When the System Believes First: AI Anchoring Bias

The most consequential effect of AI on executive decisions is not a wrong answer. It is an early one. When an AI recommendation reaches the leadership team before the team has framed the question for itself, the team’s job changes without anyone noticing: it stops deciding what problem it is solving and starts critiquing a solution it was handed. That shift is anchoring bias in AI-augmented decision-making, and it favors the answer almost every time, because most people are far better at framing a problem than at taking apart a confident answer to it.

This is not a story about bad models. The recommendation can be sound, well sourced and clearly argued. The issue is sequence. Whoever speaks first in a decision sets the terms everyone else argues within, and AI now speaks first more often than anyone in the room. What follows describes how that happens in a real capital decision, why critique is a weaker discipline than framing, and how CEOs and CFOs can put the order of thinking back where it belongs.

What is anchoring bias in AI-augmented decisions?

Anchoring is the long-documented tendency to give disproportionate weight to the first piece of information on the table. A first offer in a negotiation, a first estimate in a forecast, a first number on a slide: each pulls later judgments toward it, even when people know the anchor is arbitrary.

AI adds two things that make the anchor stronger. The first is timing. A recommendation can be generated overnight and waiting in the inbox before the meeting where the question was supposed to be discussed. The second is finish. The anchor no longer looks like a rough first guess. It arrives with a rationale, sensitivities, risks and a recommended option, which makes it feel like the end of the analysis rather than the start of it.

Together, those two features change the question a team is answering. Instead of “what decision are we facing?” the team asks “is there anything wrong with this?” Those are different questions, and the second is much easier to answer with “no.”

Why are leaders worse at critique than at framing?

Framing is generative. It asks what is happening, what it means for the business, and which problem is actually worth solving. A good framing conversation surfaces options nobody had written down, challenges the premise of the request and sometimes concludes that the question on the agenda is the wrong one.

Critique is reactive. It starts from a structure someone else built and looks for faults inside it. The options are already listed. The criteria are already chosen. The risks are already named, which quietly implies that the unnamed risks are smaller. A critic can test the numbers in a model; it is much harder to notice that the model answers a question the business should not be asking.

There is also a social cost. Challenging a framed answer means arguing against something that looks finished, in front of colleagues who are ready to move. The person who says “I’m not sure this is the right question” sounds less prepared than the person who suggests adjusting an assumption. Over time, teams learn which kind of contribution is rewarded. As one character in a recent book observes, executives “can abdicate not by clicking approve on a single case, but by accepting the first clean frame.”

What does an early answer look like in a capital allocation review?

Consider a composite case. A mid-sized manufacturer holds its quarterly capital allocation review at 10:00 on a Tuesday. At 8:00, an AI-assisted analysis lands in every executive’s inbox. It compares three investment proposals, ranks them on projected return and payback, flags the main risks and recommends funding the second proposal, a capacity expansion for the company’s fastest-growing product line.

By 10:00, everyone has read it. The meeting opens with the recommendation on screen. The discussion is serious and competent: the CFO questions a working-capital assumption, the head of operations notes a supplier constraint, and the ranking holds. The expansion is approved, subject to a revised timeline.

What did not happen was the conversation the meeting was scheduled for. Nobody asked whether the company should be adding capacity to its current product mix at all, given a shift the commercial team had been describing for two quarters. Nobody asked whether the right comparison was among three proposals or between all three and doing something different. Those questions would have come up in a framing conversation. They had no natural place in a critique of a ranked list. The system had believed first, and the team spent its best hour checking the system’s work.

How does the Progression of Meaningful Response protect decisions from the first answer?

The Progression of Meaningful Response describes the order of thinking that good decisions follow: Sense-Making (“What is happening?”), Meaning-Making (“What does this mean?”), Framing (“What do we need to solve?”) and only then Solving (“How do we solve this?”). Each stage depends on the one before it. A solution built on an unexamined frame can be internally excellent and still be the answer to the wrong problem.

An early AI recommendation skips straight to Solving. It does not do this out of malice; it does it because it was asked for an answer, and answers are what it produces. The risk is that the leadership team follows it there, treating the first three stages as already done because the output looks as though someone did them.

Used deliberately, the same discipline gives leaders a practical test. Before a team reviews any AI-generated recommendation, it should be able to state in its own words what is happening, why it matters, and what problem it has decided to solve. If it cannot, it is not ready to evaluate an answer, however good that answer is.

How can CEOs and CFOs keep AI from anchoring major decisions?

The remedy is not to keep AI out of strategic decisions. AI analysis can widen the options, stress-test assumptions and surface patterns no team would find by hand. The remedy is to control when the answer enters.

Frame before you generate. For consequential decisions, hold a short framing session before any AI recommendation is produced or circulated. Write down the problem the team has agreed to solve and the criteria that matter. Then ask the system to work inside, and against, that frame.

Separate analysis from recommendation. Ask AI tools for the evidence, the scenarios and the sensitivities first. Release the recommended option later, or not at all, so the team builds its own view before seeing the machine’s.

Ask for the competing frame. Require every AI-assisted analysis of a major decision to include at least one alternative framing of the problem, not just alternative options within the same frame.

Assign a framing owner. Give one executive explicit responsibility for asking whether the question on the agenda is the right one. Rotating the role keeps it from becoming one person’s reputation.

Record what the team believed before it saw the answer. A one-paragraph note of the team’s initial view, written before the recommendation is opened, makes anchoring visible in hindsight and helps the team learn whether it is deciding or deferring.

None of these practices slow a good decision much. They protect the part of leadership that cannot be delegated: deciding what the organization is actually trying to solve.

Frequently asked questions

Does AI make anchoring bias worse in business decisions? It can. AI recommendations tend to arrive early and fully finished, with rationale and risks already laid out. That combination makes them stronger anchors than a rough first estimate. Teams then shift from framing the problem to critiquing the answer, and critique rarely overturns a confident, well-structured recommendation, even when its frame is wrong.

Should executives see AI recommendations before strategy meetings? For routine decisions, yes; pre-reading saves time. For consequential decisions, release the evidence first and hold back the recommended option until the team has agreed what problem it is solving. That keeps the analysis useful while preventing the machine’s answer from setting the terms of the discussion.

What is the difference between framing and critiquing a decision? Framing defines the problem: what is happening, what it means and what needs to be solved. Critiquing evaluates a proposed solution inside a frame someone else chose. Framing can conclude that the question itself is wrong. Critique almost never does, which is why the order in which a team does them matters.

How do you reduce anchoring bias in AI-assisted capital allocation? Agree on the investment question and decision criteria before generating any AI ranking. Ask for scenarios and sensitivities before a recommendation, require at least one alternative framing, and record the team’s initial view in writing. Those steps keep AI analysis in service of the decision rather than in charge of it.

---

Leadership teams that want to examine how AI is reshaping the way their organization frames decisions can start a conversation with Anker Bioss about governing judgment and accountability in AI-shaped work. Related reading: The Progression of Meaningful Response: Sense-Making Before You Solve and The Point of Reliance.

Jose J. Ruiz is CEO and Managing Partner of Alder Koten and Chairman of Anker Bioss. He develops these ideas at book length in AI in the Org Chart: A Leadership Guide to Implementing AI Without Losing Human Judgment, Accountability, and Trust, published by Elavant Press and available on Amazon.