How to Prompt AI to Find a Business Plan's Blind Spot

Most AI reviews of a business plan ask the wrong question. They ask what could go wrong, and get back a list of a dozen scattered risks, none of them decisive enough to act on. The sharper question is narrower: what is the one thing this plan assumes will keep being true, that if it stopped being true, would break most of the rest of the plan along with it. Every business plan is a bet on some single load-bearing assumption, whether the founder wrote it down or not. A channel that keeps converting. A vendor price that holds steady. A customer segment that hasn't churned yet, staying put. Prompting AI to find that one assumption, instead of brainstorming a pile of failure modes, gets you a sharper answer and gets it to you faster.

This Is Not a Pre-Mortem

Swarmz has a separate guide on imagining a wide set of ways a project could fail, and that technique is deliberately broad. It wants twenty scenarios so you can triage them by likelihood and severity. This technique is deliberately narrow. It wants one sentence: the assumption that, if wrong, takes the whole plan down with it, not just one line item.

Run both if you have the time. But run this one first, because it tells you whether the pre-mortem is even worth doing in depth. If the load-bearing assumption is shaky, most of the twenty downstream scenarios are academic until that's resolved.

Every Plan Is Standing on Something

Plans read as declarative sentences. "We'll acquire customers through content and convert at a healthy rate" reads like a fact. It's actually a bet: that content keeps working at the same efficiency, that the conversion rate doesn't decay as the audience widens, that competitors don't outspend on the same channel before you get traction.

The words "assuming this holds" rarely appear next to the actual bet, because saying the bet out loud makes it visible, and visible bets invite pushback. That's exactly why AI is useful here. It has no stake in the plan looking finished, and no incentive to leave the load-bearing assumption unstated the way the person who wrote the plan might, even unconsciously.

The Prompt Template

Paste the plan in full, not a summary of it, then use a prompt that forces a single answer instead of a list. Something close to this:

"Read this business plan in full. Do not summarize it and do not list multiple risks. Instead, find the single assumption that the entire plan depends on, meaning: if this one thing turned out to be false, most of the rest of the plan would stop making sense. State it as one sentence in the form 'This plan only works if ___.' Then explain in two or three sentences why the plan is more exposed to this assumption than to any other, pointing to specific numbers or claims in the text."

A few things make this prompt do its job:

1. Paste the whole plan, not a summary. A summary already strips out the specific numbers and phrasing the assumption is hiding behind.

2. Ask for exactly one sentence. The moment you allow a list, the model reverts to hedging with three or four assumptions instead of committing to the one that matters most.

3. Only after it commits, ask for the second most load-bearing assumption as a separate follow-up. Asking for both at once produces two half-hearted answers instead of one sharp one.

4. Require it to cite a number or a specific claim from the plan, not a generic statement about markets or hiring in general.

What a Weak Answer Looks Like

A weak response usually restates something already sitting in the plan's own risks section, rather than surfacing something the plan left unsaid. It reaches for something generic, and it hedges instead of committing. It sounds like this:

"This plan assumes the market stays favorable, that hiring goes as planned, and that customers keep responding well to the product."

That's three assumptions dressed up as one answer, none of them tied to a specific number in the plan, and none of them telling you which one to actually go check first.

What a Strong Answer Looks Like

A strong response commits to one sentence, ties it to a specific figure or claim in the plan, and explains what's missing that would tell you whether the assumption already broke. It sounds more like this:

"This plan only works if the current month-over-month waitlist growth continues at a similar rate once paid acquisition replaces the founder's personal network as the main channel. The year-one revenue target is built directly on the waitlist growth curve, but signups driven by founder outreach tend to convert differently than signups driven by paid channels, and the plan doesn't show any test of a paid channel yet."

Notice the difference. The strong version names a mechanism, ties it to the specific number the plan leans on, and flags exactly what evidence is absent. You can act on that. You can't act on "the market could change."

Turning the Assumption Into Something You Can Check

Once you have the sentence, the natural follow-up is: what would we observe in the next thirty, sixty, or ninety days that would tell us this assumption is holding or already breaking. That turns a one-time exercise into an operating checklist rather than a philosophical debate you have once and forget.

It also helps to run the resulting assumption back through a model instructed to argue against it rather than confirm it. A model told to agree will usually agree with whatever hedge you propose next, so getting AI to push back instead of agreeing is a useful second pass once the assumption is named, so it gets stress-tested rather than just restated back to you in softer language.

If the load-bearing assumption turns out to be a vendor cost holding flat for the next two years, that's a case for revisiting the vendor relationship directly, including how you'd prompt AI to negotiate a lower price with that vendor, rather than leaving the whole plan's margin resting on someone else's pricing staying still.

Where This Fits in a Broader Review

This is one narrow technique inside the broader discipline of prompt engineering, and it works best as a first pass, not a replacement for the rest of a review. Use it to find the one thing worth arguing about before you spend an afternoon brainstorming everything else that could conceivably go wrong.

Frequently Asked Questions

Does this replace a pre-mortem or a SWOT analysis?

No. It's a fast filter that tells you where to spend your attention. A pre-mortem or SWOT still has value for surfacing the wider set of risks once you know which single assumption matters most.

What if the AI names an assumption I already knew about?

That's still useful. Knowing about an assumption and having it stated as the single thing the plan depends on are different. The second version is easier to test, discuss, and put a number on.

Can I run this on a plan I didn't write myself?

Yes, and it tends to work even better there, since you have less attachment to the wording and are less likely to unconsciously nudge the model toward a softer answer.

How often should I re-run this prompt?

Any time the plan changes materially, and at minimum whenever you update the numbers behind revenue or cost projections, since those are usually where the load-bearing assumption lives.

Does the plan need to be finished before I try this?

No. A rougher, earlier draft often makes the load-bearing assumption easier to spot, because it hasn't been polished into the kind of confident language that buries a bet inside a declarative sentence.