The most powerful use of AI at the office has nothing to do with writing copy or summarizing your inbox. It's the question nobody asks out loud: does my plan actually hold up? According to McKinsey's 2025 State of AI research, 88 percent of organizations now use AI regularly in at least one business function. Almost all of that use goes toward execution, rarely toward stress-testing decisions before they ship. Five structured questions to an AI model can surface, in twenty minutes, the blind spots your team won't name, because group dynamics, politeness and hierarchy tend to beat the desire to be honest.
Why won't your team tell you what's wrong with your plan?
In 1972, psychologist Irving Janis coined the term groupthink: the tendency of teams to rank consensus above quality. More than fifty years later, not much has changed. In most organizations, real criticism of a plan shows up only when it's already too late, when a client walks away or a competitor gets there first.
The pattern goes wider than any one team. A separate MIT study found that roughly 95 percent of organizations see no measurable return from their generative AI investment, part of what researchers now call the GenAI Divide. One reason: the tools get pointed at the easy tasks, while the hard questions go unanswered.
This rarely comes down to incompetence. The mechanism is social. Someone who says "I see three reasons this fails" in a meeting quickly becomes the team's designated pessimist. That's especially true in flatter organizations, where consensus is the unwritten norm. The result: your team nods the plan through, you send the proposal, and three months later the client asks exactly the question nobody in the room was willing to raise.
An AI model doesn't have that social problem. It has no stake in the outcome, no position to protect, no relationship to manage. You can ask it to tear your plan apart with the instruction "be as critical as possible," and it does exactly that, without hedging. For context, none of this is an argument for cutting your team out of the loop. It's a way to have the conversation your team is avoiding.
What's a red-team exercise?
The term comes from the military: a red team is tasked with undermining a plan, hunting for weak points and testing the defense. In cybersecurity it's now standard practice. Companies pay ethical hackers tens of thousands of dollars per engagement to find vulnerabilities before real attackers do.
The same method works on business decisions. Take your business plan, your proposal copy or your product launch and hand it to someone with the explicit brief to find the holes. The problem for a small team: you rarely have someone you can free up for this without it feeling like a personal attack on whoever wrote the plan. AI doesn't have that problem. It delivers criticism without ego.
Here's the thing: it's a bit like having an outside accountant audit your books, except this time it's auditing the plan you haven't executed yet. The costs you already see are the ones you already knew about. The investment is zero dollars on top of a subscription most people running a business already pay for. ChatGPT Plus runs $20 a month, Claude Pro the same. For that price you get a sparring partner available at any hour who has zero need to be liked.
Which five questions does your team not ask?
Not every question to an AI model produces something useful. The quality of the answer depends on the sharpness of the question. These five are tuned to the kind of decisions small business owners make every week: proposals, investments, hiring and product launches.
1. What assumptions am I making that I haven't backed up?
Every plan rests on assumptions. The gap between a solid plan and a risky one lives in the quality of those assumptions. Paste your plan into an AI model and ask: "List every implicit assumption in this plan that isn't backed by data or a source."
Typical results: you're assuming your target audience will pay a certain price, that implementation takes three months, that your current team has the capacity, or that the market stays stable. An AI model names this within a minute. Your team usually catches it only after the assumption has already collapsed.
2. What are the three most likely reasons this fails?
This is the pre-mortem method, developed by psychologist Gary Klein and later popularized by Nobel laureate Daniel Kahneman in Thinking, Fast and Slow. The question isn't "what could go wrong" (too broad), it's "imagine this plan has failed six months from now: what caused it?"
AI is particularly good at this because it has no emotional stake in your plan succeeding. Where your team responds with optimism ("we'll figure that out"), an AI model hands you a dry list of failure scenarios, including the ones nobody in the room says out loud.
3. What costs or risks are missing from my plan?
The most commonly missed costs in small business plans aren't direct expenses, they're indirect ones: the time your team loses to onboarding, the revenue you forfeit while everyone's busy with implementation, the subscriptions you'll need but haven't budgeted for. Ask AI for an analysis of everything that isn't explicitly in your plan.
TheAIDaily previously calculated what AI use actually costs a team. The same logic applies to any project: the costs you don't name explicitly are the ones that ambush you later.
4. How would a competitor solve this with half my budget?
This question forces lateral thinking. In most plans, the approach is just a continuation of how the company has always done things. AI can offer a genuinely different angle: "If someone were seeing this problem for the first time and had only half the budget, which steps would they skip, and what would they do instead?"
The result isn't a blueprint to copy, it's a mirror. You start to see which steps in your plan exist because that's how it's always been done, not because it's the best approach. That's precisely the insight your team can't offer, because the habit is invisible to whoever is inside it.
5. What's missing from this proposal that my client will notice after three months?
This is the single most valuable question for anyone who sends proposals for a living. Hand your proposal to AI with the instruction: "Read this like a skeptical client. What's missing that only becomes visible after three months of working together?"
Typical blind spots: fuzzy success metrics, an undefined project scope, vague delivery milestones, a pricing model that rewards more hours instead of better outcomes. A client doesn't see this on day one. By day ninety, they do.
When does this save you the most?
Not every decision justifies a red-team exercise. The rule is simple: the harder a decision is to reverse, the more an AI check is worth. Three moments pay off the most.
Before you send a proposal. A missed blind spot in a ten-thousand-dollar proposal costs you the deal, or worse, a client who walks away unhappy three months in. Twenty minutes of AI review saves weeks of rework.
Before you commit to a tool or platform. Plenty of small businesses sign a year-long contract based on a demo and a sales pitch. Hand the vendor's materials to AI and ask: "What are the five questions I should ask this vendor before I sign?" TheAIDaily covered how to pick the right AI use case for your business earlier.
Before you hire someone, or automate instead. HR research puts the average cost of a bad hire at roughly one and a half years of salary. A bad automation investment costs months of work with nothing to show for it. In both cases, a structured AI analysis helps you weigh the decision properly. The five questions that help you choose between hiring and automating follow directly from this one.
What can't AI judge here?
The five questions above are powerful, but they have limits. Three things AI can't assess about business decisions.
Relationship dynamics. AI doesn't know your client just went through a rough quarter, that your supplier is always two weeks late, or that your employee is halfway out the door. Human context stays human work.
Timing and market feel. "Is this the right moment to launch?" requires industry knowledge, instinct and network intelligence. AI can produce a market analysis, but recognizing the difference between "too early" and "exactly on time" takes experience.
Internal politics. Inside organizations, decisions are rarely purely rational. AI optimizes for logic, not for buy-in. The advice can be correct on paper and still be unworkable if two departments see it differently.
Use AI to sharpen your own judgment, not replace it. The questions improve your thinking. The decision stays yours.
A conversation before you hit send
You don't need to ask all five questions at once. Build it into a habit: before you send an important document, proposal or decision, open ChatGPT, Claude or Gemini and run two or three of the questions from this list. Twenty minutes. No cost beyond the subscription you're probably already paying for.
The math is straightforward. McKinsey's own numbers show that the AI "high performers," roughly 6 percent of organizations, attribute more than 5 percent of profit to AI. Everyone else is leaving that value on the table. One avoided mistake in a proposal, one well-considered investment, one carefully weighed hire is worth a multiple of twenty minutes.
Most organizations already use AI in some form. The question isn't whether you use it anymore, it's what for. As long as most people use AI to produce text while the same tool could be auditing their plans, its strongest use case sits idle. These five questions are a start.
