6 thinking traps almost every SME has run into.

What this is about

After more than two decades advising software teams, marketing departments, and executive boards, I see the same phenomenon every day: what makes digital business projects fail is rarely the technology, and often not even the budget. It is the way we think. More precisely: how we systematically deceive ourselves without noticing. That is exactly why every piece of vision and strategy work we do begins with an honest look at our own thought patterns.

This observation became widely known thanks to Rolf Dobelli's "The Art of Thinking Clearly" — a book that breaks down 52 typical cognitive biases on two pages each, in an entertaining way. For the world of digital transformation in Swiss SMEs, you can carve out 13 cognitive biases that I encounter particularly often in advisory practice — and that can cost real money, time, and reputation.

"It is harder to crack a prejudice than an atom." — Albert Einstein

This series has two parts. Part 1 — the one you are reading — covers the six classics we know from everyday life that keep tripping us up in software projects and digital strategies. Part 2 dives into seven advanced pitfalls around incentive systems, outcome bias, scarcity, anchoring — and now also a cognitive bias that has only become a mass-scale reality with the arrival of generative AI in 2023–2026.

For each bias, you get the same practical structure:

1. What it is — what exactly is the bias?
2. Why does this happen to us? — the psychological root
3. Examples from business & SME practice — modern, concrete cases
4. How to prevent it? — counter-strategies that work in advisory practice

1. Overconfidence Bias — when your own ability gets wildly overestimated

What it is

A typical pattern: an executive team has had a few successful years, the app on the smartphone looks "doable, actually," and suddenly the sentence "We could build something like that — and better" sounds like a reasonable strategy. Overconfidence does not only hit us as clients, but also as project managers, developers, and designers — especially in effort and time estimates.

Why does this happen to us?

Our brain tends to filter out complexity as soon as a product's surface looks simple. On top of that comes the Dunning-Kruger effect: precisely in fields where we have little real expertise, we feel most certain. And with effort estimates, incentive sensitivity gets mixed in (see Part 2): whoever estimates more tightly seems to have a better shot at the contract — even if reality later turns bitter.

Examples from business & SME practice

• The "we'll do it better than Instagram" pitch: At AHEAD OF TIME, the request "We have this social platform idea, but better than Facebook — how expensive would it be?" lands regularly. What most people don't realise: Facebook has developed tens of millions of lines of code over the years, and a social network lives on network effects, not features. Without a sufficient user base, no value simply emerges.
• AI hype 2024–2025: Several Swiss mid-sized companies, after a few ChatGPT demos, decided to train their own foundation model to "become independent." The result: million-franc investments that fizzle out somewhere between "produces nice email drafts" and "replaces our sales process."
• Your own effort estimate: We all know the phenomenon — we calculate from best-case scenarios, forget reserves for the unforeseen, and are then surprised when a project takes 40% longer. That has nothing to do with sloppiness, but with the psychological tendency to see the optimum.

How to prevent it?

• Start from the worst case, not the best case. Pessimistic scenarios are statistically often the more accurate ones
• Get diverse input — different people estimate differently. Have someone who is not in tunnel mode on your project check your assumptions, ideally without showing them your own conclusion first
• Make assumptions explicit and test them against measurable goals. Implicit expectations are the start of many conflicts
• Plan in fixed reserves. A 20–30% buffer for the unforeseen in software projects is not incompetence — it is experience
• Socrates already knew: "I know that I know nothing." A dose of humility beats even the best method

2. Sunk Cost Fallacy — when the past hijacks the future

What it is

You are stuck in a project that just won't work. The numbers don't add up, the team is drained, the market is not responding. Still, you keep going — "because we have already invested so much." You give past expenditures a weight in decisions about the future that they rationally don't deserve.

Why does this happen to us?

