Signal, not noise.

Decision quality

Information changes a decision only when it stands where the decision is made.

Lately, I have read a lot of claims about improving decisions. None of them say how you would know. They leave you with “better decisions” and “decision quality.”

Risk managers embrace objectives as though they were new, though COSO ERM and ISO 31000 have said so for years. Data managers promise trustworthy data for better decision quality. The AI market promises to automate the process and leave humans the final say. Decision science promises better judgment under uncertainty.

Different disciplines, same claim, barely any contact between them.

So I went back through my own cases. Some where information genuinely changed a decision, some where it didn’t. The difference wasn’t the quality of the data, nor a professional opinion.

It was where the information stood when the decision was made.

One that worked: a portfolio of regulatory programmes. Two numbers finally placed side by side. €1m to implement, against a regulatory exposure capped at €100k. Nobody had hidden that. Nobody had asked. That is a real choice: pay for good citizenship, or take the hit knowingly.

One that did not: an operational register (BPM or GRC) and a dashboard (BI platform), living in a different system from the process they describe. Using them means leaving the work, switching format, reloading context. So people decide on intuition, because it is faster and cheaper at the moment of deciding.

What I mean

Take a common objective: growing the customer base. You monitor the opportunities and the uncertainties to secure the target.

One uncertainty is legitimate customers, better known as know-your-customer: how many prospects you can legally onboard, and which customers you can no longer keep. The same applies to uncertainties like fraud or a sales pipeline backlog.

An obvious influence on the growth objective. Then it gets its own resources, mandate, report, indicators and committee. Experts go deeper, the material gets richer. Rightly so. But the distance to the objective it was meant to inform grows. So does the distraction.

No step in that sequence was a mistake. Each was justified when someone took it.

Context, legacy, pressure, interests and complexity make this hard. All legitimate reasons. None of them an excuse for the distance.

So, practically

Every decision already has an artifact. A business case. A target. A screen someone is looking at when they say yes.

Put the information there. Not in a new register, not in a new dashboard, not in a pack read once. In the thing already in front of the person deciding, every day. Start with the end.

And if you cannot find the artifact, that is the finding. The decision is not being made anywhere you can reach.

ReflectionWhich decision, made by whom, would have gone differently?

Why AI (risk) needs judgment, not just models

Three models, three answers, all statistically valid. The value is in the assumptions, not the number.

The hype cycle promised: “AI will solve decision-making. Algorithms will remove human bias.” The reality cycle: “My AI and their AI gave opposite answers.” Who’s right?

The problem

One AI analysis says “This decision will work.” Another says “It won’t.” Both are statistically valid. Both used different data. Both have different assumptions baked in.

A kitchen-table example

Jane wants to go to the Maldives in six months. Monthly income: €9k. Current savings: €15k. She feeds the exact same prompt to three different models:

• Model 1 (Claude Haiku): save €300 a month
• Model 2 (Copilot): save €1,200 a month
• Model 3 (Gemini): save €2,000 a month

All three are statistically valid. All three follow the framework perfectly. All three come with detailed breakdowns, assumptions and explanations.

So what do leaders (or you) do? They pick the result that matches their intuition. Which means they didn’t need the AI.

What’s actually happening

AI is democratising statistical analysis. That’s good. You don’t need a PhD to run sophisticated models anymore.

But democratisation has a cost. People now use complex tools without understanding them. They hide behind “the algorithm said so” instead of explaining their thinking. Result: decision quality gets worse, not better.

The real value isn’t the number

It’s understanding:

• What data did we use, and not use?
• What question did we actually ask?
• What assumptions are baked in, even if unstated?
• What could make this analysis wrong?
• Can I explain this to someone who disagrees?

That’s not AI work. That’s judgment work. Thinking. Experience.

Why this matters now

Every organisation is deploying AI and asking: “How much do we trust this?” The answer: use AI as one lens, but keep human judgment about the assumptions underneath. And make sure whoever makes the decision faces the consequences of it.

If you can’t explain the AI’s analysis to someone who disagrees, you don’t understand it. And if you don’t understand it, you’re betting.

Judgment isn’t the opposite of data. Judgment is what you do with data.

What actually works

When someone brings you an AI analysis, ask three questions:

1. “What assumptions did you make?” For Jane’s trip, it came down to assumptions about accommodation costs and the size of her emergency fund. Do they match Jane’s reality? Her tolerance? Her preferences?
2. “What would prove this wrong?” Model 1 (€300): “If flights cost more than €1,500, this breaks.” Model 3 (€2,000): “If Jane can cut discretionary spending by €1,500 without burning out, this works.” Now you know the sensitivity.
3. “Would you stake your reputation on this?” If the person presenting won’t bet their bonus on it, why should you bet the organisation’s money?

Why your decision-makers ignore the data

And what actually works. The gap between analysis and decision isn’t a data problem. It’s the conversation.

A C-level executive brings data to a strategy meeting: “Market demand is down 15%. Our growth assumption breaks.”

The CEO says: “I disagree. We’re staying the course.”

The data gets filed. The strategy doesn’t change. The assumption holds, until it doesn’t.

This happens constantly. And it’s not because executives aren’t smart. It’s because data alone doesn’t change minds. Data without context feels abstract. Data without an invitation to challenge feels like criticism.

Here’s what I’ve learned:

The gap between data analysis and decision-making isn’t a data problem. It’s a culture problem.

You can have perfect analysis. You can have sophisticated models. You can have statistically sound predictions, and a genAI storyline. None of it matters if:

• People can’t safely challenge the underlying assumption
• The decision-maker feels attacked by the data
• The data is presented as “the algorithm says” instead of “here’s what we found, what do you think?”
• There’s no way to test whether the assumption holds before committing any budget

What actually works

Strategic (objective) inversion. It’s not a new framework. It’s a different question structure. It also addresses what Nassim Taleb calls skin in the game: those who decide have a personal stake in the outcome. When compensation reflects accuracy rather than optimism, incentives shift towards honest analysis.

Instead of: “Here are the uncertainties to achieving your market growth.”

Ask: “What has to stay stable for this to work? Market demand? Customer spending? Our cost structure? Competitor pricing?” Then ask: “What evidence do we have that these will hold? What would break them?”

This moves the conversation from “you’re wrong” to “let’s test this together.”

The executive doesn’t say “Your growth assumption is broken.” They ask: “For 25% growth, we need market condition X. Do we have evidence it’s stable?”

Now it’s collaborative. Now it’s testable. Now the CEO can actually hear it.

Why now

Organisations are drowning in new decision governance: data, AI, cyber, ESG, climate. Each has its own framework. Each adds weight. Each produces stand-alone decisions. Nobody has time to master them all, while each affects their goals, directly or indirectly, without anyone seeing or feeling it.

What you actually need is a different way of thinking about uncertainty in decisions and goals. One that works across all decision governance.

Not more frameworks. Better judgment.

Your data is probably fine. Your analysis is probably sound. What’s broken is the conversation between decision-makers and the experts and analysts who inform them. And the accountability: if the leader’s forecast doesn’t hurt them when it’s wrong, why take it seriously?

Fix the conversation. Fix the incentives. Decisions improve.

ReflectionWhen was the last time data genuinely changed your mind about a decision?