Short answer: optimize for purchase if the ad set can collect about 50 purchases in seven days. If it can't, optimizing for initiate checkout is a fair first step, because checkouts happen more often and Meta can actually learn from them. But set the exit rule before you launch: once checkouts pile up with no purchases, around 20 in and none out, switch the event back to purchase.
I work on Everix, and moving a small account one step down the funnel is advice I give often. The part I used to leave out is when to undo it. This page is that part.
What each event asks Meta to find
The optimization event is an instruction about who to show the ad to. Optimize for purchase and Meta looks for people who resemble your buyers. Optimize for initiate checkout and it looks for people likely to open the checkout page. Those groups overlap, but they are not the same: some people reach checkout only to see the final price, shipping or delivery date, and leave. On a small account, Meta can get very good at finding exactly them.
That is the trade the event ladder in why a new pixel can't run purchase-optimized ads makes on purpose: more events to learn from, aimed one step further from revenue. It is a good trade while that step still predicts sales.
When initiate checkout makes sense
- The ad set can't reach about 50 purchases a week on its budget. At $30 a day that would take a cost per purchase near $4, which few products hit. The learning limited arithmetic walks through it.
- Purchases are tracked but rare, and checkouts happen several times as often.
- Your checkout itself works. Buy something through your own ad in the Facebook or Instagram in-app browser first. If that fails, no event choice will help, and these are the usual checkout breaks.
The exit rule: 20 checkouts, zero purchases
Write the exit rule down before launch, not after two weeks of watching. Mine: once the ad set has sent 20 people to checkout and none has paid, the experiment has answered, and the answer is no. If even 1 in 5 of your checkouts normally ends in a sale, 20 in a row with none would happen by chance only about 1% of the time. Past that point, more checkouts won't change who Meta is finding. They just bring more of the same people.
Two checks first, so you switch for the right reason: confirm the purchase event is still arriving (Events Manager shows when Purchase was last received), and run one test purchase through the in-app browser. If both are fine, the event is the problem, not the store. For what this pattern looked like across real accounts, see add to carts, no sales in 17 small stores.
Check it in Ads Manager: add the Checkouts initiated and Purchases columns at the ad set level and read them since the last significant edit. Checkouts initiated at 20 or more with Purchases at 0 means the exit rule applies.
How to switch back without wasting the week
Meta counts a change of optimization event as a significant edit, so the ad set starts learning again either way. I'd create a new ad set optimized for purchase with the same budget and creatives, pause the checkout-optimized one, and leave the new one alone for seven full days. On a small budget it will show Learning limited. That label means slower, not off: Meta keeps delivering, it just won't call the learning finished.
Judge it on purchases once spend reaches about two to three times what you can afford per sale (the arithmetic behind that number). If it sells, keep it and stop editing. If it still doesn't, the problem is more likely price, trust or the offer than who sees the ad, and that is where the next week goes.
What Everix does here
Everix reads each lane's funnel every day: clicks, adds to cart, checkouts, purchases. When a sales lane keeps collecting carts and checkouts with no purchases past a spend floor, it flags the break in that day's opinion, says whether people leave before entering payment details or enter them and can't pay (when your store reports that event), and asks before pausing anything. Which event a lane optimizes for stays your call; noticing the moment it stops working is ours.