a counter-question: What would have to be true about our design for this behaviour to be rational?

Leaders who take that question seriously stop managing resistance—and start designing transformation.

Explore further

TU Wien Academy’s three-day compact programme Effectively Transforming Organizations examines how leaders can create orientation, distribute responsibility and organize change as a structured process of learning and experimentation.

For leaders responsible for implementing data and AI initiatives, the Executive MBA Operational Excellence & AI connects technology, processes, economics and leadership accountability.

What is change resistance?

Change resistance describes behaviour that delays, avoids, criticizes or rejects a planned change. The term does not explain the cause: the behaviour may result from uncertainty, a genuine conflict, insufficient value or poor transformation design.

How can leaders tell whether resistance is caused by poor transformation design?

Test whether daily value, roles, decision rights, accountability, data and knowledge quality, participation, training and escalation routes are clear. If non-use varies strongly by role, language, location or task, poor fit is more plausible than a general negative attitude.

How should resistance be managed during change?

First separate observed behaviour from assumed motives. Develop competing explanations, involve affected people with real design influence, and test a reversible change on a limited scale. Then decide whether the need is communication, capability building, redesign or an explicit conflict decision.

Why does employee participation matter in digital transformation?

Employees understand tasks, exceptions and risks that central project teams may miss. OECD case studies, opens an external URL in a new window indicate that direct involvement can reduce job-loss anxiety and increase willingness to engage with AI.

Is low AI adoption evidence of resistance to AI?

No. Low use may mean that the AI offers insufficient advantage, fails to fit a role or language, depends on poor knowledge sources or creates unclear risk. A recent HR case study found selective use by situation rather than a simple decision for or against AI.[3]

What is the difference between change communication and co-design?

Change communication explains a change. Co-design gives affected people influence over tasks, workflows, rules or safeguards. Feedback without a visible ability to alter the design is not full participation.