AI gives you answers. But who makes the decisions in your team?

AI produces information and answers, not decisions. A decision means choosing one option, giving up the rest, and owning the consequences, which requires a clear human owner. That is why a team's decision bottleneck is not fixed by a new tool, but by clarity about who decides.

Last night we were at an event about AI and the future of decision making, here in Cluj. At some point, someone in the room asked the question that had been hanging in the air for an hour: "If AI gives me the answer in three seconds, what is left for me to decide?"

It is a fear we hear more and more often, in training rooms, but especially in management meetings. And it is not a foolish fear. For the first time, every person on the team has instant access to an answer that sounds intelligent. And yet something doesn't add up: companies have more information than ever, but they are not making visibly better decisions. Teams debate just as much, get stuck just as often, postpone just as hard.

Because an answer is not a decision. And team decision making has never been, at its core, an information problem. Let's look at what it actually is, and what remains your job, with or without AI.

TL;DR

• An answer is not a decision. Information is what we know, the answer is what seems right, the decision is what we choose to do knowing we might be wrong. AI is spectacular at the first two and leaves you alone at the third.
• AI adoption does not produce results on its own: 88% of organizations already use AI in at least one business function, up from 78% a year earlier, yet only 39% report a real impact on company-level financial results (McKinsey, 2025).
• The bottleneck is not information, it is the owner of the decision. Most decisions drag because nobody knows for sure who has the right to say "done, this is what we do."
• Who is allowed to disagree is the second thing AI cannot fix. If a team is already silent in meetings, an answer from a tool becomes the perfect excuse for nobody to think, and the silence now looks objective.
• The decision map: 60 minutes, no new technology. Write down your 5 to 7 most important recurring decisions and, for each, who makes it, who is consulted before, and who finds out after.
• Commitment first, automation second. AI can document and track what people committed to, but only after they have committed. Never the other way around.

What does a good team decision actually mean?

A decision is the moment a team gives up every option except one and owns the consequences. Data, analyses and AI-generated answers are raw material. A good decision has a clear owner, was argued against before it was taken, and ends with who, what, and by when.

Simply put, there is a huge difference between three things we confuse daily:

• Information is what we know.
• The answer is what seems right.
• The decision is what we choose to do, knowing we might be wrong.

AI is spectacular at the first two. At the third, it leaves you alone, because a decision requires something a statistical model does not have: ownership. Someone has to say "we go with this option" and put their name next to it.

For example, a chatbot can list every argument for and against a team restructuring in ten seconds. But it will not stand in front of your people on Monday morning, explaining what changes. You will.

Why doesn't more technology mean better decisions?

Because the tool does not decide, people decide. AI adoption has become close to universal, but reported financial impact remains a minority outcome. The difference is not made by who has the best AI, but by who can turn it into decisions the team actually executes.

The numbers confirm the paradox. McKinsey's latest global survey, "The State of AI", shows that 88% of organizations already use AI in at least one business function, up from 78% just a year earlier. And yet only about a third have started scaling AI beyond the pilot stage, and only 39% report a real impact on company-level financial results. The rest stay stuck in experiments that look good in presentations but change nothing in practice.

Notice what that number actually says: the tool is already everywhere. The results are not.

We have seen the same pattern for years, long before AI. A company buys a new instrument, a management tool, a methodology, an OKR (Objectives and Key Results) system, and expects the instrument to solve the problem. But the instrument does not decide. People decide. And if the team did not know how to make good decisions before, AI will simply hand it more elaborate arguments for postponing. This is exactly the pattern we set out at length in why training programs fail: buying content or tools where the architecture is missing.

But it doesn't have to be this way.

Decision architecture: three things AI cannot do for you

Who decides, who is allowed to disagree, and how discussion becomes commitment. These are the three pieces most often missing from a stuck team, and none of them is an analysis problem. All three are organizational design choices, made by people.

When we work with management teams, we do not start from "what tool do you need?". We start from "how do you make decisions now?". And almost every time we discover that the problem is not the analysis. It is the architecture.

1. Who decides: decision rights

Most blockages do not come from a lack of information but from a lack of clarity: nobody knows for sure who has the right to say "done, this is what we do." That is where you get meetings that end with "let's talk again," decisions made three times over, and the quiet resentment of the person who thought it was their call.

