Compare the Mortgage · Meta · prepared 18 Nov 2026 · internal

Testing plan: fund every test, judge on qualified leads

How we keep testing while A3 runs, get the money spent on the test ads, and move lead flow towards leads that clear the bar (refi $400k+, first-home any size).

In one line: A3 brings the volume, but only about 1 in 5 of its leads is clearly over $400k. So we stop judging ads on Meta's cost per lead and judge them on cost per qualified lead from SettledLoop. Every test gets its own budget so Meta can't starve it, and the next round of creative names the loan size it wants, the way the lead sellers already do.

1. Where we are

NZ$17A3 cost per Meta lead, 7 days
21%A3 leads clearly over $400k (35 of 163 since 19 Oct)
~NZ$70A3 cost per lead clearly over $400k (est.)
~NZ$48T2 first-home cost per qualified lead (6 of 6 qualified)
Ad (completed SettledLoop leads, 19 Oct – 17 Nov)LeadsUnder $400kExactly $400kOver $400kFirst-home
A3 owing-band (Launch)16310126350
B1 stale rate (retired)85120
T2 first yes (test cell)60006
T1 retention desk (retired)42200
Other / untagged84121

Three things the data shows

  1. A3 asks for the leads we don't want. Its grid of balance bands starts at "Under $300k", and its median lead owes NZ$250k.
  2. The "exactly $400k" leads may not be real answers. The pay-less balance slider starts at 400,000, the qualifying line, so anyone who never moved it lands on the threshold and counts as qualified. Those 26 leads can't be trusted.
  3. T2 died when it moved into Launch. In its own test ad set it spent NZ$289 for 6 qualified first-home leads. After joining A3's ad set on 16 Nov it got 36 impressions and NZ$1.44, because Meta put the shared budget into A3. That's the starving problem in one example.
Live state, read 18 Nov ~4pm. Launch NZ$80, test cell NZ$30 and Buying NZ$50 = NZ$160/day, matching Greg's Ads Manager edits today. The Buying campaign is on but its ad set is still paused, so A11 and A15 aren't spending. A16 was rejected on 17 Nov and is marked "with issues". T2 is paused everywhere.

2. What the research says

2a. One ad per ad set: Meta's own docs

2b. Practitioners in 2026

They split roughly in half. For forced-spend testing: Ben Heath, Ty Pryor, TheOptimizer and Motion's guide. Against, letting Meta allocate: Brad Ploch ("ad sets are just folders"), Barry Hott ("mostly a CBO maxi" now), Andrew Foxwell ("funding your losers") and Charley T. Almost all of them sell training or software. Two points they agree on at small budgets: test 2 ads at a time, not 5, and each test ad must be clearly different.

2c. Platinum Finance, reverse-engineered from our ad library

Source: 155 ads on their page in the adlib DB, scraped 14 Nov 2026. They don't tag their links, so their ad sets and budgets can't be seen; everything below comes from creative and timing patterns.

What we seeEvidenceWhat it likely means
Lots of creatives, almost no copy variation155 live ads and 114 distinct creatives, but 2 headline/body pairs cover 140 of them ("Apply in just 2 minutes" 75, "Let's make life easier" 65)They test the visual and keep the words fixed, so each new creative is a cheap test of one variable
They re-launch winners41 ads (26%) are fresh copies of a creative that's already running, launched 40–100 days after the originalDuplicating a proven creative into a new ad set or campaign to get fresh delivery. Probably their version of "one ad per ad set" for scaling, not testing inference
Losers are left runningAll 155 ads live; oldest 276 days; median 44 daysThey let Meta stop spending on losers rather than switch them off. Our DB may only hold ads that were live when scraped, so this list is survivors only
New ads arrive in dropsAbout 3 new ads a week: 1–2 most days, plus batches of 5–8 on 7 Aug, 15 Sep and 27 Oct (roughly every 6 weeks)A steady trickle of variants, with a bigger refresh every 6 weeks
One landing page, every placementEvery ad goes to start.platinumfinance.co.nz; Advantage+ placements (FB, IG, Audience Network, Threads, WhatsApp)Funnel held constant; only the creative changes
Copy names the amount they want"Need $10,000, $20,000, or even $30,000?"The ad does the qualifying

Lesson for CTM: copy the system: a steady cadence, one variable per test, re-launching winners, the amount named in the ad. Don't copy the sameness. Greg's contact says they're stuck at about NZ$200k/month and ~NZ$25 CPL not observed, and our July diagnosis put that ceiling down to running only 3 messages.