Quitting means admitting we were wrong earlier. That is cognitively exhausting and emotionally painful. We humans also want to come across as consistent — externally and internally. Abandoning a project mid-stream feels like a break in that consistency, even when it would be the right decision. Better an end with horror than horror without end — the saying captures the core of this bias quite precisely. That a bold cut pays off is shown by our art24 case, where the move to a modern platform ended years of friction.

Examples from business & SME practice

• BlackBerry 10 — the textbook case: BlackBerry pumped hundreds of millions into its own BlackBerry 10 operating system and would not let go, even though Apple iOS and Google Android had long dominated the market. An earlier migration to Android would probably have saved BlackBerry from insolvency. But every step back would also have been confirmation that years of investment had been a mistake.
• The webshop platform that has been a nuisance for years: A classic from SME everyday life. An online shop that has caused trouble for three years — performance issues, maintenance nightmares, missing integrations. A migration to Shopify or a modern headless platform would cost a few months of pain and maybe CHF 60,000. But it would avoid the next five years of friction and lost revenue. Still, people stay — because of the already sunk investment.
• Concorde Fallacy: The Anglo-French Concorde was built for years even though it was clear that it would not be economically viable. The term "sunk cost fallacy" is often used synonymously with "Concorde fallacy" in the literature — both describe the same pattern.

How to prevent it?

• Define clear, measurable goals at the start — and tie decisions to them, not to perceived investment size
• Pull the ripcord early rather than late. The longer you wait, the more expensive the exit
• Bring in neutral voices. Whoever is not emotionally invested in the project sees more clearly
• Calculate opportunity costs honestly: what else could the budget contribute if it flowed into another project?

3. Confirmation Bias — when only what supports your own view counts

What it is

You have a thesis about your target group, your product, or market behaviour. And suddenly the whole world seems to confirm this thesis — studies appear, anecdotes fit, trends confirm. At the same time, you ignore signals that argue against your thesis, or explain them away as special cases. That is confirmation bias in its purest form. A consistent customer-first approach is a good antidote here, because it forces you to think from the customer's perspective rather than from your own assumption.

Why does this happen to us?

An existing belief is part of our self-image. Questioning it means questioning ourselves. That is cognitively exhausting and emotionally uncomfortable. The brain automatically chooses the easy path: what fits gets noticed and weighted; what doesn't fit gets talked down or ignored. This becomes particularly expensive in conversion optimisation, where selectively read data can quickly steer entire campaign budgets in the wrong direction.

Examples from business & SME practice

• Public discourse: Fascinating to observe — no matter the topic, every position effortlessly finds enough material online to feel confirmed. Anyone sceptical of a technology immediately finds the critical voices; anyone in favour finds the euphoric ones. We preferentially consume what proves us right.
• The software testing trap: A subtle case in engineering: developers tend to test that their code works — positive tests. Negative tests, which are supposed to expose weaknesses, are less popular because they feel like a "refutation" of one's own work. That is exactly why good engineering teams work with largely independent QA teams.
• App performance distortion: Marketing or product owners who collect only data that proves the success of their strategy, and either never look for signs of failure or classify them as outliers. Disconfirming evidence — data that argues against one's own assumption — is systematically filtered out.

How to prevent it?

Albert Einstein had a wonderfully pragmatic recipe: he deliberately invested energy in finding evidence against his own theses — before anyone else did. Translated into SME everyday life that means:

• Lock in hypotheses before data. Already before a test, define: "Which data would convince me of the opposite?"
• Institutionalise the devil's-advocate role. In strategy meetings, rotate who is tasked with dismantling every assumption
• Cross your data sources: GA4 against CRM against real customer conversations. Whoever uses only one source mostly confirms themselves
• Independent reviews for important decisions — from code review to strategy audit by external advisors

4. Authority Bias — questioning authorities critically, especially in the AI age

What it is

Stanley Milgram's famous 1961 experiment showed impressively how strongly people trust the judgement of an entity perceived as an authority — even against their own conscience. In the digital business context, the picture broadens: we trust tech thought leaders, industry influencers, established companies — and recently also AI tools that simply sound competent.

