What Makes a New Practice Normal

• Two studies say what is needed; neither says how it is made
• Normalisation is not adoption
• Four conditions as dimensions of normalisation
• Why AI is a convenient case and not the only one
• Candidate mechanisms
• How the study is designed
• The Nordic dimension
• Conclusion
• References

We can say which conditions have to be present for change to hold. We cannot yet say what people do to produce them, and that gap is the subject of the next study.

Two studies say what is needed; neither says how it is made

The scale of the gap is not in dispute. In McKinsey’s 2025 State of AI survey, nearly two-thirds of respondents said their organisations had not yet begun scaling artificial intelligence across the enterprise, remaining in experimentation or piloting. The same firm’s earlier survey of more than a thousand companies on operating-model change found fewer than a third had moved beyond the pilot phase. Whatever is failing at that transition has been failing for long enough, and across enough technologies, to be structural rather than specific to any one of them.

Normalisation is not adoption

The standard frame for this problem is adoption. Train the users, deploy the tool, measure usage, close the project. Usage is measurable, so it becomes the target, and organisations end up with high login rates and unchanged behaviour.

The frame the next study uses is different. A practice has become normal when nobody has to decide to use it. Not when compliance is high, but when the deliberation has disappeared: when the new way is simply how the work is done, and using the old way would require a conscious choice and an explanation.

Berger and Luckmann, whose account of the social construction of reality underpins the first Bridgium study, describe the underlying process as habitualisation. An action repeated often enough becomes a pattern, the pattern can be performed without fresh deliberation each time, and eventually it is passed on to newcomers as simply how things are done here. That last step is the one organisations care about and the one no rollout plan contains.

The practical difference between the two frames is what you would measure. Adoption asks how many people used it. Normalisation asks whether anyone still has to think about using it, whether a newcomer learns it as procedure or as an initiative, and whether the old way has actually stopped. Those are harder questions, which is presumably why they are asked less often.

Four conditions as dimensions of normalisation

The first study proposes that legitimacy, predictability, connectivity and innovation memory can be treated as dimensions of normalisation, and that the next step is to make explicit the mechanisms through which organisations produce them. That framing is what the second study inherits and what the third would test.

Set out as a research agenda, the position is unusually clear about the boundary between what is established and what is not.

Table 1. The research boundary. Conditions from Bridgium, How Innovation Happens (2026); open questions as stated in its next research direction.

Why AI is a convenient case and not the only one

Artificial intelligence tools are the obvious setting for this study, and it is worth being explicit about why, because the reason is methodological rather than topical.

A study of normalisation needs change that is happening now, in ordinary work, at enough organisations to compare. AI tools qualify on all three counts. The change is live rather than remembered, it touches daily practice rather than a specialised function, and the surrounding gap between formal deployment and real use is well documented, which gives a study something to explain rather than something to discover.

But the research question is not about AI, and treating it as an AI study would be a mistake in both directions. It would narrow the findings to one technology, and it would invite the assumption that the mechanisms are technical. The same question applies to a reorganisation, a new process, a merger, a change of supplier, or any other case where an organisation is asked to work differently. If the mechanisms of normalisation are general, they should be visible across those cases, and comparing them is how you find out whether they are.

March’s distinction between exploration and exploitation is relevant here as a caution. A tool introduced as an experiment and evaluated as an operational system will fail that evaluation regardless of its merits, and the resulting judgement will be recorded as a fact about the tool. Part of what a study of mechanisms has to separate is what the organisation did from what the technology was.

Candidate mechanisms

The following are not findings. They are the hypotheses the two completed studies point toward, stated plainly so they can be tested rather than assumed.

• Visible use by people whose judgement is trusted. Legitimacy may be produced less by announcement than by observation: the practice becomes permissible when someone whose competence is not in question is seen using it without ceremony. This is testable, and if it holds, it changes who a rollout should target first.
• A named decision that closes the trial. Predictability may depend on the existence of a moment when the old way formally stops. Where no such moment exists, both ways remain live, and running two ways in parallel is more work than either, which supplies a rational reason to revert.
• Someone who translates between local versions. Every unit adapts a new practice to its own conditions, and those versions diverge. Granovetter’s account of weak ties suggests the linking is done by people connected across units rather than within them. Whether that role is assigned or emerges, and what happens when the person holding it leaves, is an empirical question.
• A record of what was tried and abandoned. Innovation memory may operate less through documentation than through a small number of people who remember why an approach was dropped. Nonaka and Takeuchi’s distinction between tacit and explicit knowledge suggests why written records underperform here, and Cohen and Levinthal’s work on absorptive capacity suggests what an organisation loses when the memory goes: not the record, but the ability to recognise a relevant idea next time.

