Why does innovation stagnate the more suggestion boxes you open? That is the question I encountered at the end of 2023 in a metal and machinery manufacturing plant in Konya, employing 378 people. The management board had established an ‘innovation committee’ two years earlier, organized quarterly idea-collection meetings, and had even purchased software to have employees fill in digital suggestion cards. The result after 14 months: 412 suggestion cards, 3 initiatives taken forward, 1 failed pilot. Engineers stopped attending the meetings; the form was being filled in the way you fill in a weather report. The problem had nothing to do with the quality of the ideas. The problem was that running an experiment had become prohibitively expensive.The central argument of this article is this: innovation culture is not an idea-collection mechanism — it is the discipline of lowering the cost of failure. Suggestion boxes, surveys, hackathons and innovation committees are all symptom management. The real obstacle sits in plain sight: if an employee who says ‘let’s do this differently’ feels that what they stand to lose outweighs what they stand to gain, they will quietly pocket that idea no matter how good it is. This is also measurable. By 2024, teams that combine Amy Edmondson’s psychological safety framework with internal data analysis can detect where the silence norm has taken hold — through sprint retrospective data, pull-request comment patterns or Slack message cadence. If a team lead reviews a new proposal in less than two minutes and the thread goes quiet immediately after, that is not engagement; that is compliance theatre.Framed this way, the tension between two distinct management philosophies becomes clear: idea-driven innovation versus experiment-driven innovation. The idea-driven approach centres the question ‘what should we do?’, assuming that if great ideas accumulate, innovation will eventually emerge. The experiment-driven approach reverses the question: ‘How do we learn most cheaply that this idea is wrong?’ The difference between the two is structural, not cultural. In the first, failure is concealed because its cost is high; in the second, failure is shared because it has been designed to stay small. In Turkey, SME managers typically apply the first model while expecting the outcomes of the second, which is why most innovation initiatives end up as theatre.A caveat must be stated before moving on: experiment culture cannot be built at the same pace or in the same form across every sector. Running ‘rapid experiments’ on a production line in the Konya plant without proper safeguards can easily become an occupational safety issue — and that risk is real. A manager at a pharmaceutical raw-material supplier in Ankara, with 295 employees, put it to me directly: ‘We operate within GMP boundaries. Every experiment has to be documented and cleared through an approval chain.’ The ‘fail fast’ consulting rhetoric that ignores this reality never gains traction in Turkish manufacturing. The experimentation space must be designed within the constraints of the sector, in a controlled and deliberate manner — not imported wholesale from a Silicon Valley playbook.So what does reducing the cost of experimentation actually look like in practice? Three structural frames are worth applying together. The first is the ‘small bet’ structure: before any idea is taken to full implementation, it goes through a four-to-six-week pre-experiment phase with minimal resources and a single metric. The team at the Konya plant piloted this using seven-week sprint cycles; a proposal for warehouse layout optimisation was tested on a single corridor before anything else was changed. This eliminated both the physical risk and the budget anxiety. The team had sufficient data to make a full deployment decision within three weeks. The critical point is this: even if the pilot had failed, those three weeks would not have been wasted — the cost of eliminating the wrong idea early would have been paid. The second frame is ‘learning goal’ definition: at the start of every experiment, the question is not ‘what do we want to prove right?’ but ‘what do we want to learn is wrong, as cheaply as possible?’ When the direction of the question shifts, the team’s psychology shifts with it; failure becomes data rather than blame. The third frame is outcome-independent recognition: the person who designs and runs the experiment is evaluated positively regardless of whether the result is negative, while bypassing or concealing the experiment process carries a negative consequence. When these three frames are not applied together, none of them works in isolation.Turkish companies approaching this topic in 2024 face an additional pressure: the integration of AI tools into operational processes. As generative-AI-based tools spread rapidly through the market, companies without an experimentation culture are purchasing the tools but failing to produce the transformation. The reason is cultural, not technical. Teams carry the anxiety of ‘what happens if we say this tool did not meet our expectations’ before even asking ‘how might we use this tool?’ Without a structure that makes small experiments safe, AI pilot projects follow the same fate — budget approved, tool purchased, results expected in three months, disappointment experienced, conclusion drawn that ‘AI does not work for us.’ The problem is not with the tool; it is with experimentation capacity. A logistics firm in Gaziantep working on a RAG architecture broke this cycle by eventually choosing to share its pilot failure openly: the whole team learned what had not worked, and the second pilot was completed 53 days faster than the first. Shared failure is always more productive than concealed failure. This specific dynamic is part of why AI governance discussions in Turkey in 2024 — accelerated by the EU AI Act adopted in April — are increasingly focusing not just on technical compliance but on the organisational readiness to run, document and learn from small AI experiments before scaling them.Back to the Konya plant. Whether that system, the one that accumulated 412 suggestion cards across 14 months, is still in operation today, I do not know. But I do know this: the reason it failed was not that the engineers lacked ideas. The reason was that the question of who would bear the cost of a wrong idea was left unanswered. Opening space for experimentation means managing that ambiguity — making it explicit who owns both the risk and the learning. Trying to build an innovation culture without resolving this structurally is like planting seeds in untilled soil: the seed may be excellent; it will not grow. Before your company plans its next innovation workshop, one question is worth sitting with: what actually happens when someone in that meeting proposes an idea that turns out to be wrong? If the honest answer is anything other than ‘not much’, the problem is not the idea-collection mechanism — it is the cost of failure.
This article was originally published in Turkish by Gökhan MERCANOĞLU on January 2, 2024. The English edition has been reviewed and edited by the author.