The “What Am I Missing?” Audit: How Founders and PMs Use AI to Find Blind Spots

Three days before I was supposed to pitch our product to a room of investors, I did something I’d never done before.

I pasted the entire pitch deck into an AI and asked it one question: “What’s the dumbest, most obvious hole in this story?”

The answer came back in fifteen seconds:

“Your slide 6 claims a 40% conversion rate, but slide 11 shows your churn is 22% per month. Any investor will ask why people are signing up faster than your product can keep them. You don’t have a slide that answers this.”

I had stared at that deck for six months. My co-founder had reviewed it. Two advisors had reviewed it. None of us caught it.

That’s when I realized something uncomfortable: the people closest to your work are the worst people to spot what’s wrong with it. Not because they’re bad at their job. Because they have the same blind spots you do.

This blog is the system I use now. Three prompts. Five minutes. Run them before any decision, launch, or pitch that actually matters.

Why Your Brain Can’t Do This Alone

You wrote the strategy. You picked the feature. You hired the person. By the time it’s on paper, your brain has already decided the answer is right — that’s why it’s on paper.

The job of the “What Am I Missing?” audit isn’t to find new ideas. It’s to attack the ideas you already love, from angles you’d never attack yourself from.

An AI is good at this for one reason: it has no emotional investment in being right. It will tell you your favorite feature is confusing without flinching.

Prompt 1: The Three-Adversary Audit (The Hero Prompt)

This is the one. If you copy nothing else, copy this:

Prompt — Three-Adversary Audit

“I’m going to share [a pitch / launch plan / strategy doc / product spec — pick one]. Don’t give me feedback yet.

I want you to play three specific roles, in order, and stay in character for each:

Role 1 — A skeptical investor / boss who has seen 100 of these and is bored. Their job is to find the one question that, if I can’t answer it, kills the whole thing. What is that question?

Role 2 — A competitor who just got handed this document and has 30 days to crush me. Where do they attack first? What’s the weakest assumption I’m making?

Role 3 — A frustrated user / customer / employee who this affects. What’s the thing I’m saying that sounds great on paper but will quietly piss them off in practice?

Give me each role’s answer separately. Don’t soften anything. Don’t add disclaimers.

Here’s the document: [paste]”

Three roles, three angles, three different blind spots. The reason it works: you can’t be your own skeptical investor and your own angry competitor and your own frustrated user. But an AI can be all three in a single response.

The first time you run this, the answers will sting a little. That’s the signal it’s working.

Prompt 2: The Meta Question (The 80% That Nobody Asks)

After Prompt 1, paste this:

Prompt — The Question I’m Not Asking

“Based on everything I just shared, what’s the question I should be asking that I’m not? Not a suggestion. A question. The one I’m avoiding because I don’t want to know the answer.”

This is the one that’s saved me the most.

When I ran it on our pitch deck, it asked: “Are you sure you want to raise money right now, or are you raising because that’s what the timeline you set in January said you would do?”

I closed the laptop. Walked around the block. Came back and changed three slides.

You’re not going to like the question every time. That’s exactly why you need to ask it.

Prompt 3: The Specific-Disaster Prompt

Use this one for launches, hires, or any decision with a clear “go” moment.

Prompt — The Day-One Disaster

“Imagine it’s [the launch day / one week after this hire / three months into this strategy]. Something has gone embarrassingly wrong, and in hindsight, the signs were already in the document I just gave you.

What is the most likely thing that went wrong? Walk me through it in three sentences:

1. What happened.
2. The signal in my plan that should have warned me.
3. The cheapest thing I could do today to prevent it.”

This works because “imagine a disaster” forces the AI to commit to a specific failure instead of vaguely listing risks. Vague risks don’t change behavior. A specific story does.

What the Output Actually Looks Like

Here’s a real example. I ran the three-adversary audit on a launch plan for a new feature. The angry-user response came back like this:

“You’re describing this feature as ‘AI-powered insights,’ but in your spec, the user only gets to see the insight after they’ve used the product for 14 days. So for the first two weeks, the entire selling point of the feature is invisible. The frustrated user is going to churn at day 7 thinking your product is broken or empty. You should either show a placeholder insight on day 1 or rename the feature so the promise matches the timing.”

We delayed the launch by ten days and added a placeholder. Day-7 retention went up 18%.

That paragraph cost me one prompt.

Three Ways People Get This Wrong

1. Asking leading questions. “Don’t you think this is a strong plan?” will get you a strong plan. Phrase prompts neutrally, or even adversarially. You want pushback, not applause.

2. Stopping after the first answer feels uncomfortable. The first sting is the easy one. Run all three prompts even if the first one already gave you something to chew on. The meta question (Prompt 2) is usually where the real insight hides.

3. Treating the AI’s answer as the truth. It isn’t. It’s a hypothesis you didn’t generate yourself. Your job is to read it, decide which 30% is right, and act on that. The value isn’t the answer — it’s that you’d never have considered the question.

When to Use This

• Before any pitch (investor, internal, client)
• Before publishing a strategy doc
• Before a big hire (paste the JD and the candidate’s resume)
• Before launching anything
• Before making any decision you’ve been thinking about for more than two weeks (long deliberation = you’re stuck, not careful)
• Before sending any document where the cost of being wrong is more than a week of work

Try It On Something Today

Pick the document on your laptop right now that you’re nervous about. The one you keep meaning to “polish one more time.” Paste it into the three-adversary audit and let it tell you what you already know but don’t want to see.

Then come back and tell me one thing: what was the question you weren’t asking?

That answer is the most valuable thing you’ll get from any AI this month.

If you’re building an AI-native company solo, the same systems-thinking runs through The One-Person AI Business.