Each of these is plausible and none of them is established. Kerr’s observation about rewarding one behaviour while hoping for another suggests a further line of inquiry that cuts across all four: whether any of these mechanisms survives contact with a measurement system that rewards deployment rather than use.

How the study is designed

The method follows the two completed studies, because comparability matters more than novelty here. Qualitative, semi-structured interviews with people living through a transformation now or recently, reconstructing concrete episodes rather than collecting opinions: how a new technology or practice was introduced, how it was interpreted, what was tested, what was adjusted, and how it was gradually incorporated into everyday work.

The analysis then looks at how new practices are made legitimate, how they become predictable, how local interpretations get linked across teams and departments, and how learning from experiments and earlier attempts is retained, circulated and reused. Comparative analysis across cases is what allows recurring mechanisms to be identified and different approaches to normalisation to be distinguished. The stated aim is to systematise and classify those mechanisms and describe how they operate in practice.

It is worth being equally clear about what this design cannot deliver.

Table 2. Scope and limits of the planned study. Method and aim as stated in the next research direction of Bridgium, How Innovation Happens (2026); limits stated on the same basis as the two completed studies.

The Nordic dimension

The Nordic setting is a useful place to study normalisation for a reason that has come up in both completed studies, and it cuts both ways.

Flat structures, low power distance and direct cross-unit contact mean that a great deal of coordination happens by conversation. New practices spread by someone showing someone else, and adaptation to local conditions is quick because permission is not required for it. For a study of mechanisms this is close to ideal, because the mechanisms are visible in behaviour rather than buried in process documentation.

The same feature makes the fourth condition fragile. Where knowledge travels by conversation, there is little incentive to write anything down, and the organisation’s memory of how a change was made is held by the people who made it. A systematic review of empirical work on knowledge loss identifies staff turnover as one of the most frequently reported causes of organisational knowledge loss, and the effect is sharpest where retention rested on people rather than artefacts. So Nordic organisations may be unusually good at producing normalisation and unusually bad at remembering how they did it, which would explain why the same organisation can absorb one change smoothly and struggle with the next.

Whether that is true is exactly the kind of claim the study should test rather than assert. It is stated here as a hypothesis, not a finding.

Conclusion

Two studies have established a diagnostic. Innovation moves where four conditions are present, and where it stalls, one of them is usually identifiable as missing. That is genuinely useful and it stops short of the question most people ask immediately afterwards, which is what to actually do.

Answering that requires a different object of study. Not the conditions, but the acts that produce them: what someone said in a meeting, what was decided and when, who was seen using the new thing, what stopped and on what date, and what was written down about why an earlier attempt failed. Those are small, unglamorous and specific, which is why they are absent from most accounts of transformation and why they need to be collected directly from the people who performed them.

So the question this research starts from, and the one worth asking about your own organisation. Think of a practice that is now completely ordinary where you work and that did not exist three years ago. Can anyone still say what made it ordinary?

Both completed studies are open. How Innovation Happens examines the flow inside organisations:
bridgium-research.eu/innovation-report-2026
From Discovery to Practice extends the analysis across the organisational boundary:
bridgium-research.eu/startup-report-2026

References

1. Bridgium Research Team (2026). How Innovation Happens: Insights from Leading Enterprises in Times of Change. Illarionova N., Verlin K., Verlin A. Report
2. Bridgium Research Team (2026). From Discovery to Practice: How Corporate–Startup Collaboration Becomes Usable. Illarionova N., Verlin K., Verlin A. Report
3. Berger, P. L. and Luckmann, T. (1966). The Social Construction of Reality. New York: Doubleday. Publisher
4. Nonaka, I. and Takeuchi, H. (1995). The Knowledge-Creating Company. New York: Oxford University Press. Publisher
5. Cohen, W. M. and Levinthal, D. A. (1990). Absorptive Capacity: A New Perspective on Learning and Innovation. Administrative Science Quarterly, 35(1), 128–152. JSTOR
6. March, J. G. (1991). Exploration and Exploitation in Organizational Learning. Organization Science, 2(1), 71–87. JSTOR
7. Granovetter, M. S. (1973). The Strength of Weak Ties. American Journal of Sociology, 78(6), 1360–1380. JSTOR
8. Kerr, S. (1975). On the Folly of Rewarding A, While Hoping for B. Academy of Management Journal, 18(4), 769–783. JSTOR
9. Weick, K. E. (1995). Sensemaking in Organizations. Thousand Oaks, CA: Sage. Publisher
10. McKinsey & Company (2025). The state of AI: Agents, innovation, and transformation. Read
11. McKinsey & Company (2020). Breaching the great wall to scale. Read
12. Knowledge loss induced by organizational member turnover: a review of empirical literature (Part I). The Learning Organization, 30(2). Emerald