Why 'Tolerance for Errors' Keeps Me Up at Night
Navigating the impossible tension between innovation and reliability when building products people truly depend on
Let me tell you what pisses me off: this whole "tolerance for errors" bullshit that's infected tech culture. Building health tech creates an impossible situation. My users depend on our app for their well-being—we can't just "move fast and break things" like we're making some throwaway photo filter. Yet we still need to innovate as rapidly as Spotify or Netflix to stay competitive.
This isn't abstract for me. At Welltory, we have 11 million real people using our app to monitor their health. They trust us. What happens if we get it wrong?
But every time I express this concern, some smug advisor inevitably chimes in with: "You need to build a culture that's safe to fail in." As if celebrating failure was somehow the secret sauce of innovation.
The Innovation-Reliability Paradox
This fundamental conflict deserves closer examination. Let's use Goldratt's Evaporating Cloud framework to understand it better:
+-----------------------------+
| Our Goal (A): |
| Trusted & Innovative App |
+-----------------------------+
/ \
/ \
+--------------------+ +----------------------+
| Ensure Trust (B) | | Innovate Fast (C) |
+--------------------+ +----------------------+
| |
+----------------+ +------------------+
| Avoid Risk (D) | | Embrace Risk (D')|
+----------------+ +------------------+
\ /
\__________________/
CONFLICT
Let me break this down:
Common Goal: Build a product users love and trust that grows fast and sustains advantage
Requirement A: We must be reliable and trustworthy (especially in health)
Requirement B: We must move fast and innovate, or we'll be outpaced
Prerequisite to A: Be a perfectionist, validate heavily, reduce all risk
Prerequisite to B: Ship fast, take risks, experiment aggressively
Conflict: These approaches directly contradict each other
The traditional solution? "Fail fast, fail often." This mantra suggests that moving quickly and accepting failures will ultimately lead to greater innovation and success.
But something about this never sat right with me. To navigate this tightrope, I've hired people with high responsibility standards, neurotic perfectionism, and demanding self-expectations. The flip side? Lower risk tolerance and decreased velocity. Yet I simultaneously want innovation, competitive advantages, and high-speed movement.
These requirements seem contradictory. Seem being the operative word.
What really bothers me – keeps me up at night – is how the phrase "tolerance for errors" has taken root in our industry. As if failure itself is somehow a virtue. As if Edison was proud of those 10,000 ways that didn't work.
Honestly, what the hell?
Let's Examine the "Fail Fast" Assumption
To resolve the paradox, we need to challenge our assumptions. Let's start with the most widely accepted one: that failing fast and often leads to innovation and success.
1. The Birth of a Dangerous Mantra
The "fail fast" ideology has a surprisingly twisted history. It's like a game of telephone where "be methodical" somehow morphed into "set your money on fire."
It started innocently enough with Thomas Watson Sr., IBM's chairman, who reportedly said: "The fastest way to succeed is to double your failure rate."
Sound familiar? Here's the first delicious irony: Watson's IBM wasn't some hippie commune celebrating failures with participation trophies. IBM under Watson was terrifyingly disciplined, with men in pressed white shirts creating five-year plans with military precision. They dominated the mainframe era through carefully calculated, decade-long bets like the System/360 project—not through random shots in the dark while shouting "YOLO!".
Watson was essentially saying "try more things," but within a company culture that demanded those attempts be thoughtful and strategic. It was never about glorifying failure itself.
Even Edison's famous quote about finding "10,000 ways that won't work" is misinterpreted. Edison wasn't randomly failing—he was conducting methodical, documented experiments in a dedicated laboratory. This wasn't a celebration of failure. This was a scientific method with borderline obsessive record-keeping.
2. Silicon Valley's Distortion Machine
The modern "fail fast" mantra gained momentum with Eric Ries' Lean Startup methodology. The original intent was reasonable: validate hypotheses quickly through minimum viable products, gather feedback, and pivot if necessary before burning through resources.
But something got lost in translation:
2009: The first FailCon launches in San Francisco—a conference dedicated to celebrating startup failures.
2013: "Fail Fast, Fail Often: How Losing Can Help You Win" book is published, completing the transformation from methodology to mindset.
2000-2014: Facebook promotes "Move fast and break things" as their motto.
2014: Even Zuckerberg abandons this approach, stating: "We've changed our motto to 'Move fast with stable infrastructure.'" Less catchy, but more reflective of what works when users depend on you.
In just a decade, what began as "iterate and validate quickly" had morphed into "failing is inherently virtuous." The means had become the end, like a diet plan that forgets the goal is health, not just eating kale.
