Getting an analysis has never been easier. Ask an AI tool and within seconds you have projections, comparisons, scenarios. Access to information has stopped being the problem — and yet many organizations continue making the same decisions as always, with the same quality as always.

We see it frequently in consulting engagements: teams producing solid technical work — pricing models, promotional simulations, demand forecasts — but when we get to the room where the decision is actually made, that analysis is rarely on the table. Marketing looks at brand health, sales looks at the quarter, finance looks at margin. The analysis ends up as an exercise no one truly engages with, and decision-makers act under short-term pressure without having reviewed the recommendations.

The problem is not the quality of the analysis. It is that the system connecting that analysis to decisions was never designed.

Decision Analysis has a concept that has been taught in the world’s leading decision programs for decades: the value of information. The premise is straightforward: information only has value if it can change a decision. If the analysis doesn’t move anything, its value is literally zero — regardless of its sophistication, its cost, or the time it took to produce.

The mathematical foundation goes back a long way. Reverend Thomas Bayes, in his 1763 essay, established the logic by which new evidence should update what we know — and therefore what we should decide. Claude Shannon, two centuries later, formalized something complementary: the informational value of a message is proportional to its capacity to surprise. What we already know with certainty carries no relevant information.

The practical implication is direct: no one should pay anything for information about something they are already certain of. The more uncertain the starting point, the greater the potential value of the analysis. But that rule only holds if the decision-maker is genuinely open to the analysis changing their position. And that is where the real problem appears.

There is a well-documented human tendency to seek out information that confirms what we already believe — to pay more attention to data that supports our position and dismiss what contradicts it. Confirmation bias is not a failure of intelligence, but of disposition. And in organizational environments, where strong opinions are associated with leadership and doubt with indecision, that bias is amplified.

Add to this overconfidence — one of the most studied biases in decision psychology: the systematic tendency to underestimate the real uncertainty of a situation and overestimate the quality of one’s own judgment.

The result is that much of the analysis in organizations does not inform decisions — it decorates them. It is requested to validate, not to learn. And if that is what is happening, more analysis does not improve decisions. It justifies them.

There is a question that serves as a litmus test before commissioning any analysis: what would the result have to show me to change my mind? If the answer is “nothing,” the analysis has no information value. If the answer is genuine, the analysis can be worth a great deal.

The emergence of generative artificial intelligence changed the surface of the problem, but not its substance. Any executive can now have any analysis in thirty seconds. Analytical abundance is no longer a privilege of organizations with large data teams. But the bottleneck remains the same.

If you come to an AI tool with a well-defined decision — clarity on what you are choosing, between which alternatives, and on what criteria — you have the best analyst your organization has ever had. One that doesn’t get tired, has no agenda, and can explore scenarios in minutes.

If you arrive without knowing what decision you are trying to make — or worse, looking for confirmation — AI offers something equally powerful and far more dangerous: infinite distraction at unprecedented speed.

Technology did not change the question. It made it more urgent.

The shift is not technological — it is a shift in starting point.

Before opening any tool, before commissioning any analysis, it is worth pausing on three concrete questions:

What am I choosing? Not “I want to understand the market,” but “I am deciding whether to raise price in this segment, and by how much.”

What are the real alternatives? Defining the concrete options is part of framing the problem well — not a step that comes before it.

What outcomes matter when evaluating each alternative? Volume, margin, market share, brand reputation. The answer defines which analysis has value and which does not.

With that clarity, the question to AI changes entirely. Not “give me a pricing analysis.” But “if we raise price by 10% in the premium segment, what range of volume impact should we anticipate, and which variables are most sensitive?” The first question produces a report. The second produces an input for a decision.

Organizations that adopt this order — decision first, analysis second — develop something that cannot be bought with technology or replicated by installing software: the ability to ask better questions. And that capability, unlike a model or a dashboard, cannot be copied. It must be built.

In a world where anyone can run the analysis, what is scarce is knowing what you want to achieve — and which decisions can get you there. AI does not do that.

According to Gartner, by 2028, 25% of Chief Data and Analytics Officers’ vision statements will be “decision-centric” — surpassing “data-driven” slogans — as human decision-making behaviors are modeled to improve the value of data and analytics.^¹ The underlying shift has already begun.

Ernesto Weissmann
Socio de Tandem
ew@tandemsd.com

[1] Gartner, “More Than 100 Data and AI Predictions Through 2030,” June 2024.