A decision needs a clear owner. Not perfect consensus, but one person who decides and a team that knows who that is. Decision rights sounds corporate, but the question behind it is as simple as it gets: who has the final say?

2. Who is allowed to disagree: psychological safety

Google's researchers studied 180 teams in Project Aristotle and found something that surprised everyone: the strongest predictor of performance was not how smart the people were, but how safe they felt to say what they think. Psychological safety, the sense that you can raise a problem or an objection without being punished, mattered more than individual talent.

AI does not solve this. On the contrary: if your team is already silent in meetings, an "objective" answer from a tool becomes the perfect excuse for nobody to think anymore. "That's what the AI says." A good decision needs contradiction, someone who dares to say "I'm not convinced." That is a human, not a model.

3. How you move from discussion to commitment and ownership

The third piece is the most ignored: a decision is not truly made until the team knows who, what, and by when. Many teams confuse agreement in the meeting with commitment. Everyone leaves nodding and nothing happens.

This is where AI genuinely can help: documenting, tracking, reminding people what they committed to. But only after people have committed. First the commitment, then the automation, never the other way around.

What actually happens in a real team

We were working with the management team of a manufacturing company, good, professional people who wanted to introduce new analytics tools to decide faster. Their frustration: important decisions dragged on for weeks.

When we dug into the details, we discovered the tool was not the problem. They already had excellent reports, more than they used. The problem was that, for almost every important decision, three people each believed it was someone else's call. Information was not missing. The owner of the decision was.

We brought no new software. We did a single thing: together, we mapped the ten most important recurring decisions and wrote down, for each one, who makes it, who is consulted before, and who is merely informed after. A one-page map.

The lesson was not about technology. It was that they had been buying new tools to compensate for a decision architecture that simply did not exist.

Where to start: map your decisions before you buy more technology

If you are a manager or business owner and you feel your team moves slowly, do not start from "which AI would help us?". Start from a more uncomfortable question: do we make good decisions, with or without technology?

Here is the exercise you can run this week, in 60 minutes, with no new tool:

1. Write down the 5 to 7 most important recurring decisions in your team (from "we approve the campaign budget" to "we hire for position X").
2. For each one, answer three questions: Who makes it? Who must be consulted before? Who finds out after?
3. Wherever the team cannot agree even on the first question, you have found exactly where your decisions get stuck.

That is decision architecture. Only once it is clear does AI become a real accelerator, instead of an elegant excuse for postponing.

Because that is the truth we took home from the event last night: AI will make good teams even faster. And it will make confused teams even more confused, just more elegantly.

At Azimut, this is what we build with management teams: not just a training, but the architecture through which a team decides and actually follows through. We usually start with a Capability Diagnostic to see where decisions really get stuck. And we wrote separately about how you build the behaviors that sustain such decisions, in An effective team building shows up on Monday. If you want to see what this would look like for your team, talk to us.

Sources: Google re:Work, Project Aristotle, Understanding team effectiveness (180 teams; psychological safety as the strongest predictor); McKinsey & Company, "The State of AI", global survey, November 2025 (88% adoption in at least one function, up from 78%; about one third scaling; 39% reporting impact on financial results).

FAQ

Does AI replace team decision making?

No. AI produces information and answers, but a decision means choosing one option and owning its consequences. That requires a clear human owner, something a model cannot provide.

What is psychological safety and why does it matter for decisions?

Psychological safety is the sense that you can raise an objection or a problem without being punished. In Google's Project Aristotle study, across 180 teams, it was the strongest predictor of team performance, because good decisions need people who dare to disagree.

What are decision rights?

Decision rights define, for each type of decision, who makes it, who is consulted, and who is merely informed. Without them, teams make the same decision several times, or never make it at all.

How do you improve decision making without new technology?

Map your 5 to 7 most important recurring decisions and write down, for each, who decides, who is consulted, and who is informed. Wherever there is no agreement on the owner, you have found the blockage.

Sources

• McKinsey & Company (2025). The State of AI: Global Survey. Fielded June–July 2025, published November 2025; 1,993 respondents across 105 nations.
• Google re:Work. Project Aristotle — Understand team effectiveness. 180 teams; the five dynamics, with psychological safety ranked first.
• Azimut internal program data: 600+ programs delivered, 20,000+ participants, 9.27/10 average trainer score (since 2008).

By Szilárd Kacsó