XenonStack BlogProblems in Decision Making and How Leaders Fix Them

The most common problems in decision making are not a lack of data. They are human and organisational limits: people settle for the first option that is good enough, rely on mental shortcuts that bend judgement, give different answers to the same case, defer to the group, or stall because nobody owns the call. Time pressure and uncertainty make each of these worse.

Decision analysis helps by making the options, probabilities, trade-offs and owner of a decision explicit. For enterprise leaders the fix is less about one more dashboard and more about classifying decisions, naming who decides, testing plans before they fail and recording why each decision was made so it can be reviewed. That last step matters more now that AI agents recommend and take decisions too.

This guide focuses on the challenges. If you are choosing a method such as a decision matrix, Pareto chart or decision tree, see Decision Analysis Tools: Which Method Fits Which Decision. If your question is how forecasts turn into action, see predictive analytics in decision-making. For platforms that support the whole decision cycle, see the decision intelligence platforms buyer's guide.

What are the most common problems in decision making?

The table below lists the problems that show up most often in enterprise decisions, how to recognise each one and the first control that addresses it. Each is explained in the sections that follow.

Why do capable teams still make poor decisions?

Most decision failures come from predictable limits in how people and groups process information, not from a lack of intelligence or effort. Five of them account for most of the problems leaders describe.

Bounded rationality: settling for good enough

Herbert Simon received the 1978 Nobel memorial prize in economic sciences "for his pioneering research into the decision-making process within economic organizations". His idea of bounded rationality is that people have limited time, information and attention, so they look for a solution that is good enough rather than the best one. In a business this shows up as a team accepting the first supplier, site or design that clears the bar, without comparing it with alternatives that might be much better.

Cognitive biases: shortcuts that bend judgement

Amos Tversky and Daniel Kahneman's 1974 paper in Science, "Judgment under Uncertainty: Heuristics and Biases", described three shortcuts people use under uncertainty: representativeness, availability, and adjustment from an anchor. The paper's abstract notes that these heuristics "are highly economical and usually effective, but they lead to systematic and predictable errors." Typical enterprise examples are budgets anchored on last year's number, and risk estimates driven by whichever incident is most recent in memory.

Noise: different answers to the same case

Bias pushes judgements in one direction. Noise is the scatter between them. In a 2016 Harvard Business Review article, Daniel Kahneman and co-authors call the "chance variability of decisions" noise, and suggest a noise audit: have members of a professional unit evaluate a common set of cases, then measure how far their assessments differ. Pricing, underwriting, credit, hiring and claims decisions are where noise is most costly, because the same case should get the same answer.

Groupthink: agreement that hides doubt

Irving Janis developed groupthink into a testable model in his 1972 book Victims of Groupthink. The symptoms include an illusion of invulnerability, rationalising warnings away, self-censorship and an illusion of unanimity where silence is taken as agreement. His suggested remedies are practical for any leadership team: each member acts as a critical evaluator, the leader withholds their view when setting the task, independent groups work on the same problem, outside experts are invited in, and a devil's advocate role rotates from meeting to meeting.

Decision paralysis: analysis without a decision

Analysis paralysis is the opposite failure: so much analysis that no option is chosen in a reasonable time, often driven by fear of error or by too many options. In large organisations it tends to look like repeated requests for "one more data cut" on a decision that could be reversed cheaply if it turned out wrong.

How big is the decision-making problem in enterprises?

A McKinsey survey of 1,259 participants, in the field in February 2018 and published in April 2019 as "Decision making in the age of urgency", found that only 20 percent of respondents said their organisations excel at decision making. Respondents spent on average 37 percent of their time making decisions, and on average 61 percent said most of that time was used ineffectively. Only 48 percent agreed their organisations make decisions quickly, and 57 percent that they consistently make high-quality decisions.

McKinsey also sorted decisions into three types: big-bet decisions (rare and high-stakes, such as acquisitions), cross-cutting decisions (frequent, made of linked choices across groups, such as pricing) and delegated decisions (frequent and narrow, handled by one person or team). The type matters because each fails in a different way.

Which problems hit which type of decision?

Matching the decision type to the right process is one of the most effective fixes. Amazon's 2015 shareholder letter makes the same point: some decisions are "one-way doors" that "must be made methodically, carefully, slowly", while reversible "Type 2 decisions can and should be made quickly by high judgment individuals or small groups." The letter warns that larger organisations tend to use the heavy Type 1 process for most decisions, which leads to "slowness, unthoughtful risk aversion, failure to experiment sufficiently".

How does decision analysis handle uncertainty and risk?

Decision analysis is a systematic, quantitative and visual way to compare options. It distinguishes three situations. Under certainty, the outcome of each option is known, so the task is to pick the best payoff; this is rare in business. Under risk, each option has several possible outcomes and the decision-maker can estimate their probabilities, so the expected value of each option can be compared. Under uncertainty, the probabilities themselves are unknown, so scenarios, ranges and reversible first steps matter more than a single expected value.

