---
title: Getting spend onto new Meta ads — push delivery vs the creative testing tool
slug: meta-push-delivery-and-creative-testing-tool
summary: Meta's two ad-level ways to give new ads real spend without separate testing campaigns — "Push delivery to this ad" (default 20% of budget for 7 days, rolling out to ~half of Heath Media's accounts in Oct 2026) and the Creative testing tool (2–7 ads split-tested on segmented audiences) — when each fits by conversion volume, how much budget to test with (10–60%), and the short-term performance hit.
domain: marketing
tags: [meta-ads, creative-testing, push-delivery, delivery, testing-budget]
related: [meta-creative-testing-by-budget, meta-breakdown-effect-judge-in-aggregate, meta-underspend-and-never-turn-off-ads, meta-strategy-by-spend-level, meta-ads-per-ad-set-and-budget-minimums, paid-social-creative-testing-framework, creative-hit-rate-and-spend-tiers, judging-creatives-by-spend-not-roas]
sources: [yt:9mUwsvSK0NY, yt:6TRy9eUAPVg, yt:r51Tj5qQGa0]
updated: 2026-10-24
---

# Push delivery vs the creative testing tool

## TL;DR
- **The problem.** New ads dropped into a working ad set get no spend, because the incumbents keep it all. Advertisers then kill the old winners to force spend, and performance drops.[^1][^2]
- **Push delivery to this ad** (ad level → *Show more settings*; rolling out, in about half of Ben Heath's accounts as of Oct 2026). It dedicates a share of the daily budget to the ad, **default 20% for 7 days**, then delivery returns to normal.[^3][^4]
- **Creative testing tool** (ad level, below the creative). Split-tests **2–7 ads**, up from 5, on non-overlapping slices of the audience with a share of budget you choose. Compare on cost per result for your optimisation event, **never cost per engagement**.[^5][^6]
- **Choose by conversion volume.** Testing 5 ads at once at 1–2 conversions/day can take a month or more, so use push delivery one ad at a time. Dozens or hundreds of conversions a week make the batch tool practical.[^7]
- Both keep tests **inside the existing ad set**, so there's no data fragmentation from extra testing campaigns or ad sets.[^8]
- **Expect worse short-term results.** Most new ads won't beat the best 2 of the last 3 months. You test for the occasional ad that changes the business, e.g. 4.8× ROAS against a 3× norm.[^9]

## Details
**How much budget to test** (Ben; Heath Media spends 30–40% of $100M+/yr on testing):
- typically **10–50%**, up to 60%
- **100%** if nothing converts
- 10–20% when campaigns are very profitable[^10][^11]

**How long.** Base duration on conversion volume, not days. A campaign with 20 conversions in an hour is as "mature" as one with 20 in two months. Low-volume accounts may need a month. If a test ends with ~3 conversions, rerun or extend it. There's no proxy metric that lets you judge early.[^12][^13]

**Best practices** (Ben):
- **Don't test around one-off sales** like BFCM; let Meta consolidate on winners. Repeating monthly sales are an exception, since learnings carry over.[^14]
- Push only **big changes**: new format, style, hook, avatar or angle. A new background colour doesn't need it.[^15]
- Don't try to split-test your way to "the one perfect ad". Post-Andromeda delivery is personalised, so a diverse set beats a single average winner.[^16]

**Jon Loomer's flow.** For an initial batch, use the creative testing tool only if you want a true A/B test, accepting less efficient delivery. Let ads run another week after the test ends. Publish later batches straight into the ad set without a test, and judge results in aggregate after a couple of weeks. Use push delivery for a few days on an ad you feel was unfairly neglected.[^17][^18]

## Disagreements / caveats
- **Inside the ad set vs a separate testing campaign.** Dara Denney and Smart Marketer test outside the scaling campaign so new ads don't disturb it ([[meta-creative-testing-by-budget]]). Ben (at low or medium volume) and Loomer test inside it with these tools. CTC adds a separate "underspend" ASC ([[meta-underspend-and-never-turn-off-ads]]).
- Push delivery and the testing tool can't run in the same campaign at once: the test option greys out while a push is active.[^19]
- Feature names and limits are from Aug–Oct 2026 and still rolling out.

## See also
- [[meta-creative-testing-by-budget]]
- [[meta-breakdown-effect-judge-in-aggregate]]
- People: [[people/ben-heath]], [[people/jon-loomer]]

[^1]: yt:9mUwsvSK0NY @ 01:15 — Ben Heath
[^2]: yt:9mUwsvSK0NY @ 06:44 — Ben Heath
[^3]: yt:9mUwsvSK0NY @ 00:00 — Ben Heath
[^4]: yt:9mUwsvSK0NY @ 04:46 — Ben Heath
[^5]: yt:6TRy9eUAPVg @ 25:24 — Ben Heath
[^6]: yt:6TRy9eUAPVg @ 26:39 — Ben Heath
[^7]: yt:9mUwsvSK0NY @ 10:06 — Ben Heath
[^8]: yt:9mUwsvSK0NY @ 06:44 — Ben Heath
[^9]: yt:9mUwsvSK0NY @ 11:37 — Ben Heath
[^10]: yt:9mUwsvSK0NY @ 01:15 — Ben Heath
[^11]: yt:9mUwsvSK0NY @ 02:21 — Ben Heath
[^12]: yt:9mUwsvSK0NY @ 03:51 — Ben Heath
[^13]: yt:9mUwsvSK0NY @ 15:26 — Ben Heath
[^14]: yt:9mUwsvSK0NY @ 13:55 — Ben Heath
[^15]: yt:9mUwsvSK0NY @ 13:55 — Ben Heath
[^16]: yt:9mUwsvSK0NY @ 16:20 — Ben Heath
[^17]: yt:r51Tj5qQGa0 @ 03:14 — Jon Loomer
[^18]: yt:r51Tj5qQGa0 @ 06:25 — Jon Loomer
[^19]: yt:9mUwsvSK0NY @ 08:19 — Ben Heath