AI and innovation in business: the risk of the Efficiency Bubble
Category: Decision-Making | AI & Strategy
Reading time: 7 minutes
AI and innovation in business are often presented as an inevitable combination: more information, greater speed, more analytical capacity and, consequently, better ideas.
Part of this promise is real. Artificial intelligence can reduce research time, organise large volumes of information, support decisions and accelerate tasks that previously required significant effort.
But there is a less comfortable question: what happens when an organisation becomes highly efficient at producing, summarising, responding and deciding, but stops creating room to explore unexpected hypotheses?
The risk is not that AI will generate a wrong answer. It is that it will generate an answer that is plausible, fast and good enough for no one to feel the need to search for a less obvious alternative.
This is the risk I call the Efficiency Bubble: a situation in which the continuous pursuit of cognitive efficiency may progressively narrow the range of perspectives, questions and possibilities that enter strategic decision-making.
What is an Efficiency Bubble?
The Efficiency Bubble is neither a clinical diagnosis nor a phenomenon that has already been conclusively validated in organisations. It is a conceptual proposition developed as part of my research in Management to describe an emerging strategic risk.
The idea is simple: when an organisation uses AI primarily to reduce effort, accelerate synthesis, automate recommendations and make work flow more smoothly, it may gain speed without necessarily increasing the quality or diversity of strategic thinking.
The problem emerges when efficiency ceases to be a capability serving decision-making and becomes the sole criterion of quality. At that point, the organisation may continuously improve what it already knows how to do without noticing that it has stopped exploring what it does not yet know how to ask.
A note on rigour. The Efficiency Bubble is presented here as an analytical lens and a testable organisational hypothesis. It does not claim that AI inevitably makes companies less innovative. It proposes that the uncritical adoption of AI may reduce exposure to the unexpected and narrow the range of alternatives considered.
The tension between efficiency and discovery
A recent study on AI-mediated subscription services helps frame this tension. Kim and Park analysed the relationship between algorithmic curation, technological fluidity, cognitive efficiency and serendipity — the discovery of unexpected and potentially relevant information.
In the model analysed, cognitive efficiency was the main predictor of retention in the service. Serendipity, in turn, had no significant effect in the aggregate model. The authors also identified a negative trend between technological fluidity and serendipity: the more fluid and effortless the experience appeared, the lower the exposure to the unexpected tended to be.
The context of the study is not the strategic management of companies. It concerns the consumption of AI subscription services. Even so, it raises a relevant question for any organisation: when everything is optimised to be immediate, relevant and predictable, what happens to discoveries that depend on a certain degree of drift, contrast or interruption?
Not all friction is useful. Unnecessary bureaucracy, confusing processes and rework do not produce innovation. But some productive friction — pauses, constructive challenge, confrontation between perspectives and exploration of hypotheses — may prevent the organisation from mistaking the first plausible answer for the best available decision.
How this risk may appear in an organisation
The Efficiency Bubble does not appear as an obvious error. It normally emerges in teams that seem to be working well: they respond quickly, produce consistent documents, use good tools and have access to more information than ever.
The signs become visible when:
- meetings begin with an AI-generated summary and end without questioning the assumptions behind that summary;
- teams continuously improve existing processes but rarely explore structurally different alternatives;
- the same indicators, sources or frameworks are used to interpret every problem;
- AI recommendations are treated as the end point rather than as initial material for discussion;
- no one can explain which relevant hypothesis was rejected, why it was rejected or who challenged it.
This does not mean that the organisation is making poor decisions. It means that it may be deciding within an increasingly narrow range of possibilities.
This risk complements what we explore in Artificial intelligence in decision-making: greater clarity or greater error?. AI can reduce noise and increase capacity. But without criteria, context and human validation, it can also make errors more difficult to recognise.
Efficiency is not the same as exploration
In their research on hybrid problem-solving, Raisch and Fomina distinguish three ways of combining human and artificial intelligence:
- Autonomous search: AI explores solutions that are further removed from the team’s initial experience.
- Sequential search: AI helps refine or develop solutions that are close to what is already known.
- Interactive search: humans and AI influence one another and recombine knowledge throughout the process.
Sequential search is useful. It is often faster, less demanding and appropriate when the problem is well defined. The risk emerges when it becomes the only available way of working.
A team that uses AI solely to improve its current solution may enhance execution without expanding its search space. An organisation that only improves what it already does may become highly efficient within a model that is no longer sufficient.
