Curiously Dangerous: The Questions Leaders Should Ask Before Data and AI Drive the Decision
1965 words/8.25-minute read
Most leaders approach data and AI cautiously. I'm asking you to approach them curiously dangerous: curious enough to ask better questions, and dangerous enough to act on the answers.
Think about the last time you checked the weather or the traffic before leaving the house. You weren't looking at data for the fun of it. You had a decision to make: what to wear, when to leave, which route to take. The data informed the decision. The data wasn't the decision.
Somewhere between that morning traffic check and the boardroom, we lose that clarity. Leaders today have more dashboards, more reports and more AI tools than ever. Teams can produce a polished analysis in an afternoon that used to take a month. And yet I still talk with executives who walk out of meetings less certain than when they walked in. The slides looked great. The numbers sounded confident. But nobody could quite say what the organization should do next, or how sure they really were.
That isn't a technology gap. It's a curiosity gap. Data alone does not create insight. Insight comes from asking better questions, and that's a leadership skill, not a data science skill.
AI raises the stakes in both directions. Every major technology shift creates two groups: those chasing tools and those chasing value. AI can generate more options, more analysis and more confident-sounding numbers than any team could check by hand. The leaders who get value from it aren't the ones with the most tools. They're the ones asking the best questions.
What it means to be curiously dangerous
Curious means giving yourself permission not to know all the answers, and being curious enough to go find some. It means challenging what you think you know about your customers, your operations and your data, and asking the question everyone else skipped.
Dangerous means acting on what you learn. Experimenting. Making a recommendation and backing it. Playing it completely safe with data and AI is its own risk: decisions stay slow, technology investments sit underused and good ideas never get tested.
I live in Hawaii, and I love to surf. One of the biggest lessons the ocean teaches is that you don't paddle for every wave. You sit, you watch the set come in, you see who makes it and who wipes out, and you choose. Often the hardest decision isn't how to ride the wave. It's which beach, which board and which wave is worth riding today.
Data and AI work the same way. There's no shortage of opportunities. The leadership question is which ones are worth riding, and which ones to let pass. You don't need to become a data scientist to make that call. You need a handful of questions you ask every time, and a team that knows you'll ask them. Here they are, organized by the three moments that matter most.
Get curious before the analysis starts
The most expensive data mistakes happen before anyone opens a spreadsheet. A fuzzy question produces a fuzzy answer, no matter how sophisticated the tools are.
1. What decision are we actually trying to make, and why?
Know your why. Not "let's look at the customer churn data," but "should we invest in a retention program next quarter, and at what level?" When the decision is named up front, teams know what to look for and what to ignore. When it isn't, you get a beautiful report that answers a question nobody asked.
2. What does a good outcome look like, and how will we know?
Agree on the desired outcome and the measure of success BEFORE the findings arrive. Otherwise it's far too easy to decide what success means after you've seen which numbers look good.
3. Are we asking about yesterday, or about tomorrow?
Most reporting is descriptive or diagnostic: what happened, and why. Some is predictive: based on the past, what's likely next? All useful. But I relied on the past to plan my 2020 travel, and we all know how that went. In a volatile environment, the past only takes you so far.
The questions that pay for themselves are prescriptive: What would help the business tomorrow? What can we MAKE happen? Not "what have customers bought?" but "if we offer a coupon next month, can we move the product that isn't selling, and what would that be worth?" That one shift turns a reporting exercise into a business case.
Stay curious when the findings arrive
This is the moment where one well-placed question saves the most money. The recommendation is on the screen, the room wants to move, and the numbers look convincing.
4. How do we know?
It's the simplest question in the room and the most powerful. A strong team should be able to answer it in five parts:
• Source: Where did this come from, and is it current?
• Sample: How much data, over what time, and who's missing from it?
• Assumptions: What are we taking for granted?
• Confidence: How sure are we, and have we said so plainly?
• AI's role: What did AI do, and what did a person verify?
That last one is new, and it matters. AI tools can produce a statistic, a benchmark or a projection that looks authoritative and doesn't exist anywhere in your data. AI-generated numbers aren't data. They're claims to verify. AI can generate. People must evaluate. Make it normal, and safe, for your teams to say exactly how AI was used and what they verified.
5. Is this a cause, or just something that happens at the same time?
Correlation means related. Causation means caused. When I teach this, I use a simple example: employees with the most help desk tickets also logged the fewest training hours. Ask a room of analysts whether that's correlation or causation and you'll hear their favorite answer: it depends. Now imagine an AI summary that confidently states, "More training will cut tickets by 30%." New hires, for example, have both less training and more questions. Investing in the wrong cause is one of the most common, and most expensive, ways good data leads to a bad decision.
