A faster report helps only if the team can use it to make a sound decision. This guide follows one AI answer from the screen to the result.
One example, from answer to result
A regional retailer needs a new supplier for the holiday season. The AI tool compares five suppliers and recommends one. It is the cheapest, and its quality scores are good.
The answer. The recommendation is clear and well written. It took the AI a few seconds.
The check. The manager asks what the comparison left out. The operations team says the supplier has never delivered to two of the company's markets.
The action. The manager approves a small first order for one market. She names two conditions for a larger order.
The ownership. Operations owns delivery tracking. A named person handles late orders.
The result. The first deliveries arrive on time. The team expands the order with evidence, not hope.
This is a fictional example. Each step after the answer was a human decision.
Judgment is needed throughout
Judgment with AI means deciding what to use, what to check and what to do next. It is not a final proofreading step. It starts before the AI works: someone sets the objective, the permissions and who reviews the output.
Economists Ajay Agrawal, Joshua Gans and Avi Goldfarb describe AI as a tool that makes forecasts better and cheaper. They call deciding what outcomes are worth "judgment". In their model, the two work together, as long as the judgment is not too difficult.
Some AI answers help, and some mislead
In a 2023 experiment with 758 BCG consultants, those using GPT-4 completed 12.2% more tasks and worked 25.1% faster on tasks the AI handled well. On one business case designed to be outside its ability, 84.5% of consultants without AI reached the correct recommendation. In the two groups using AI, 60% and 70.6% did. Read the research summary.
The same tool helped on some tasks and misled people on another. Knowing which is which is part of the work.
Results need more than adoption
PwC's 2026 CEO survey asked 4,454 chief executives about AI. Of these, 56% said it had brought neither higher revenue nor lower costs in the previous 12 months. 12% reported both. The survey does not explain why. It shows that using AI and getting a business result are different things. Read the survey.
Checking also takes time. BCG's 2026 survey covered 11,749 employees, managers and leaders. In that survey, 52% agreed AI had increased the time they spend reviewing and correcting AI output. Read the research summary.
What a manager can do next
- Pick one AI-assisted decision your team makes each week.
- Write down who checks the answer, who decides and who owns the result.
- Ask what the answer could leave out, and what that would cost.
- Agree when the team should review the decision again.
Read next
- Questions to ask before acting on an AI answer
- What does judgment with AI mean?
- Try a simulation: approve or stop an AI rollout
Sources
- Agrawal, Gans and Goldfarb (2018). NBER Working Paper 24243. NBER.
- Dell'Acqua and others (2026). Navigating the jagged technological frontier. Organization Science 37(2):403–423. PDF.
- PwC (2026). 29th Annual Global CEO Survey, 4,454 CEOs in 95 countries and territories, September to November 2025. Survey.
- BCG (2026). AI at Work 2026. Slides.