TikTok Spark Ads: boost a post that already works or make fresh ads?

Spark Ads let you run a TikTok post that's already getting organic traction as a paid ad, under the original handle, with the likes and comments still showing. Plenty of people treat it as the cheapest way to get proof into paid, since the social proof is already there. My take is that it only works when the organic post had real watch time, not just a pile of likes. A post with 40 likes and a 3 second average watch is still a weak ad, and boosting it just buys weak views faster. Boosting whatever happened to get lucky is a bad habit for the same reason. Has anyone seen Spark Ads beat fresh in-feed ads on CPA, or does it mostly lower CPM and leave conversions flat?

Watch time is the right filter, but who's watching matters just as much as the number. A post that pops organically usually skews toward people who already follow the account. Say a dental clinic's before-and-after video does well with its existing followers. Boost it to a cold audience and the CPM can look fine while conversions are bad, because the people who loved it were never the cold audience you're paying for. TikTok's analytics splits viewers into followers and non-followers, so check that split before judging Spark on CPA.

The follower split tells you whether the hook works for strangers. A before-and-after opening on the result works for people who already trust the clinic, but cold viewers need the context in the first two seconds. If non-follower drop-off is steep, re-cut the opening before you touch the budget.

Re-cutting is the right fix for cold viewers, but you give up the one thing Spark was supposed to give you. A re-cut is a new post, so it starts with zero likes, comments or watch-time history. A cleaner test is to keep the original as a Spark and run the re-cut as an in-feed ad against the same cold audience. If the in-feed version wins on CPA, the opening was the problem. If it loses, the social proof was carrying the Spark, and re-cutting alone won't fix that.