This bias has two modern variants:

1. The classic: blind trust in human authorities (industry heavyweights, consultants, textbooks)
2. The new one: blind trust in AI-generated answers because they sound confident, structured, and plausible

Why does this happen to us?

Deference to authority is evolutionarily anchored — groups with clear hierarchies survived better. Our brain therefore likes to take mental shortcuts: if a source "seems important," we spare ourselves our own analysis. With AI tools, something new is added: the language itself feels authoritative. Precise phrasing, structured arguments, source citations (sometimes invented) — all of that activates the same trust heuristics as a good human expert. Plus: if we have invested hours in a ChatGPT prompt, we want the answer to be correct. Cognitive dissonance further reinforces our trust.

Examples from business & SME practice

Classic authority examples:

• Zuckerberg's chatbot prophecy 2016: At the Facebook F8 conference, Mark Zuckerberg announced that smart chatbots would largely replace native apps within a few years. The result: hundreds of companies — also in Switzerland — jumped on the train, founded chatbot agencies, pushed native roadmaps aside. By 2026 we see: the app market continued to grow happily, and truly widespread chatbots are few. A single tech authority was enough to mislead an entire industry.
• Blockchain and crypto hype 2017–2022: Another classic. Swiss mid-sized companies built blockchain strategies around the statements of crypto thought leaders, without critically scrutinising the statements. Many of these strategies never led to a viable business model.

Modern AI authority examples (NEW 2024–2026):

• Mata v. Avianca (New York, 2023): A US lawyer had ChatGPT research precedents for a legal brief — and submitted the citations to court unchecked. The problem: the AI had invented the cases. Sanctions, ruined reputation, international cautionary tale. The central weakness was not ChatGPT's hallucination (we'll get to that in Part 2, bias 13). The central weakness was believing it unchecked — classic authority bias with a new sender.
• Sam Altman, Dario Amodei & co. as guides: Statements from AI lab heads ("AGI in 5 years," "50% of all knowledge work replaceable by 2030") massively influence executive decisions in Swiss SMEs. Maybe they're right. Maybe not. Critically thinking through both possibilities is the mandatory homework — not amplifying the echo of the loudest voice.
• ChatGPT statistics in strategy presentations: "According to ChatGPT, 73% of Swiss SMEs already use AI." Source: nowhere verifiable. Still, the number lands on executive slides, in talks, even in press releases. That is not AI hallucination as the problem — the problem is following the AI unchecked.
• Junior developers with full Copilot/Cursor trust: A young team member adopts AI code suggestions unchecked, without really understanding why a particular solution was chosen. The codebase collects subtle bugs that only a strong senior review culture can catch.

How to prevent it?

For classic authorities:

• Check your own assumptions. Was this a decision from your own analysis, or from trust in an authority? Which authority was that, and where do I get my belief in its credibility?
• Actively seek dissenting voices. For every hype topic there are serious sceptics — they are just less loud than the enthusiasts
• Diversity in decision bodies. A homogeneous group amplifies authority bias. Heterogeneous teams question more readily

Specifically for AI tools (this is new):

• Treat AI output as a first draft, not as a final product. What you cannot verify yourself does not belong external (to clients, into contracts, into press releases) unchecked
• Cross-check with primary sources for every statistic, every quote, every legal or technical statement
• Critical-prompt pattern: After the first AI answer, follow up concretely: "Where could you be wrong? Which assumptions did you make that I should check? List your uncertainties." Surprisingly often, usable self-corrections emerge
• Establish a team rule: No AI output that has not been understood and verified by a human goes external. Understanding before speed
• Multi-model cross-check on critical statements: the same question to Claude + ChatGPT + Gemini. With significant divergences, deeper checking pays off

5. Survivorship Bias — when only the winners' story gets told

What it is

You read the next success story of a startup, a product, an app — and think: "Should be doable, look how they pulled it off." What you don't see: the hundreds or thousands of projects that crashed and burned with the same idea, the same setup, the same strategy. The successful ones enjoy the visibility, the failed ones disappear quietly in the background.

