Founders have a habit of looking more at dashboards than the spreadsheets or raw numbers. Someone else built the model, chose the cohort, decided what counted as “active,” and picked the date range that made the chart more readable. The founder’s job, in theory, is to decide what to do next. And, it makes sense for founders to outsource and delegate, especially if they’re a non-technical founder.
But in practice, a lot rests on trusting that whoever built the analysis asked the right question in the first place. After all, for what the founder lacks in technical analyst skills, the analyst lacks in business intelligence and context relative to the founder.
Most founders have no reliable way to check. Not because they can’t read a chart, but reading a chart and interrogating the method behind it are very different things. Only one of them gets taught.
The failure is deceiving because it’s not that the answer (analysis and dashboard) is bad, but that it’s a good answer to the wrong question.
The sums, not the story
The clearest example of what gets missed is in a 1973 admissions dataset from Berkeley. In aggregate, the university had acceptance rates of 44% from male applicants and only 35% from female applicants that year. It was a big, concerning gap that looked like clear evidence of bias. But when broken down, the picture began to reverse. Women applied in far larger numbers to hyper-competitive departments that had overall acceptance.
Pool the departments together and the aggregate number found a pattern that doesn’t really exist in any single department. In a business context, the dashboard and top-line figure in isolation didn’t tell the whole story. When the founder is presented with only half the story, they can draw the wrong conclusions.
The data that never made it back
The same failure runs the other way too. Not a total that hides a pattern, but a dataset that’s missing the cases that actually mattered the most. In 1943, the US Army wanted to know where to add armor to its bombers, and the obvious approach was to study the bullet holes on planes that returned from missions so it could reinforce the areas that were hit most often.
The statistician Abraham Wald, working with Columbia University, pointed out the flaw in this approach: the returning planes were, by definition, the ones that survived their damage.
So the areas with the fewest bullet holes on returning aircraft weren’t actually the safest places to get hit, they were the most devastating places, because a hit meant the plane never made it back to be counted at all.
So the armor went on the gaps in the data rather than the marks on it. Wald wasn’t working from a bigger dataset than the officers that were already studying the same bombers, he was asking a different question of the identical numbers.
A founder analysing churn might focus completely on exit surveys from departing users, missing the fact that the most critical failure signals are found in the lost prospects who abandoned the product before reaching the onboarding phase to leave any feedback at all. Once you pose the right question, the absence of data is telling in and of itself, but it not an intuition that many have naturally (due to human biases), so it must be taught.
Where the skill actually gets taught
Interrogating a data set and asking what’s missing, what’s been pooled together that shouldn’t be, what assumption the model is making, does not come naturally to humans, but it’s a teachable skill.
It shows up as a deliberate part of some business degrees rather than something students pick up accidentally. Schiller International University’s degree in international business is an example of building that kind of analytical questioning into its coursework along with the more obvious international business skills.
For students or founders who want that training, an american university in Germany means not settling for a translated course, but having it natively created in English, while also benefiting from studying in an internationalised environment.
A confidence problem, not a skills one
This isn’t so much a founder-specific failing, nor does it fade with seniority. A KPMG global survey of executives (ten countries) found:
- 70% believed using data and analytics exposed their organisation to reputational risk. This mistrust tracked all the way through the process
- 10% of respondents said their organisation excelled across every stage of managing data and analytics
- 16% were confident in the accuracy of the models they were producing
- 18% had governance frameworks they’d confidently describe as adequate across the board
It became clear that data sourcing had the most trust, but confidence soon fell away at every stage downstream. Founders of large organisations report the same gap a first-time founder has – it’s not a lack of dashboards, but a lack of grounds to trust their foundations.
The founders who ask the better questions usually aren’t smarter, but simply just learned, at some point, to stop reading the answer before they understand the method that produced it.
Picture Tumisu / Pixabay
Author Dino Bozzi
Statements of the author and the interviewee do not necessarily represent the editors and the publisher opinion again.