3. The VC Factor: Portfolio Math vs. Founder Reality
Perhaps the most fascinating aspect is who became the loudest proponents of "fail fast" culture: venture capitalists. And this makes perfect mathematical sense—for them.
A VC's entire business model is built on portfolio theory: if you invest in enough companies, a few massive winners will offset the many losers. According to Cambridge Associates data, VCs typically anticipate that 65-75% of their investments will return 1x or less, while just 5-10% need to deliver those coveted 10x+ outcomes.
"The harsh reality is that VCs aren't in the business of minimizing failure; they're in the business of finding unicorns. Everything else is marketing." — Jason Lemkin, SaaStr Founder
What's rarely discussed is how VCs weaponize the "failure is normal" narrative as marketing to their limited partners. When portfolio companies fail (as most do), these failures are framed as "expected" and "educational"—not as poor investment decisions. This clever positioning helps VCs maintain credibility when raising subsequent funds despite mediocre performance.
The data confirms this reality. Carta's research shows that "less than 10% of 2021 funds had reported any cash distributions to investors after three years." Meanwhile, Silicon Valley Bank's LP survey revealed that 75% of LPs received fund extension requests, with many funds significantly exceeding their promised timelines.
This narrative serves VCs brilliantly, but it's terrible advice for individual founders, who don't have the luxury of a 20-company portfolio where 17 can fail. It's like a general telling soldiers it's okay if 90% of them die in battle because the army as a whole will be fine.
"When VCs praise failure as 'learning,' they're speaking as portfolio managers, not as individuals with one career and limited time. Founders should remember this crucial distinction." — Fred Wilson, Union Square Ventures
The psychological impact on founders is significant. First Round Capital's research found that 70% of founders experience depression or anxiety—nearly twice the rate of the general population. They bear the full weight of failure without the portfolio diversification that makes "fail fast" comfortable for their investors.
When VCs advise founders to "fail fast," they're essentially encouraging high-variance strategies that maximize the chances of outlier outcomes. For the VC with 30 companies in a portfolio, this approach is rational. For the founder with just one company and career, perhaps less so.
What Science Actually Says About Failure
When I first started questioning the "fail fast" dogma, I assumed someone must have done the research that supports it. Surely Silicon Valley's most sacred mantra was backed by mountains of peer-reviewed studies, right?
So I dug into the research. And oh boy, was I in for a surprise.
The Hard Data: Does Previous Failure Predict Future Success?
Here's a sobering reality check: multiple academic studies have tried and failed to find evidence that previous entrepreneurial failure leads to future success. Not only is there no positive correlation – in some cases, there's actually a negative one.
A comprehensive Harvard Business School study by Gompers, Kovner, Lerner, and Scharfstein analyzed hundreds of venture-backed startups and reached a conclusion that flies in the face of Silicon Valley wisdom:
"Entrepreneurs with a track record of success were much more likely to succeed than first-time entrepreneurs and those who had previously failed."
Their data showed that previously failed entrepreneurs were statistically no more successful in their next ventures than first-time entrepreneurs. You can find the study here.
Even more revealing was Professor Francis Greene's research at the University of Edinburgh. His team studied nearly 8,400 German startups over several years and found something shocking:
"The idea that you need to fall in order to rise is a myth. If you've fallen once, you're more likely to fall again."
Greene's team found "no indication that business failure triggers a reflection process in which entrepreneurs look back on mistakes they have made and adapt their future behaviour accordingly." You can read about this research here.
Additional research from the Centre for European Economic Research (ZEW) puts hard numbers on this phenomenon, showing the very small difference in success rate between experienced and failed entrepreneurs:
Failed entrepreneurs: 17.7% success rate in next venture
First-time entrepreneurs with the same VC backing: 14.3% success rate
These findings directly contradict the comforting narrative that failure is some entrepreneurial badge of honor or a necessary stepping stone to success. It's as if we've been telling runners that breaking their legs is great preparation for winning a marathon.
Normalizing vs. Analyzing: The Critical Distinction
In 2018, the Journal of Business Venturing published what I consider the smoking gun in this debate. Researchers Danneels and Vestal studied 106 manufacturing companies to determine what actually made the difference in innovation outcomes.
They distinguished between two approaches to failure:
Normalization – simply accepting failures as natural and not punishing them (basically what most "fail fast" advocates preach)
Analysis – systematically examining failures to extract actionable insights (what actually works)
The results? Companies that merely normalized failure without proper analysis showed no significant improvement in innovation outcomes. Zero. Nada. Might as well have done nothing.