2d. The biggest NZ lead-gen accounts we can actually see: OneChoice and NZ Seniors

Both hard-code their campaign names into their links, so their structure can be read directly from the ad library. Neither uses one ad per ad set. Their answer to starving is one campaign, with its own budget, per concept:

NZ Seniors campaign (all "Lead · Daily Cost Cap")Live ads
…_Age-2291
…_Creative-Themes-2105
…_Personas95
…_ASC (Advantage+ scale)64
…_Financial-Commitment · _Families · _Death-of-a-Friend · _Trends13–25 each
…_Old-Assets (old creatives re-launched, from 28 Oct)21

2e. The mortgage lead sellers state the loan size in the ad

Seller · brandLine in the ad
Alfred · Mortgage Connect"For homeowners with $400k+ left on their mortgage and a fixed rate rolling off."
Alfred · Mortgage Connect"For mortgages over $500k, this one's worth 60 seconds."
Alfred · Mortgage Connect"One of the banks is currently offering 1.25% cashback. On a $750,000 loan, that's over $9,000."
Impact Leads · Mortgage Easy (reSure's ~$105/lead benchmark)"On a $500k loan, every 1% you shave off puts roughly $5,000 back in your pocket a year."

Alfred's are mostly video, 14 of its 49 live ads name an amount, and the $400k+ line is the newest. Policy risk Meta's Personal Attributes rule bans implying you know someone's finances ("your mortgage is over $500k"). Third-person framing ("For homeowners with $400k+…") is the safer way to write it. Run any new copy past reSure compliance as usual.

3. The plan

3a. Structure: one budget per job, each test funded on its own

Same NZ$160/day total Greg set today. No change to reSure's agreed spend.

Launch · refi scale

NZ$80
CBO · unchanged
A3 owing-bandVolume engine. Nothing new goes in, because adding an ad resets its learning.

Test cell · refi tests

NZ$30
ABO · 1 ad per ad set
Now: A14 self-employed aloneA13 off. At NZ$30 there's room for one clean read.
January: 2 ad sets × NZ$15–20Next refi tests that state "$400k+" and look nothing like A3.

Buying · first-home + next-home

NZ$50
ABO · 1 ad per ad set
Ad set 1 · T2, NZ$25Rebuilt from its existing post ID so it keeps its likes and comments. The proven qualified-lead ad.
Ad set 2 · A15 pre-approval tag, NZ$25The new idea most different from T2.
On the benchA16 (after its rejection is fixed) rotates in for whichever loses at 7–10 days. A11 is held: it's the same "first yes" angle as T2.

Why this shape: one ad per ad set guarantees each test is funded, which is what killed T2 in Launch. Capping each campaign at 2 ad sets keeps each one at NZ$15–25/day, about one lead a day. That's the floor for a useful read at our CPL. And separate campaigns for refi and buying copy OneChoice's "one campaign per idea" at a size our budget supports.

3b. How we judge a test

WhenLook atKill if
Day 3–5 (~2–3k impressions)Link CTR, CPC, cost per landing page viewClearly worse than A3's soft metrics (A3 CTR ~2.3–3.5%)
After spend of 2–3× targetCost per qualified lead from SettledLoop, not Meta's CPLWorse than A3's ~NZ$70 per lead over $400k (first-home: worse than T2's ~NZ$48)
WinnerTakes its place in Launch / Buying via "Use existing post"Must take spend within ~1 week or it gets backed out

The bar is a qualified lead. An ad at NZ$40 per Meta lead where 80% clear $400k beats A3 at NZ$17 per lead where 21% do.

3c. Sequence

  1. Now (before the ~25 Nov pause): unpause the Buying ad set; split it into the two ad sets above with T2 from its post ID; switch A13 off. This gets T2 spending again and gives first-home a week of data before Christmas.
  2. Funnel fix: change the pay-less balance slider default from 400,000 to below the line (e.g. 250,000), so an untouched slider can't count as qualified. Then re-read A3's real qualified share.
  3. Fix A16's rejection so it's ready to rotate in.
  4. Build a qualified-refi batch for January: 2 creatives, each its own brief and design world (rule since 30 Oct). Both state the band in third person ("for homeowners with $400k+ left…") and use cashback-on-a-big-loan maths, the angle Alfred and Impact run. Ideally one video, because the lead sellers' winners are mostly video.
  5. January restart rhythm (Platinum's cadence at our size): one new test every 1–2 weeks into an open test slot, and a fresh batch about every 6 weeks. Retired ads that once performed get a later second run (the "Old-Assets" idea).
  6. Report on qualified leads: extend the daily read so every ad shows SettledLoop leads split into under $400k / over $400k / first-home, joined on utm_content. Meta's own lead count stays secondary.

4. Decisions for Greg

  1. Go on the Buying split (T2 from post ID + A15, NZ$25 each) and on switching A13 off?
  2. Change the slider default from NZ$400k to NZ$250k (funnel change on pay-less)?
  3. Write January's refi copy with "$400k+" in the ad (third person), subject to reSure compliance?
  4. Try Meta's built-in Creative test in the January test cell, as a cheaper alternative to one ad per ad set? It needs checking in Ads Manager first.