Why does this happen to us?

Success is visible, failure is quietly wound down. Media, LinkedIn, conference talks — everything favours the winners' stories. Our perception is optimised for what is visible, not for what disappears invisibly. From this comes a systematic distortion: we overestimate our own chances of success because we see only other people's successes. Anyone who plans growth in a structured way instead — for example along a growth strategy matrix — forces themselves to also honestly evaluate the risky fields.

Examples from business & SME practice

• Rovio and Angry Birds — the true story: The Finnish game studio Rovio may look like an overnight success. It wasn't. In 2009, Rovio was on the brink of bankruptcy and still employed 12 people. 51 games they had developed before — none successful. Angry Birds was game number 52. The 51 failures are practically unknown today, because Rovio became the remembered brand through Angry Birds. Had they given up after game 30 or 40 — we wouldn't know today that Rovio ever existed.
• The tech graveyard: Nokia, BlackBerry, Microsoft Windows Phone, MySpace, Xing, Google+, Google Stadia, 3D TVs, Quibi, Theranos, FTX. Every one of these brands was once celebrated as "the next big thing." Today they are case studies — if remembered at all. Whoever orients themselves by successful brands should mandatorily study this list too.
• AI startup hype 2023–2026: A few visible successes (OpenAI, Anthropic, Mistral, Lovable) obscure hundreds of failed AI startups that barely made it past the seed phase. An SME planning its "own AI strategy" while orienting only by the surviving stars systematically calibrates its expectations too high.

How to prevent it?

• Study the graveyard too. Before any major undertaking, find three comparable failures in the same market and honestly analyse — what didn't work there?
• Calculate your own probability of success conservatively. If the "comparable successes" are more exception than rule, that should be visible in the plan
• Make assumptions explicit that must hold true for success — and regularly check whether they still hold
• Play out the worst-case scenario. If we're going to estimate optimistically anyway, better to hedge twice than to hope once
• Pre-mortem exercise: Play it out in the team — "The project has failed in 12 months. What was the most likely cause?" Brings surprisingly realistic answers

6. Swimmer's Body Illusion — confusing cause and effect

What it is

You look at successful swimmers and see that impressive, athletically shaped body. The obvious conclusion: "If I swim enough, I'll look like that too." What is easily overlooked: that body type was probably already the precondition for this person becoming a professional swimmer — not the result of training. Cause and effect are confused.

Why does this happen to us?

Our brain loves simple causal chains — A leads to B. More complex or reversed connections are cognitively exhausting. On top of that: books, talks, and consulting offers that sell simple success recipes are much more popular than honest analyses that admit that success is multi-causal, often luck-dependent, and rarely replicable.

Examples from business & SME practice

• "7 steps to success": There are countless books and trainings claiming to have found the success recipe — "The Google Principle," "How Apple Thinks," "The 10 Habits of Successful Founders." What the books rarely mention: hundreds of companies have applied the same principles and still failed. Because it wasn't the principles that made the difference, but dozens of other factors.
• Waterfall versus Agile — the eternal method debate: Some wildly successful products emerged in strict waterfall mode. Other equally successful ones with a fully agile setup. Anyone claiming one method is the key is running Swimmer's Body Illusion. In truth, team quality, market timing, and customer-problem clarity contribute far more than the choice of project method.
• Startup founder stories 2023–2026: "How we built Stripe / Notion / Lovable" — almost every one of these stories is shot through with survivorship bias and retrospective smoothing of failure phases. Reading recommendations are fine — but not as 1:1 recipe books.

How to prevent it?