But companies that rigorously analyzed their failures – within a culture of open discussions and "constructive conflict" – demonstrated substantially higher rates of successful product innovation.
This is like discovering that going to the gym doesn't make you stronger – it's the specific exercises you do while there that matter. Just showing up and hanging out by the water fountain doesn't build muscle.
The conclusion is clear: it's not about failing more; it's about learning more from each failure. Quality over quantity, people.
Our experience: Not All Failures Are Created Equal
One of the biggest flaws in the "fail fast" dogma is that it treats all failures as equally valuable. But in reality, some failures teach us volumes while others teach us nothing. Some are cheap; others are catastrophically expensive. Some are reversible; others are permanent.
Here's a story that taught me this lesson the hard way:
Early in Welltory's development, we ran two parallel experiments. First, we tested a small UI change in our onboarding flow—moving a button and tweaking some copy. No big deal if it failed. Second, we updated our stress assessment algorithm to make it scientifically more robust and less sensitive to data fluctuations.
The UI experiment failed but gave us clear data on user preferences. A $500 mistake that saved us $5,000 in future development. Classic "good failure."
The algorithm experiment proved far more dangerous. From a scientific perspective, the new algorithm was superior—it reduced sensitivity to random fluctuations and provided more stable measurements. But what we didn't anticipate was the subtle impact on user experience. The changes in value distribution varied across different user cohorts and emerged gradually over time—making the problem exceptionally difficult to detect.
Users couldn't see how their stress levels shifted in response to daily events—like a tense conversation with their mother-in-law or a moment of relaxation. The core issue wasn't about psychological satisfaction; it was functional. Users rely on these dynamic measurements to learn how their bodies respond to different situations throughout the day. They need sensitivity to these fluctuations to build body awareness and make meaningful connections between their activities and physical responses. The challenge wasn't choosing between scientific accuracy and user preference, but finding a way to serve both the learning process and scientific validity.
We first noticed unusual patterns in our financial metrics, but the cause remained elusive. By the time we connected the dots, the experiment had cost us hundreds of thousands of dollars. We ultimately solved this by creating a separate "wellness" parameter that remained stable for scientific accuracy, while preserving the responsiveness users expected in our main stress metric.
The critical lesson wasn't just about user preferences—it was about risk categorization. Changes to core algorithms require fundamentally different approaches than surface-level experiments. They need extensive modeling, careful rollouts, and sophisticated monitoring systems capable of detecting subtle, cohort-specific shifts.
Same "fail fast" approach, wildly different consequences. It was like comparing a paper cut to a chainsaw accident.
The Core/Edge Architecture of Risk
Think of your product like a body: you can safely experiment with your haircut, but not with your heart.
In any serious product — especially in health — you need a clear split:
• The Core: things like scientific analytics, user trust, and security. These require surgeon-like precision, careful changes, and close to zero tolerance for failure.
• The Edge: UI, engagement tricks, growth tests — here, bold experiments and fast iteration are not just welcome, but necessary.
Jeff Bezos described this distinction in his 2015 Shareholder Letter:
“Some decisions are consequential and irreversible – one-way doors – and these must be made carefully, slowly, with great deliberation. But most decisions aren’t like that – they’re two-way doors. If you’ve made a suboptimal Type 2 decision, you can go back.”
The message: Protect the core. Experiment at the edges. Know the difference.
Revisiting Our Paradox: New Solutions
Now that we've questioned the overly simplistic mantra of "fail fast," it's time to return to our original paradox: how can we reconcile the demand for both reliability and innovation—especially in consumer health products where lives and trust are at stake?
The key is understanding that not all decisions carry the same weight. A structured approach based on two dimensions — reversibility and impact—provides a way to move fast where it’s safe, and tread carefully where it’s not.
1. The Reversibility-Impact Framework
Inspired by Jeff Bezos’s distinction between "one-way" and "two-way" doors [2015 Shareholder Letter], this model categorizes decisions into four types:
Type A: High reversibility, low impact like A/B testing a homepage headline. Move fast with minimal oversight.
Type B: High reversibility, high impact like changing pricing with a rollback option. Test cautiously but decisively.
Type C: Low reversibility, low impact like restructuring backend infrastructure with limited user exposure. Use research-informed judgment.
Type D: Low reversibility, high impact like migrating platforms or shifting product strategy. Requires extensive diligence and alignment.
This matrix closely mirrors the [Speed-Reversibility Matrix], which categorizes decisions by risk and reversibility into four modes: fast mode, multiple testing, gradual rollout, and slow mode.
