Founder Journey

Module 9 · Grow

Meta ads

How the algorithm works, what to trust, and why you don't touch a winner.

Meta is an impulse channel. Nobody's searching for you. They're scrolling, your ad stops them, and they decide in seconds.

How Meta actually works

  • The algorithm targets. You don't. Broad targeting with strong creative and clean conversion data usually beats hand-picked interests.
  • It learns from your conversion data. Every purchase teaches it who buys. Broken tracking means it learns who clicks.
  • Creative is the lever. See the creative module.

What you control: creative, offer, budget, landing page, and data quality. What Meta controls: who sees it, where, and what each impression costs.

Keep the structure simple

  • One campaign optimized for Purchase
  • One or two ad sets, broad targeting
  • 3–6 creatives per ad set
  • A landing page that matches the ad

More campaigns split your budget and your data. The algorithm learns faster with fewer, bigger buckets.

The learning phase

A new ad set starts in a learning phase while Meta figures out who converts. Meta's guidance is that it needs about 50 optimization events in a week to exit.

During learning, results bounce around. That's normal. Don't judge an ad set in its first few days.

Significant edits can send an ad set back into learning: big budget changes, targeting changes, new creative, a new bid strategy, or pausing it for a while.

Rule: don't touch a working campaign

If a campaign is working, leave it alone.

Every edit risks resetting what the algorithm learned. The most common way founders kill a winner is by "optimizing" it: bumping the budget hard, swapping creative, tightening the audience.

To scale a winner:

  • Raise the budget in small steps, days apart, and watch cost per purchase
  • Or launch the winning creative in a new campaign and leave the original untouched
  • Test new creative in a separate test campaign, not inside the winner

Before you change anything, ask: is it actually broken, or am I bored?

What to trust and what not to

Don't take at face value Why Trust instead
Meta's reported purchases and ROAS Meta credits itself, including people who only saw the ad, depending on your attribution setting Shopify orders and revenue
Today's numbers Attribution keeps updating for days The trend over 7+ days
Recommendations and optimization tips They're suggestions, and many push more spend or broader delivery Your own cost per purchase versus your margin
High Event Match Quality It means Meta can match your events to people. It doesn't prove the events are correct or unduplicated A real test order that arrives once
Click-through rate alone Clicks from people who don't buy are expensive Cost per purchase
One great day Small numbers swing hard Consistent results across a week

The number that tells the truth

Blended return: total revenue in Shopify divided by total ad spend across every platform.

If Meta says it's crushing it but blended return isn't moving, Meta is taking credit for sales that would have happened anyway.

Read the ad in this order

  1. Hook rate: did people stop scrolling? If not, fix the first two seconds.
  2. Click-through rate: did they care enough to tap? If not, fix the message or offer.
  3. Landing page conversion: did visitors buy? If not, fix the page, price, or trust signals.
  4. Cost per purchase: is it below what a customer is worth to you?

Each step points to a different fix. Don't change the creative when the landing page is the problem.