• Before copying, ask: What was cause, what was effect? Did the successful ones have preconditions I don't have? Which ones?
• Be critical of simple recipes, especially when sold as "the" key to success. Complex reality needs differentiated analysis
• Success is often hard work plus luck plus timing. Whoever internalises this calculates more realistically and is less susceptible to inflated expectations

Interim conclusion Part 1

These six biases are the pitfalls that come up most frequently in SME advisory practice. Overconfidence, clinging to sunk costs, selective perception, deference to authority (classic and new in AI), the lure of the winners' stories, and confusing cause with effect — they have all cost concrete projects concrete money in recent years.

The good news: those who know them spot them earlier. Those who spot them early can steer against them.

What these six have in common: they are general biases that have been described for decades. In AI cognitive biases we go deeper into more specific pitfalls — incentive systems, outcome bias, illusion of control, cognitive dissonance, anchoring effect, scarcity fallacy. And one bias that has only become a mass-scale reality with generative AI in 2023–2026: the hallucination of AI systems and how SMEs deal with it.

How to apply this with your team

Run a short 60-minute check with your team:

1. Go through the six and honestly ask: "Which two of these are hitting us hardest right now?"
2. For each prioritised bias, agree on one concrete counter-strategy for the next four weeks — with a responsible person and check-in date

Small and concrete beats big and vague. Whoever wants to address all six simultaneously will fail at step one. If you want a neutral outside view on your decision processes, our advisory mandates help exactly at this point.

AI as a reflection tool — with a clear caveat

AI tools can help to spot your own biases — provided the team knows how to use them critically. What structured knowledge transfer for this looks like is shown by our HSO coaching mandate. Practical applications in 2026:

• Pre-mortem with Claude or ChatGPT: "Here is my project. Imagine it has failed in 12 months. List the 10 most likely causes, ordered by probability." Often surfaces points you wouldn't have thought of yourself
• Disconfirming-evidence search: "Here is my thesis on [market / product]. What counter-arguments exist? Which data would refute my thesis?"
• Bias audit: "Which cognitive biases could distort my view on [topic]?" — AI knows the cognitive-bias vocabulary well and can point things out surprisingly precisely

Important: AI has its own biases. See Part 2, bias 4 above (authority bias with AI) and bias 13 (hallucination). The tool is useful, but not neutral.

Q&A — the most frequent questions

Which bias hits SMEs most often? In my advisory practice: sunk cost fallacy coupled with confirmation bias. SMEs stick too long with tools, suppliers, or strategies that no longer fit — and at the same time always find reasons why the data "actually does kind of fit."

How do I recognise that I'm currently stuck in a mental trap? Watch out for decisions that feel emotionally right but are rationally hard to justify. Phrases like "We've always done it this way" or "We've come this far already…" are statistically reliable early warning signals.

Why a new edition on cognitive biases right now? Three reasons: first, the old biases have been amplified, not weakened, in the digital context. Second, AI tools bring their own new variants that didn't exist three years ago. Third, in advisory work in 2026 I see significantly more Swiss SMEs concretely suffering from these patterns.

Are there tools against biases? Yes — structured decision templates (pre-mortem, decision matrix, devil's advocate), clear KPI dashboards (no vanity metrics), external advisory as regular outside view, and increasingly in 2026 also AI tools as sparring partners (with a caveat).

Sources and further reading

• Dobelli, R. (2011). Die Kunst des klaren Denkens. Hanser. — Grundlage für die in diesem Beitrag besprochenen sechs klassischen Cognitive Biases.
• Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux. — Vertiefung zu System-1- vs. System-2-Denken.
• Milgram, S. (1963). Behavioral Study of Obedience. Journal of Abnormal and Social Psychology, 67(4), 371–378. — Klassisches Experiment zum Authority Bias.
• Tversky, A. & Kahneman, D. (1974). Judgment under Uncertainty: Heuristics and Biases. Science, 185(4157), 1124–1131. — Fundamentale Arbeit zu Cognitive Biases.

This list covers the primary sources for the concepts discussed in this article. It is not exhaustive but should provide an entry point for those interested in going deeper.