It also aligns with the [A3 Life Design decision matrix], which uses consequence and reversibility as key axes.
While we weren’t able to verify the often-cited Wharton statistic about 37% revenue uplift from structured decision-making, related insights from McKinsey emphasize that creating clear categories for decisions—like big bet, cross-cutting, delegated, and ad hoc—helps leaders move faster and more effectively.
That’s not just a productivity tactic—it’s a cultural operating system. When a product team proposes a change, the first question should be: "What type of decision is it?". It’s not that easy, sometimes it’s hard to understand that you are actulally making a decision.
Other helpful decision-making models include the [Cynefin framework] for navigating complexity and ambiguity, and the [RAPID framework] for clarifying roles in high-stakes decisions.
The bottom line: Smart product leadership isn’t about avoiding failure. It’s about managing the type and cost of being wrong.
2. Learning Velocity: The True North Star Metric
Instead of obsessing over how often you fail, elite product teams focus on how quickly they learn. This distinction is critical.
As Chris Jones from SVPG explains, the idea of "failing fast" has often been misinterpreted. The real goal isn’t to fail frequently — it’s to maximize the rate of learning. High-performing teams don't celebrate failure for its own sake; they celebrate what they learned and how quickly they can apply it to improve outcomes.
Netflix provides a great example. Their culture isn’t about tolerating failure — it's about enabling informed, responsible risk-taking. They emphasize clarity on what’s reversible and what’s not, and they optimize for learning velocity — the speed at which they extract actionable insight per unit of time and resources.
That mindset helped them evolve from DVD rentals to global streaming and original content creation — not by failing fast, but by learning fast and making informed nig bets. Like an original content creation was a bet for $5B. Expensive bold bet.
3. The Experiment Design Revolution
The most innovative companies don't run haphazard experiments; they design them with scientific precision. Consider how Airbnb approaches experimentation:
They use a process called "Experiment-Driven Development" where every test must have:
A clear, falsifiable hypothesis
Predefined success metrics
Minimum sample size calculations
Mechanism to isolate variables
When Airbnb tested their host verification process, they didn't just "try something new." They carefully crafted multiple variants, rolled them out to statistically significant sample sizes, and measured not just conversions but downstream effects on trust.
The result wasn't just a 43% improvement in verifications; it was a comprehensive understanding of why the improvement occurred, which they could then apply elsewhere.
This is worlds apart from the "spray and pray" approach suggested by simplistic "fail fast" advocates.
4. Engineering a Culture of Intelligent Risk
Translating these principles into organizational culture requires more than just lip service. It demands deliberate design of team structures, incentives, and communication patterns. Not all teams should approach risk the same way:
Growth & Conversion Teams
Focus: Funnels, A/B tests, onboarding
Approach: Fast iterations, many small tests, minimal supervision
Mantras: "Optimize Relentlessly. Break Nothing." and "Small Moves. Fast Gains."
Product Feature Teams
Focus: Core features, main screens
Approach: Balanced risk, medium-sized bets
Mantras: "Think Bigger. Miss Less." and "Be Bold. Be Right."
Strategic Bet Teams
Focus: New directions, architecture, business models
Approach: Thorough research, few well-chosen bets
Mantras: "Outthink. Outlearn. Outlast."
Stripe exemplifies this model. Their core payment infrastructure teams operate with extreme caution, while their new financial products teams move faster with more experimentation. Both are valued equally, but with different metrics and expectations.
That’s it: Resolving Our Paradox
Returning to our original Evaporating Cloud paradox, we can now see the hidden assumptions that created the false conflict:
False Assumption #1: "To be innovative, we must accept a high failure rate"
Reality: Innovation requires smart experimentation and learning, not blind tolerance for failure.
False Assumption #2: "To be reliable, we must never make mistakes"
Reality: Reliability means making mistakes in safe areas while protecting core functions.
False Assumption #3: "All parts of our product have the same risk profile"
Reality: Different aspects require fundamentally different approaches to risk.
Once we recognize these false assumptions, the paradox dissolves. We don't need to choose between reliability and innovation – we need to apply the right approach to each domain of our product and business.
So, let’s put “fail fast” to rest — and build something better in its place.
The truth is, failing isn’t the goal. Learning is. And not all failure leads to learning. If it did, we’d all be geniuses by now.
Great teams don’t just tolerate mistakes — they design smarter experiments, move with informed speed, and make sure they extract meaning from every outcome. They build systems that turn failure into insight, and insight into advantage.
Because in the end, you don’t win by failing faster.
You win by thinking better, deciding faster, experimenting smarter, and learning relentlessly.
That’s the edge.