6. What would change our minds?
People get creative with data. Anyone ever seen an engagement survey that came back low, so the rating scale quietly changed from one-to-ten to one-to-three the next year? Ask your team to argue against their own recommendation. What would a skeptic say? What data would prove us wrong? Watch for a percentage with no "out of how many," a timeframe chosen because it tells the best story, an average hiding a wide spread, or a chart whose axis makes a small change look dramatic. A team that has stress-tested its own conclusion is a team you can back with confidence.
Get dangerous before you commit resources
A sound finding is only half the job. The value shows up when the organization can act on it.
7. What exactly are we recommending, and what will it take?
Ask for the recommendation as an action statement: "As our goal is to ___, we should ___." For example: As our goal is to connect with a younger customer base, we should invest in mobile solutions. Notice it doesn't say "build an app" or "hire a vendor" yet. It gives you a clear direction to say yes or no to. It also gives you instant scope clarity. When someone suggests a website redesign six weeks later, everyone can ask: does that help us invest in mobile solutions?
Then put it in the context of the business. What if we've never built anything mobile before? Is that a gap or an opportunity? Do we outsource the first time, or build in-house because this is becoming a competitive advantage? Many good decisions stall not because the data was wrong, but because nobody planned for the people, process and technology the decision required.
8. Who owns it, and when will we check whether it worked?
Three of my favorite questions:
Who owns the data?
Who makes the decision?
Who gets the bill?
Answer them BEFORE you charge ahead.
Then set the date to re-measure before you commit, not after. Here's a question I love asking leadership teams: how often do you go back and compare the business case to the actuals? A project pitched at $15 million in profit that delivers $100,000 is a lesson the organization should learn from before the next pitch. And ask one more: will this decision still look smart in five years, given everything we can't predict? That's not agile. That's agility.
Which waves to ride, and which to let pass
The line isn't "AI or no AI." It's about matching your speed to the cost of being wrong.
Ride the wave: move fast | Let it pass, or check first |
Using AI to brainstorm questions, options and scenarios | Any number AI produces that will drive a budget or a target |
Exploring alternative and substitute data sources to spot patterns worth investigating | Claims that one thing causes another |
Drafting summaries, first-pass analyses and stakeholder messages | Decisions that are expensive or hard to reverse |
Asking AI to argue against a recommendation | Decisions that affect customers, employees or compliance |
Small, reversible experiments with a clear checkpoint | Putting personal, customer or confidential data into any tool that isn't approved |
Challenging assumptions the organization has never questioned | Projections presented as if they were results |
Being curious pays off in that left column. At one point, my company wasn't tracking the customer behavior we needed. The help desk, though, had website traffic, how customers reached our app, how long they stayed and what they purchased. They didn't think of it as customer data. It was gold. That's the left column at work. The right column is where your questions protect your credibility, your customers and your technology investment.
How leaders get curiously dangerous this quarter
You don't need a new platform to make better decisions with data. You need a few clear expectations, modeled from the top.
1. Make "How do we know?" a standard question. Ask it in every decision meeting until your teams start answering it before you ask.
2. Ask for the decision before the dashboard. Every analysis request names the decision it supports and what a good outcome looks like.
3. Set clear guardrails for AI and data.Make sure people know which tools are approved and what data can go into them. Uncertainty here slows adoption more than any policy does.
4. Measure the value added, not the work removed.Say a new AI note-taker saves each of five people 20% of their week. The tempting math says that's one position you no longer need. The curious question is better: what is each of those people now doing with that time? AI doesn't create innovation. It frees up space for it.
5. Invest in the people, not just the tools.Your technology investment pays off when people know how to frame questions, validate outputs and connect the work to value.
AI can accelerate the work. It cannot define value for us. That still takes leaders who ask better questions, and teams who know those questions are coming. So be dangerous because you ask better questions. Dangerous because you recognize the opportunities others miss. Dangerous because you know which waves to ride, and which to let pass.
I'll leave you with one of the best questions I know, and one worth asking at the end of every decision meeting: What questions do we have that we haven't asked yet?
Arrive curious. Leave dangerous.
Jamie Champagne helps organizations maximize the value of their technology by bringing people, process and AI together through keynotes, custom training and consulting. Want your leaders and teams asking better questions of their data? Connect at champagnecollaborations.com.
View our article sharing and reuse policy here
