99% of executives say data and AI are top priorities. Sebastian Wernicke, author of the book Data Inspired and a partner at Oxera, likes to sit with that number for a second. You can’t get 99 percent agreement on much of anything. Ask people whether they enjoy music, and you’ll get 98. So somewhere in that survey is an executive who doesn’t care for music but still wants to do more with data.
Then comes the second number. Ask those same executives whether data has genuinely transformed their business, whether it has changed their efficiency or their business model in a way they’d call meaningful, and the figure collapses to about 10 percent. Everyone else describes something smaller. A dashboard they’ve grown comfortable with. An optimization here and there. The occasional new idea.
The usual headline is that most data projects fail. Wernicke thinks that framing misses what’s actually happening.
“Almost everybody’s saying, I’m getting a modest amount of value out of data. But if you ask them, ‘Are you getting transformative change? Is it deeply affecting your efficiency and your business models? Only about 10 percent will say, ‘Yeah, that’s us.’ And everybody else will be saying, well, we’ve been making this a top priority for years, we’ve been pouring billions into this, and all we’re getting is this bit of incrementalism.”
His conclusion is the uncomfortable part. The projects didn’t fail. They worked. They just worked toward the wrong end. “If you’re not happy with your modest increases, you’ll probably just stay the same,” he told me. “And that’s what data made you do.”
I’ve watched this play out with leaders I respect. Capable people who put in real money and years of effort, and ended up with a slightly more efficient version of the company they already were. Wernicke’s book gave me language for why.
Most data strategies rest on a premise so reasonable that nobody questions it. He calls it the data deficit theory. The belief that the only thing standing between your people and better decisions is the right information at the right time. Give them the numbers, and they’ll act differently.
Decades of psychology say otherwise. He points to a Stanford study he traces back to the year he was born, one that took students with strong views on capital punishment and showed them a fabricated body of research, some of it supporting their position, some cutting against it. The point was to see whether conflicting evidence would pull them toward the middle. It pulled them apart instead. People who started in favor became more in favor. The opponents dug in harder. Faced with data that challenged them, they didn’t reconsider. They went looking for reasons to throw it out. As Wernicke describes the reaction, people land on “yeah, don’t trust the source, the data seems sketchy, I’m not gonna do that.”
If that’s how human beings handle inconvenient evidence, and it is, then pouring more data into an organization doesn’t produce better thinking. It produces more ammunition for whatever people already believe. The fix isn’t technical. It’s the slower work of building places where a leader can look at a number that contradicts three years of their own conviction and say out loud that they were wrong, without it costing them anything. That kind of admission doesn’t happen on its own. It happens when leadership makes it safe and treats it as a sign of strength.
Most Decisions Are Made Before the Meeting
There’s a diagram in nearly every business book about decision-making. Understand the situation. Lay out the options. Evaluate them. Pick the best one. Implement. Wernicke includes a version in his book and labels it the typical but wrong model.
It’s wrong because nobody decides this way. He walks through a Stanford experiment in which researchers wired up a monkey’s brain during a simple task and looked for the moment of decision. There wasn’t one. What they saw was noise, a jumble of competing signals that slowly resolved into a clear one, until a few seconds before the monkey acted, the researchers already knew what it would choose. No clean step from option to choice. Just chaos settling into direction.
Organizations work the same way, he argues, and most of us know it from experience. “Most meetings are decided before the meeting,” he said. “Most meetings are decided when you’re not even in the room, because you then have to get other people on board.”
This has direct consequences for anyone who cares about evidence-based decision-making. If you treat data as something you slot in at step three, the part where you calmly weigh options, you’ve already lost. The decision was forming long before the meeting, in hallway conversations and one-on-ones, and the slow business of lining people up. Data introduced late is data introduced after the fact. To change a decision, the information has to reach people early, while the signal is still taking shape, and in a form they can carry into the rooms you’re not in.
Why the Numbers Look Good While the Business Drifts
One of the cleaner distinctions in the conversation is between two kinds of metrics. Steering metrics are inputs. Things you control directly. In sales, that’s how many customers you call, how fast you respond, and how quickly you turn around a quote. Success metrics sit at the other end. Revenue. Deals closed. They look backward, and you can’t move them directly. You can only move the inputs that feed them.
The trouble starts when leaders steer by the success metric. Paying a salesperson on revenue feels obvious, almost too sensible to question. Wernicke’s warning is that it reliably goes wrong. This is Goodhart’s law, the old observation that when a measure becomes a target, it stops being a good measure. He uses GE as the cautionary case, a company that steered toward stable, predictable profits for years while those smooth numbers quietly hid the problems underneath.
“When you use success metrics for steering, in the long term, it will always drive people and organizations to behave badly. They start to manage the number, but they forget to manage the business.”
It rarely looks like cheating. People aren’t acting in bad faith. They’re responding to what gets measured and rewarded, and there are always more ways to move a number than to do the real work behind it. The numbers stay green while the business drifts somewhere else.
The Tool We Trust and the Tool We Should
The part of the conversation that’s stayed with me longest is about AI, and it runs against intuition.
Wernicke draws a hard line between machine learning and generative AI. Machine learning trains an algorithm toward a narrow, defined goal, and it’s grounded in statistics. Within its lane, it often outperforms us. A model trained on tens of thousands of skin images eventually spots a tumor more reliably than a trained physician. Generative AI is a different animal. It was trained, more or less, to produce output that pleases the person using it. Fluent, confident, agreeable, and considerably less reliable.
We distrust the system that has earned trust and trust the one that hasn’t. People resist machine learning even after they’re shown it makes better calls, the same instinct that makes us recoil from a self-driving car after a single accident, while we shrug off the steady toll of human drivers. Generative AI gets the opposite treatment. It sounds like a person, so we lean in. Wernicke put it more bluntly than I could: “AI tells you you’re brilliant, while machine learning tells you you’re wrong. Of course, you would trust the one that tells you you’re brilliant.”
For any leader putting these tools into real decisions, that’s the trap worth naming. The fluency of a generative tool is not evidence of its accuracy. The discomfort a rigorous model gives you might be the most honest thing in the room.
The thread running through all of this is a kind of courage that never shows up on a dashboard. The companies that get real value from data, in Wernicke’s telling, are the ones that use it to find better questions instead of more reassuring answers. They’re willing to be wrong. They make room for a leader to see the data, disagree with it, commit to a different call, and own the result either way.
That runs against nearly every operating pressure, and Wernicke is honest that he feels those pressures too. The hundred emails. The pull in twenty directions. The urge to skip the hard thinking and just ask for the recipe. His answer is that the recipe is the problem, a fancier way of saying you sometimes have to slow down to speed up. The companies worth copying didn’t get there by copying anyone.
What stuck with me most was something he said about case studies. The best ones he’s been part of, the ones that taught him the most, didn’t have a clean answer. The value wasn’t in learning what the leader did. It was in the argument that broke out afterward, when everyone in the room had to say how they would have handled it, and in the process found out what they actually believed.
That’s the test worth applying to your own data. Not whether it confirms what you already think. Whether it can start an argument good enough to change your mind.
Brandon Laws is the host of Transform Your Workplace and a team member at Xenium HR in Portland, Oregon.