When Research Lies to Us
Why do we make the wrong decisions even with the "right" research? Mistaking “confirmation bias” for research.
This Is A Continuation Of The Previous Article
One pattern has appeared in project after project. A direction is chosen first, research is conducted second, and the findings almost always seem to support the original decision. The team becomes increasingly confident that they are building the right product, yet the final result still struggles in the market.
After writing the last article Rethinking Market Opportunity, I was questioning again why people collect “false evidence” for assumptions.
A common journey that I have experienced:
The decision (about the product) comes first.
The research comes second.
The confidence comes third.
The market comes last.
The issue is, companies decide what they want to build, then research teams naturally collect evidence that supports the decision.
The Real Enemy: Confirmation Bias
Humans naturally look for evidence that supports what they already believe. We also have a remarkable ability to hear what we want to hear, even when the words suggest something quite different.
Psychologists call this confirmation bias, which means our tendency to interpret new information in ways that reinforce our existing beliefs.
The problem is that this bias doesn't disappear when we conduct research. If we're not careful, it becomes part of the research itself.
Imagine a company decides that AI should be the next big feature in its product. The strategic direction has already been set, and research is brought in to validate the decision.
Researchers begin interviewing users, asking questions like: "Would AI be useful?" Most people would probably say yes.
Then the research team draws the conclusion:
Customers want AI.
The team feels confident. The data appears convincing. The strategy seems validated. But that isn't what the research actually proved.
No one asked whether AI would challenge users' behavior. No one asked whether it would influence a purchasing decision or solve a problem they experience today.
Questions like these reveal something entirely different:
Would you pay more for this product if it included AI?
Would you choose us out of the others because we have AI?
What was the last moment you wish for AI while you are using this product?
The answers to these questions are telling us different information. The first one certainly tells us that people would buy this idea. The second one tells us whether the feature creates enough value to encourage decision making. The last tells us whether we're solving an existing problem or creating a new one.
When we confuse these answers, we don't just misinterpret the data, but we fool ourselves into believing we've validated the product. In reality, we've only confirmed that people are open to an attractive idea, not that they'll pay for it, switch because of it, or continue using it once the novelty wears off.
We can easily turn what we hear into what we want to hear.
This is one of the biggest dangers in research that I have observed. We interpreting the right data in a way that confirms what we already wanted to believe.
The Reason: Organizational Incentives
Sometimes, even though people know the evidence is insufficient, they have good reasons not to say so, given their roles:
Researchers don't want to contradict senior leadership.
Agencies don't want to tell clients that their favorite idea is unlikely to succeed.
Product managers have already promised a roadmap.
Executives have announced a strategy to investors.
Designers don't want to be seen as "negative" after months of work.
Even students don’t want to fail the course.
Everyone has an incentive to keep the project moving.
As a result, research often shifts from testing assumptions to supporting them. Contradictory findings are softened. Uncertainty is presented as confidence because confidence is easier to sell than doubt.
The irony is that everyone is trying to reduce risk, but not always about the project, but relationships, budgets, or careers. And then the organization ends up taking on a much greater risk: investing in a product built on assumptions that were never seriously challenged.
The Cost Of Trial and Error
Many startups proudly say:
We'll let the market decide.
If your prototype costs $20,000 to build and you can release three versions within six months, real customer behavior is probably more valuable than months of additional interviews.
A friend of mine used to work in the sex toy industry. Interestingly, most of the product developers are men, designing products primarily for women. Conducting meaningful user research isn't easy since it's a highly personal topic, and what people say doesn't always reflect what they actually want or use.
Instead of relying heavily on pre-launch research, the company often releases products, observes how the market responds, and learns from real customer behavior. This works because the cost of failure is relatively low. If a product doesn't succeed, they can iterate and improve the next one.
Not every industry has that luxury. When developing a humanoid robot, a medical device, or a new vehicle platform, every mistake can cost years of development and millions in investment. In those cases, research has a much greater responsibility: not to justify decisions, but to challenge them before it's too late.
Here I want to propose my point of view:
The need for research is proportional to the cost of being wrong.
Every wrong assumption becomes extraordinarily expensive. When the cost of failure is enormous, you need research that genuinely challenges your assumptions before you commit.
So what can we do?
Find the Cheapest Way to Learn
Research is not free. Neither is failure.
The question is:
Can we afford to be wrong?
If the answer is yes, don't over-invest in trying to predict everything. Build, launch, observe, and iterate. Sometimes the market is the fastest and most honest source of feedback.
If the answer is no, then invest much more effort into reducing uncertainty before making the commitment. Challenge assumptions, test hypotheses, and make sure you're solving the right problem before investing years of development and millions of dollars.
One of the best ways, is to find the cheapest way to learn.
Sometimes the cheapest classroom is the market.
Sometimes it's a prototype.
Sometimes it's an interview.
Sometimes it's a simulation.
Finally, Ask One Difficult Question:
What evidence would make us change our minds?