Online Ad Test

An online ad test is a paid ad you point at a not-yet-real thing. You change one variable, the headline, the image, or the audience, and let the spend tell you what pulls. You buy clean traffic on purpose so the read is controlled, not whatever showed up organically.

That's the whole move. You're not measuring how clever the ad is. You're measuring which version of it makes a stranger reach.

The question it answers

Run it to learn two things off one budget. Do people want this at all, and which way of describing it pulls hardest. That pair, demand and message, is why it earns its own page.

The taxonomy underneath this experiment comes from David Bland and Alex Osterwalder's Testing Business Ideas, which catalogs more than forty of these tests by the kind of evidence each one buys you. The online ad test is one of the cheapest tests in that catalog for proving people want it, and the cleanest one for proving which words land, because you're paying for the exact slice of audience you want to hear from instead of hoping the right people scroll past.

A fair flag before you go further. This overlaps the social-media teaser hard, and most of the time it's the same experiment. The difference is paid reach plus targeting versus organic. The teaser fishes the audience you already have. The ad test buys the audience you want to test, and controls who sees what. That control is the reason to reach for this one when the teaser can't answer your question.

How to run it cheaply

The minimum version that raises real confidence is small, fast, and disciplined. You don't need a campaign. You need one variable and one number.

1. Write down the variable. Change exactly one thing across your ad versions: the headline, the image, or the audience. One. If you change three things and one wins, you've learned nothing about which of the three mattered.

2. Write down the success number before you launch, so it can't bend afterward. The number that matters is cost per signup, ad spend divided by signups, not raw clicks, because a cheap click that never converts is just a cheap click.

   Two benchmarks tell you whether the ad is healthy enough to produce a good cost per signup. Click-through rate, the share of people who click the ad after they see it, around 1 to 2%. And signup conversion, the share of those clickers who actually sign up, above 5%. Multiply those together against what you paid for the clicks and you get your cost per signup. So the two percentages aren't a rival number, they're the two dials that move the one number you wrote down. Treat them as a common starting benchmark, not a finding. Your real target gets written for your audience, in your context.

3. Point a small spend at it for a fixed window. A week. Set the end date now. An experiment without an end date is a pet project with a cooler name.

4. Read the one number you wrote down, and decide. The version that wins on cost per signup is your answer. Ship it. Then change the next single variable and run again.

That's it. No engineering, no product behind the button yet, just the ad, the promise, and a page that records who meant it.

Worked example (illustrative)

Everything in this example is invented to make the mechanics concrete. The method is real. The benchmarks are real and credited. Nothing here is a result I'm claiming.

Picture a meal-kit service aimed at people who work twelve-hour hospital shifts. The messaging is already settled from an earlier test, so the open question now isn't what to say, it's who to aim at. You've got one ad that pulls. You don't yet know which slice of that audience is your beachhead, the first narrow group you win before you widen out to everyone else.

So you run the winning ad against three audiences, changing only the targeting. Night-shift nurses, EMTs, and on-call residents. Same ad, three targets, one variable.

The success read is cost per signup, written down first. Say it comes back like this. Nurses sign up at $6 a head, EMTs at $14, residents at $11. (Audiences and numbers all illustrative.) The nurses are your beachhead, and you didn't argue your way there in a meeting. You bought the answer. That's segmentation tested with money instead of opinion, for the price of a week of ad spend, before you've signed a single supplier, and it cost less than the catering for the persona workshop would have.

When to reach for it, and when not

Reach for it when you want a controlled read on one variable and you're willing to pay for clean traffic to get it. Targeting questions are the sweet spot. Who should we aim at, which headline wins, which image pulls. The paid reach is what makes the read clean, because you choose who sees the test instead of taking whoever drifts by.

Reach for the social-media teaser instead when you just want a fast, cheap read on demand and you don't yet need the targeting control. If you're already changing one variable and reading one number inside your teaser, good. You don't need both. The discipline is the point, not the label.

Reach for the landing-page test when the question has moved past which words to will the whole pitch convert, and you want a real page behind the click, not just an ad. The ad test usually feeds the landing page. The winning angle and the cheapest audience you find here are what you put on the page you build next.

And don't reach for any of them before you've named the riskiest assumption in your idea, the one that kills the whole thing if it's wrong. Find that, name the question underneath it, and pull the cheapest experiment that answers it. Sometimes that's this one. Sometimes it isn't. The catalog has the rest.