Premortem prompts: stress-test your plan with AI in 5 steps

Every plan looks good on the day you write it. That's the problem.

When you're excited about a launch, a new hire or a budget, your brain fills the gaps with optimism. The risks are there. You just can't see them from where you're standing.

A premortem flips the question. You assume it's six months from now and the plan has already failed. Then you explain why.

That small change works because explaining a failure feels like analysis, while predicting one feels like disloyalty. People who'd never say "I think this will flop" in a meeting will happily write the autopsy.

Start with step 1 of the Premortem series here, then come back to see where the next four steps take you.

Who the premortem prompts are for

Anyone with something real at stake and a date on the calendar:

• Founders about to commit cash to a launch, a market or a first sales hire.
• B2B SaaS teams shipping a pricing change, a new tier or a big feature.
• Managers with a budget to defend, a hire to make or a strategy to present.

If you can describe the plan in a few paragraphs, you can run it through all five steps in under an hour.

The one rule: keep one chat open

The five prompts are a chain. Each one reads what the previous step produced, so run them in order in the same ChatGPT or Claude conversation.

If you have to start a new chat, every step has an optional slot where you paste the earlier output back in.

To make this concrete, we'll follow one example the whole way: a B2B SaaS team planning to launch a new "Growth" pricing tier between its Starter and Pro plans, with a launch date eight weeks out.

Step 1: Find the 7 likeliest ways your plan will fail

Step 1 casts the AI as a forensic failure analyst. You paste in the plan (goal, timeline, budget, team, and what success looks like in numbers) plus any context: market, constraints, past attempts, worries.

Then it imagines the plan died completely and writes the autopsy. You get 7 causes of death, ranked by likelihood, each with a month-by-month timeline, the fatal assumption behind it and the first warning sign you'd have spotted.

It also ends with one line naming the detail you should have given it and didn't. Pay attention to that line.

In our example: the SaaS team describes the Growth tier, its price and the target of moving existing Starter customers up. The autopsy's top cause is that Pro customers downgrade to Growth because the feature gap is too small, so revenue falls while sign-ups look healthy. The missing detail it flags: the team never said what share of Pro customers use the features held back for Pro.

Step 2: Get a blunt verdict on the fatal flaw

Seven risks is a list. A list is easy to nod at and ignore.

Step 2 forces a choice. Same chat, you add your risk appetite (low, medium or high) and the things you can't change, like a budget cap or a fixed launch date.

It comes back with four short sections: the most likely cause, the most dangerous cause (often a different one), the hidden assumption you didn't know you were making, and a fatal flaw check that starts with either "Fatal flaw:" or "No fatal flaw".

In our example: most likely is slow Starter upgrades. Most dangerous is the Pro downgrade wave from step 1. The hidden assumption is that customers pick plans on features, when the team's own plan shows most buy on seat count. Verdict: no fatal flaw, unless the Growth tier ships with the same seat limit as Pro.

Step 3: Rebuild your plan so it survives its own autopsy

Now you fix it. Step 3 turns the AI into a plan surgeon that has already seen how the plan dies.

You paste in the constraints you must keep and the things you're willing to change: scope, pricing, sequencing, people, tools.

You get back a full rewritten plan in the same shape as your original, a change log table (failure mode, what changed, why it closes the gap, what it costs), a pre-launch checklist of 3 to 5 tests with clear go and walk-away results, and a residual risk section listing what's still open.

In our example: the rebuilt plan caps Growth at a lower seat count than Pro and moves one reporting feature back up to Pro. One checklist item: show the new pricing page to 10 current Pro customers before launch. If more than a few say they'd switch down, walk away from the tier as designed.

Step 4: Attack your plan as the rival who wants it to fail

Your plan now survives your own worst-case thinking. Step 4 tests it against someone else's.

You name a rival (the market leader, a competing team inside your company, or let the AI pick the most dangerous one), your market and your launch date. The AI reads your rebuilt plan and plays that rival, in character.

It ranks the 3 weakest points from the rival's side, plans launch week day by day, names the blindside move you'd be least likely to see coming, then steps out of character to give the cheapest defence for each numbered attack.

In our example: the rival is the category leader. Its blindside is a limited-time discount on its own mid-tier plan, timed for the team's launch week, aimed at prospects comparing prices. The cheapest defence: lock in a founding-customer price for anyone who signs up in the first month.

Step 5: Set tripwires that warn you before the plan fails

The last step makes sure you'll notice trouble while you can still act. Step 5 casts the AI as an early-warning analyst.

You give it your start date, the data you can already see (sales dashboard, sign-ups, ad spend, customer emails) and how often you'll check.

For every failure mode and every rival attack in the chat, it sets one tripwire: a measurable signal with a threshold, where you'll see it, the week to check, and the first action to take within 48 hours if it trips. The output is one table sorted by check week, plus a 3-line summary for your diary.

In our example: one tripwire watches Pro-to-Growth downgrades in the billing dashboard from week 2, with a threshold the team sets from its current monthly churn. If it trips, the team pauses Growth promotion to Pro customers and reviews the feature split within 48 hours.

What you walk away with

After five prompts in one chat, you have a ranked list of failure causes, a clear verdict, a rebuilt plan with a pre-launch checklist, a defence against your sharpest rival and a dated table of warning signals.

That's a plan you can take into a board meeting, a team kick-off or your own diary with your eyes open.

Pick the plan that's keeping you up at night and run step 1 of the Premortem now. Keep the chat open, and work through to step 5.