AI adoption is accelerating across the workplace, but business value is not keeping pace.
McKinsey’s 2025 State of AI survey found that 88% of respondents said their organizations regularly use AI in at least one business function. Only 39% reported an enterprise-level EBIT impact.
The gap is not simply about access to better technology.
Most professionals can open an AI tool, write a prompt, and get a polished answer in seconds. The harder skill is judging whether the question is right, the evidence supports the answer, and action is justified.
I learned that discipline long before generative AI. At Arthur D. Little and McKinsey, good analysis started by defining the decision and testing the assumptions behind it.
AI makes those habits more important, not less.
To understand why, start with the most common failure: using AI to solve a problem the team has not defined properly.
AI Can Help Teams Solve the Wrong Problem Faster
AI becomes risky when a team starts looking for answers before it understands the problem.
Let’s say employee engagement drops. The team turns to AI and gets sensible responses such as pulse surveys, manager training, recognition programs, flexible working, and better communication.
But those ideas only help once the team knows why engagement dropped.
The real issue could be:
- Weak onboarding for new hires
- Higher workloads after a restructure
- Poor communication from one management layer
- Low trust in promotion decisions
- A problem concentrated in one team
Each situation needs a different response. First, look at where the problem appears, who is affected, and what has changed. Then use AI to explore causes, test explanations, and develop options.
Used in the right part of the workflow, AI can improve everyday employee performance by helping people research, analyze, and work through information faster.
Better AI Adoption Starts With Three Judgment Checks
In practice, judgment shows up at three points in an AI-assisted decision.
1. Before Using AI: Define the Decision Clearly
Start by clarifying what the team needs to decide.
Instead of asking:
“Our customer support team is becoming less productive. How can we improve performance?”
define what has changed:
“Customer support resolves fewer tickets per employee than six months ago, despite stable staffing. Which factors should we investigate before deciding whether the issue comes from process, workload or capability?”
This habit sits at the heart of using AI to structure business problems. Once the problem is clear, AI can break it into components, explore hypotheses and identify questions for deeper analysis.
2. While Using AI: Treat the First Explanation as a Hypothesis
A plausible AI answer gives the team somewhere to start. It should then become something they test.
Suppose an AI analysis suggests that declining sales are linked to higher prices. Before changing the pricing strategy, the team needs to understand why that explanation fits the evidence and what else could explain the same result.
They could compare price changes with conversion rates, check whether demand has fallen across every customer segment, and review channel performance. If another explanation fits the data better, the recommendation changes.
A simple question keeps the analysis open: “What else could be happening here?” From there, evidence and business context help determine which explanation holds up.
3. Before Acting: Make the Reasoning Defensible
Before a recommendation moves forward, the person presenting it should be able to defend the reasoning behind it. For AI-supported recommendations, that means answering four questions:
- What evidence supports this recommendation?
- Which assumptions still need attention?
- What new information would change our conclusion?
- What is the business risk if this recommendation is wrong?
The NIST AI Risk Management Framework takes a similar approach to AI governance. It calls for clear responsibilities, human oversight and organizational practices that support critical thinking around AI systems.
That oversight works when reviewers know what evidence, assumptions and risks they are responsible for checking.
Workplace Culture Shapes How People Question AI
Employees notice what their leaders reward, question, and ignore.
If managers praise speed but rarely ask how a conclusion was reached, teams start optimizing for fast answers. If uncertainty is treated as a problem, people become less likely to challenge an AI output that looks convincing.
That is one reason leadership in the AI era depends on creating room for questions, judgment, and open reasoning.
Leaders reinforce those habits through everyday signals:
- When speed receives all the recognition, teams prioritize completion over review.
- When uncertainty is treated as weakness, employees delay raising concerns.
- When managers ask only for the final answer, people stop explaining their reasoning.
- When leaders discuss assumptions and trade-offs, teams become more comfortable doing the same.
Managers can reinforce judgment in project reviews, meetings, and one-on-one meetings. Ask why the team believes an explanation and what information they checked. Recognize people who catch weak assumptions early.
AI Training Needs Real Decisions, Practice and Feedback
Many organizations start AI training with the mechanics of the tool. Employees learn how to write prompts, summarize information, create drafts, and automate routine tasks. That gives people a useful starting point, but real workplace decisions demand more than tool familiarity.
Tool training only matters if employees can apply it to real decisions. People improve when they can practice a skill, receive feedback, and use it again in real situations. That is why training designed to drive behavior change works best when learning moves beyond explanation and into repeated application.
Leaders and L&D teams can build it into training in five practical ways.
1. Give Employees AI Outputs to Critique
Give employees a realistic AI-generated recommendation with gaps in the reasoning. Ask what they trust, what needs verification, and what information is missing. This tests whether they can evaluate a polished answer before accepting it.
2. Practice Problem Definition Before Prompting
Give employees a messy business situation and ask them to define the problem before using AI. They should clarify what changed, what decision needs to be made, and what information they need. Managers can review the framing before solutions are generated.
3. Require an Alternative Explanation
When AI produces a convincing explanation, ask employees to develop at least one credible alternative. This simple habit prevents teams from settling too quickly on the first answer and helps them compare explanations against the available evidence.
4. Train Managers to Review the Thinking
Managers should review how employees reached a recommendation. Ask how they defined the problem, which assumptions they tested, and why they chose one option. This shows where judgment is strong and where coaching is needed.
5. Define Where Human Review Is Required
Set clear rules for which AI-supported decisions require human review, especially those affecting employees, customers, financial commitments, sensitive information, or legal risk. Employees should know when they can act independently and when another review is required.
Make Human Judgment Part of AI Adoption
As AI becomes more embedded in everyday work, organizations need to strengthen the judgment surrounding it.
Employees should define problems clearly, test explanations, check evidence, and understand the consequences of their recommendations. Managers can reinforce those habits through coaching and reviews.
Successful AI adoption depends on employees knowing what to ask, verify, and challenge. They also need to turn useful AI output into sound business decisions.
Views: 275