Why More Data Does Not Produce Better Decisions: Kevin Hanegan on Bias, Questioning and Exponential Thinking
Most businesses now collect more data than they can use, and most have bought the tools to analyse it. Why, then, do the decisions coming out the other end so rarely improve? Kevin Hanegan, chief learning officer at a data analytics company, locates the answer downstream of the technology, in the part no dashboard covers: how a person reads a number and decides what to do about it.
The work he describes doing is training. Advising the company he works for and its customers on using data strategically, he then teaches their employees the skills that requires. Much of it comes down to interrogating a number before acting on it: where it came from, what it leaves out, and which existing belief it quietly confirms.
Three steps in his reasoning are worth following. Almost nobody in a business consumes raw data; they consume insights already shaped by someone else. Because the questioning that tests those insights has been trained out of most organisations, artificial intelligence arrives at an awkward moment. The same linear habit also caps how far a company can reimagine its own model, which is where the money sits.
Why Data Is Broader Than Numbers and Statistics
When Hanegan raises the subject in person, he watches half the room disengage, because people hear data and think mathematics and statistics. His own definition is much wider. Data, in his framing, is the lowest level of knowledge content, which puts the reviews someone reads before booking a holiday rental in the same category as a spreadsheet.
Volume is what turns that into a problem. Since so much information arrives at once, the mind economises, and one of its economies is to treat whatever it holds as the complete set. Hanegan describes the mechanism this way:
the brain gets overwhelmed and it takes shortcuts and I think that’s why misinformation is a problem now is we see surveys we see statements we see numbers and our brain thinks it’s a complete puzzle but it’s missing pieces
For owners and executives, the consequence concerns who gets left out of the data conversation. Anyone who declares themselves uninterested in numbers is still making data-shaped decisions all day, without a method. The choice is whether data capability is a specialist function or something every manager practises.
Most Decisions Rest on Someone Else’s Insight
Almost no one in a business works from raw material. Someone runs the analysis, forms an insight and circulates the conclusion. What concerns Hanegan is that the failure point sits upstream of that conclusion, in the inputs: an insight can be accurate and still answer a question nobody asked, or rest on a sample that never represented the population it describes.
He recalled hearing about early autonomous vehicles trained on a data set of Caucasians, which, as he remembers it, meant the system did not register a person of colour as human. A parallel case runs through lending, where automation trained on wealthy borrowers declines applicants with fewer assets and, on his account, snowballs the discrepancy it inherited. The suspicion belongs further back than most people put it, in the data feeding the insight.
Technology executives buying or building AI systems face a diligence question in that. Model performance is the easy thing to test; the provenance of training data is harder, and it is where the failures he describes originate. Investors assessing an AI-enabled business look at the same question from outside.
If questioning is the defence against a weak insight, the obvious puzzle is why so little of it happens.
Why Organisations Suppress the Questioning They Need
Hanegan’s explanation starts developmental and turns institutional. Children ask why relentlessly; adults stop, partly because a well-stocked memory says the brain already holds the answer, and partly because the environments they pass through punish the habit. A student who challenges a teacher is told not to talk back; an employee who challenges a manager is told, in his example, that long experience settles it.
The timing is what troubles him.
we suppress curiosity we suppress critical thinking and we’re at the same time asking everyone to use those more and more with AI and Automation and data there’s a disconnect there’s a gap
That reframes a training budget. Where managers have spent a career learning that challenge reads as insubordination, no tooling will generate the questioning AI-assisted decisions require; the remedy sits in how meetings are run, not in what software is bought.
Practical Ways to Catch Bias Before Deciding
The first obstacle is admitting the problem exists at all. Describing the workshops he runs, Hanegan estimates: “I will do a workshop and train a 100 people and on estimate there’s probably about 25 or 30 of them that will not believe we have bias”. Those, in his experience, are the hardest people to reach.
What breaks the pattern, on his account, is a jolt that forces the brain to register a disconnect. Short of that, the substitutes are procedural: canvassing opinions from people likely to disagree, and having someone play devil’s advocate. For weightier decisions he describes structured frameworks of about ten steps, requiring two alternative solutions to be weighed, so the mind has something to compare against.
Advisers and executives can read that as a specification for a decision process rather than as self-improvement. If a board paper carries a single recommendation, the structure guarantees the bias; requiring two viable options with their trade-offs changes what the room can see.
Using AI to Question a Decision, Not Make It
Two different uses of the same technology are worth keeping apart. One is generative, producing content and drafts. The other is decision support, and it is the one Hanegan has adopted: he feeds an insight and its assumptions into a general-purpose chatbot, then asks it to apply the Socratic method and interrogate him rather than answer him.
Asked for an example, he sketched a coffee business: the capital to raise, the site and its demographics, his expected path to profitability, and an instruction to prompt him with the questions that would expose where he is misinformed. No decision comes back, only the question he had not thought to ask: which assumption is doing the work, which variable is missing.
Founders should recognise the value, because it stands in for something early-stage companies rarely have: a board or peer group willing to attack a plan before capital goes in. He is clear that the technique supplements the human input he relies on most rather than replacing it.
Linear Thinking and the Limits of Business Models
The habit that lets a flawed insight pass unexamined also caps how ambitiously a company can reimagine itself. As Hanegan frames it, “technology is exponentially growing but our brains are not exponentially evolving at the same rate our brains are evolving linearly”. Handed a problem, people reach for cause and effect, and for increments.
Hotels spent decades adding a pool, a business centre, a restaurant. A different question, about whether a hotel could exist without owning a building, produced Airbnb. Taxi operators optimised routes and took cards; a transport company owning no cars produced Uber. The technology to work that way already exists, in his view, and the barrier is emotional: the exponential question sounds absurd until it works. Nor is it taught, he points out, in science or in MBA programmes.
For investors, that works as a screening lens. The question to ask of a plan is whether it improves an existing model or removes one of its structural assumptions, because the two carry very different capital requirements.
One qualification runs underneath all of it. Data does not settle an argument by itself, since two people can read the same numbers, both reason validly, and land differently according to their tolerance for risk. His own worry is not that machines will run amok, but that overloaded people are being handed more information, faster, without the skills to sort it.
What separates one company from another, on his reasoning, may have little to do with how much data it holds or which models it licenses. The divide falls between organisations that question an insight before acting and those that treat an analysis as an answer: the first can change its business model on evidence, the second can only optimise the one it has.
Kevin Hanegan set out his reasoning at greater length in Mastering Data Literacy: Overcoming Cognitive Bias and Misinformation on the GrownLearn podcast.
Listen on Spotify Apple Podcasts SoundCloud YouTube