The working steps are the same in each case: define the decision and the objective, list real alternatives, gather the information that would change the choice, build a simple model (often a decision tree or weighted matrix), estimate the expected value or range of each option, and decide. The decision analysis tools guide compares the methods for each step.

Two classic examples show the problems in practice. A real estate firm deciding whether to build a shopping centre may weigh traffic, demographics, local competition and the success of similar centres; bounded rationality and anchoring on one favourable data point are the main risks. A company holding a patent for a product with a short sales window may compare selling the patent with building the product itself; a decision tree with branches for price, cost and timing exposes the trade-offs that a gut call would hide.

What changes when AI agents take part in decisions?

AI removes some old problems and adds new ones. Models and agents can apply the same rule to every case, which reduces noise in frequent delegated decisions, and they can work under time pressure. But they also create automation bias, unclear authority over who approved an action, and decisions that are hard to explain after the fact.

Regulators now name these risks. Article 14 of the EU AI Act requires high-risk AI systems to be designed so that the people overseeing them can understand the system's limits, remain aware of "the possible tendency of automatically relying or over-relying on the output", interpret its output, decide not to use it or override it, and stop it safely. Under the AI Omnibus that entered into force on 27 July 2026, obligations for Annex III high-risk systems, such as credit scoring of individuals, apply from 2 December 2027, as reported by White & Case. In the US, the NIST AI Risk Management Framework (AI RMF 1.0, released January 2023 and now being revised) is the common reference for managing these risks.

The practical question after any AI-assisted decision is which context and policy it relied on, and who had the authority to act. ElixirData Context OS compiles decision-grade context and enforces policy and authority before an agent executes, and keeps a decision trace of the context, policies, authority, action and evidence for each decision.

Video: an escalation decision made under time pressure, with a P1 declared in nine minutes (ElixirData).

How should leaders fix problems in decision making?

A practical sequence that works for most enterprises, starting with the decisions that matter most:

1. List and classify the decisions. Identify the ten to twenty recurring decisions that drive cost, risk or revenue, and mark each as big-bet, cross-cutting, delegated or reversible.
2. Name one owner per decision. Record who decides, who is consulted and the date by which the decision is due.
3. Run a premortem on big bets. Gary Klein's premortem (Harvard Business Review, September 2007) asks the team to assume the project has already failed and list the reasons, which gives people with doubts a safe way to raise them early.
4. Audit noise in frequent decisions. Give several people the same set of cases, measure the spread, then introduce rules or scoring where the spread is large. AgentAnalyst lets teams ask questions of enterprise data in plain language, which helps when the audit needs quick comparisons across cases.
5. Build challenge into meetings. Rotate a devil's advocate, ask the senior person to speak last and collect independent estimates before discussion.
6. Record the decision and its reasons. Keep the options considered, the information used and the expected outcome, for human and AI decisions alike.
7. Review outcomes on a schedule. Compare results with the expected outcome and update rules and criteria. The Decision Review Agent analyses outcomes, detects patterns and recommends improvements, so each decision feeds back into the process.

When is a heavier decision process the wrong answer?

More analysis is not always better. For reversible, low-cost decisions, a formal process adds delay and can cause the very paralysis it is meant to prevent. Structured rules can also hide judgement that should stay with people, for example in rare cases the rules were not designed for. And a decision record only helps if someone reviews it. Match the weight of the process to the cost of being wrong and the cost of reversing the decision, and keep human override for anything that affects people's rights, money or safety. For enterprise programmes, see our approach to intelligence-driven decision making and data intelligence.

To map one high-stakes decision, the people who own it and the context and policy it depends on, book a Context Workshop. See what happens in the workshop.

Frequently asked questions

What are the main problems in decision making?

The most common are bounded rationality (accepting the first good-enough option), cognitive biases such as anchoring and availability, noise (different answers to the same case), groupthink, decision paralysis, time pressure, uncertainty and unclear ownership of the decision.

What is bounded rationality in decision making?

It is Herbert Simon's idea that people have limited time, information and attention, so they look for a solution that is good enough rather than the best one. Simon received the 1978 Nobel memorial prize in economic sciences for his research into decision making in economic organisations.

What is the difference between bias and noise?

Bias pushes judgements consistently in one direction. Noise is the unwanted variability between judgements of the same case. A noise audit, where several people evaluate the same set of cases, measures how large that variability is.

How can teams avoid groupthink?

Irving Janis suggested asking every member to act as a critical evaluator, having the leader withhold their view at the start, using independent groups on the same problem, inviting outside experts and rotating a devil's advocate role.

Do AI agents remove problems in decision making?

They can reduce noise and delay in frequent decisions, but they add automation bias, unclear authority and decisions that are hard to explain. Article 14 of the EU AI Act requires people overseeing high-risk AI systems to be able to interpret, override and stop them, so decisions need policy checks and a record of why they were made.