Four practices for introducing productive friction
1. Separate moments of efficiency from moments of exploration
Not every task requires prolonged debate. Use AI to accelerate synthesis, documentation, preparation and initial analysis. But reserve specific moments for exploring alternatives that do not arise directly from the first answer produced.
A decision-making meeting may, for example, begin with an AI-assisted summary and then include a compulsory second stage: “Which non-obvious alternative are we failing to consider?”
2. Request challenge, not just confirmation
The most useful prompts do not only ask for solutions. They ask for opposition, limitations and failure scenarios.
- Which assumptions behind this recommendation may be wrong?
- What relevant information is missing?
- Which alternative would an organisation with an opposing business model choose?
- In which scenario would this decision produce the opposite of the intended effect?
The aim is not to create paralysis. It is to prevent the team from mistaking textual coherence for strategic robustness.
3. Diversify the people and sources involved in the decision
The best alternatives rarely emerge when everyone involved shares the same references, metrics and assumptions. Exposing a decision to people from different departments, customers, operations, HR, technology or sales may reveal variables that a single framework cannot see.
The same applies to the sources used by AI. An answer based only on generic or historical information tends to reproduce what is already dominant. Decision quality improves when the team introduces internal data, field experience, specific evidence and perspectives that the model would not receive by default.
4. Record the role of AI in relevant decisions
In decisions with significant impact, retaining the final recommendation is not enough. The organisation should be able to answer four questions:
- Where was AI used in the process?
- Which sources, data or assumptions informed the analysis?
- Which alternative was rejected and why?
- Who validated or challenged the outcome before the decision?
This record should not be used to create bureaucracy. It should make organisational learning visible. Without a memory of how a decision was made, the organisation can only repeat outcomes; it cannot improve its decision-making process.
The role of leadership and performance practices
This is not an exclusively technological issue. It is a matter of leadership, development and performance management.
When an organisation rewards only speed, output volume or the rapid achievement of targets, it naturally encourages more automatic use of AI. When it recognises the quality of judgement, the ability to question assumptions, risk identification and the exploration of alternatives, it creates different incentives.
This does not mean assessing every employee according to how sophisticated their use of artificial intelligence is. It means ensuring that, in roles where AI influences analysis, decision-making and coordination, the organisation does not lose the human capacity to interpret context, assume responsibility and recognise limitations.
It is the same difference between measuring and managing. Indicators, dashboards and outputs do not automatically improve execution. What improves execution is the organised response that the team builds from them, as we explain in Organisational performance: why measuring is not the same as managing.
AI and innovation in business should expand the decision space, not just accelerate the response
The relevant question is not whether AI reduces effort. It does, and that is one of its main benefits.
The question is different: what is the organisation doing with the capacity it has freed? Is it using that time to analyse more thoroughly, explore alternatives and make better-qualified decisions? Or is it simply producing more quickly within the same familiar categories?
Competitive advantage will not lie solely in the tool selected. It will lie in the design of the process around it: who asks questions, who challenges, what information enters the process, which alternatives are considered and who assumes responsibility for the final decision.
At VAGAR, we work with organisations that want to integrate AI, leadership and management practices without losing judgement, autonomy or the capacity for renewal. This involves developing teams, clarifying decisions and designing performance systems that value more than speed.
Explore VAGAR’s work in Organisational Development and Training and in Results, Performance Management and Recognition.
Evidence base
Duede, E., Teplitskiy, M., Lakhani, K. R., & Evans, J. (2024). Being together in place as a catalyst for scientific advance. Research Policy, 53, 104911.
Pires da Silva, L. (2026). Inovação e Estratégia na era da IA: a tensão entre Cognitive Efficiency e Serendipity e o risco de uma Efficiency Bubble nas organizações. Academic paper, PhD in Management, University of Évora.
Kim, B., & Park, J. (2026). Driving service stickiness in the AI subscription economy: The roles of algorithmic curation, technological fluidity, and cognitive efficiency. Journal of Theoretical and Applied Electronic Commerce Research, 21, 30.
Raisch, S., & Fomina, K. (2025). Combining human and artificial intelligence: Hybrid problem-solving in organizations. Academy of Management Review, 50(2), 441–464.
Lebovitz, S., Lifshitz-Assaf, H., & Levina, N. (2022). To engage or not to engage with AI for critical judgments: How professionals deal with opacity when using AI for medical diagnosis. Organization Science, 33(1), 126–148.
Leave a Reply