Before Spending on Ads, I Asked AI to Imagine the Campaign Failing

“A month from now, this campaign has failed”

The copy is ready. You've chosen an audience and prepared the landing page. Before launching, hand the brief to AI with one extra question: “Imagine it's a month from now and this campaign has missed its target. What did we overlook?”

A premortem starts with an imagined failure and works backward to possible causes. Teams use it to review risks and responses before a project begins; Atlassian includes the approach in its Team Playbook. Here, we'll apply that idea to a marketing campaign.

Give AI a brief it can actually challenge

Imagine a Korean brand called Afternoon Shop launching single-serve pour-over coffee bags. Its marketer wants to reach office workers who enjoy afternoon coffee. “What do you think of this ad?” leaves too much open. Start with the offer and the behavior you want to encourage.

• Product: a box of 10 single-serve coffee bags for ₩18,000. Shipping costs extra and appears at checkout.
• Audience hypothesis: office workers who drink coffee at their desks in the afternoon. Customer research hasn't validated this yet.
• Ad copy: ‘Make an ordinary afternoon special. Discover premium coffee.’
• Landing page: a styled product photo first, then the coffee's origin and brand story. Pack size, price and brewing instructions appear farther down.
• Goal: 100 new-customer purchases in four weeks, with an ad spend cap of ₩1.5 million. Purchases are counted using orders confirmed in the store.

If all 100 purchases were acquired through advertising and the full budget were spent, ad cost per purchase would be ₩15,000. A ₩18,000 selling price doesn't tell you whether that's affordable. Check what's left after product, packaging, shipping contributions, returns and other costs. An expectation of repeat purchases isn't validated revenue yet.

Ask where each possible failure comes from

Go beyond “be brutally honest.” Each critique needs a link to the material you've supplied so you can decide what to change. Start with three possible risks to keep the review manageable.

That last request matters. Imagining failure directs attention toward weaknesses. Following every criticism could lead you to replace good work too. Ask what to preserve and what evidence would make the critique no longer apply.

“Shipping is the problem” isn't a finding yet

The brief could prompt the following critiques. What matters is how you would check them. An AI statement that office workers are price-sensitive doesn't establish why your customers buy.

You can check when shipping costs appear by inspecting the page. Whether they cause customers to leave is a separate hypothesis. High abandonment could also reflect payment errors or other issues. If you have reviews or support questions, supply the original material and its date range. Distinguish feedback about this product from feedback about similar products.

Revise three parts of the plan

A review doesn't mean rebuilding the audience, product and offer. Afternoon Shop could keep its brand photography while making purchase information easier to find.

We don't yet know whether the revised copy will sell more. It does make the product, occasion and purchase requirements more explicit. Avoid adding unverified claims such as ‘ready in three seconds’ or ‘a taste everyone loves.’

Leave one question for the next experiment

Revising a plan and learning which change caused an improvement are different tasks. If you change the ad and landing page together and purchases rise, you can't isolate which revision made the difference.

This time, test whether copy describing a specific usage occasion leads to purchases. Give both ads the same landing page with accurate product and shipping information. Keep the audience, image, offer and delivery conditions as comparable as possible. Where supported, use an A/B testing feature to compare the copy and reduce the effects of routine ad allocation and other variables.

1. Write the hypothesis: an afternoon cup at your desk will communicate the buying occasion better than the word ‘premium.’
2. Define the difference: compare abstract copy with usage-focused copy using the same image.
3. Choose the decision criteria in advance: review purchases and cost per purchase alongside CTR.
4. Set a spending cap, duration and stopping conditions: account for typical purchase volume and conversion reporting delays. Resolve payment or tracking errors first.
5. State the limits of the result: if there are too few purchases, leave the outcome undecided. Check whether an observed difference could be due to chance.

There isn't a universal minimum number of days or purchases for every campaign. The budget and target in this example aren't benchmarks. A small test limits spending; it doesn't make a small amount of data conclusive.

Bring past campaign records into the review

You can start a premortem with a brief alone. If you've run related campaigns, you have more evidence to compare. Connect an ad account to Pango Neuro, specify the campaigns and period, and ask it to retrieve relevant performance data to review your assumptions.

Platform-reported purchase conversions may differ from new-customer orders in your store. You need separate order and cost records to identify new customers and evaluate first-order profit. Reading historical ad performance doesn't reveal customers' thoughts or predict the next campaign's outcome.

Questions about AI campaign premortems

Change one question in your next pre-launch meeting. Ask, “If we miss the target, what should we check now?” A useful premortem leaves you with a revised brief and questions to investigate this week.