5. What's verified and what isn't

Appendix: research write-up — one ad per ad set

Mac Mini research lane, 18 Nov 2026. Tags: VERIFIED-DOC = Meta's own documentation; PRACTITIONER = named person with date; UNVERIFIED = secondary source or the lane's own reasoning.

Tags: VERIFIED-DOC = Meta's own Help Centre, developer docs or Meta for Business post (help pages carry no date; text pulled 18 Nov 2026). PRACTITIONER = named person, with date. UNVERIFIED = secondary or blog claim, or my own reasoning.

Executive answer

1. What Meta's own documentation says

(a) How budget is split among ads in an ad set (VERIFIED-DOC, high confidence)

(b) Ad set spending limits in Advantage+ campaign budget (CBO) (VERIFIED-DOC, high)

(c) Built-in Creative test (VERIFIED-DOC, high) — Set up a creative test in Meta Ads Manager (1423851372208214):

2. Practitioner views (2025–2026)

Who Position Date Sells
Ben Heath (Heath Media) Pro: a separate testing campaign (about 20% of budget) so new ads aren't starved by the winner, with several ad sets if budget allows; low budgets should run 2–3 ads 2025 videos (summaries, early 2025) Agency, course
TheOptimizer Pro: ABO with one creative per ad set at US$20–50/day for "clean data" ~mid 2026 SaaS tool
Ty Pryor (lead-gen guide) Pro: ABO with one concept per ad set, cost cap at 50% of target CPL, 20–30% of budget for testing ~Nov 2026 Agency
Motion guest guide (Ben&Vic) Both: CBO or ABO with one creative or concept per ad set; "ABO will allow you to get quicker results" 2025 Motion is creative-analytics SaaS
Andrew Foxwell / Foxwell Digital Hybrid: CBO by default with occasional forced-spend ABO tests. Forcing spend means "funding your losers"; on a tight budget, consolidate 13 Mar 2026; 24 Jun 2026 Community, courses, agency
Brad Ploch (Scalability School) Against: "meta is going to be right on the allocation more often than we are"; ads moved from CBO losers into forced-spend ABO "every single one of them fail"; ad sets are "just folders" 9 Apr 2026 Foxwell education
Barry Hott Changed sides. Earlier: control budget "to get an even read" (ABO). 2026: "mostly a CBO maxi"; CBO with cost caps, new creative as new ad sets, judged by spend at campaign level X post Jul 2026; Foxwell NYC recap ~Nov 2026 Consulting, course
Charley T (Tichenor) Consolidation: 3-2-2 dynamic or flexible ads; "Andromeda 2" means one CBO with a scale ad set of proven post IDs plus 2 testing ad sets; "fewer, better-fed ads" 3-2-2 from Feb 2023 (possibly outdated); Andromeda structure 2026 via toco.marketing Meta MBA course
Jon Loomer Uses native tools: the Creative test "transformed my process" (Oct 2025), then after Push delivery appeared he "no longer recommends starting new ads with a creative test" (Jun 2026). Calls 2–6 ads per ad set normal and one ad "risky" Oct 2025 – Jun 2026 Membership, courses
Dara Denney A separate testing campaign with one ad set per test and 2–4 variations each; doesn't require statistical significance 2024, possibly outdated; her 2025–26 structure is known only second-hand Consulting, YouTube
Sarah Levinger Consolidate; Meta "prefers ad families" that differ in hook, tone and format 31 Jul 2025 Tether Insights consulting
Nick Theriot Against volume testing without strategy: "I'm anti-WASTEFUL testing" 5 May 2025 Agency, course

The case for: it guarantees spend and data on each ad, avoids winner-takes-all delivery, and lets you read each ad on its own. The case against: it forces money into ads Meta predicts will lose. Splitting the conversions gives each ad set fewer and puts them all into permanent learning. Meta's own ad-volume page says "Combine ad sets." Because similar creatives share learning anyway, separate ad sets don't produce truly independent reads. And a winner found under forced spend may still fail once it competes in normal delivery (Ploch, Foxwell).

3. Mechanics and failure modes

4. Alternatives

  1. Meta Creative test (best fit). It runs in the existing test ad set (the 20% guidance matters less when there are no other active ads) or inside Launch with about NZ$20–30/day for testing. Constraints: Highest volume bid strategy only, it reports no confidence level, and Leads support is unconfirmed. (VERIFIED-DOC)
  2. A/B test (Experiments). Separate audiences with no overlap. Best for one challenger against the champion. It is heavier to set up and slow at this budget. (VERIFIED-DOC)
  3. Ad set spending limits inside CBO. Put the test ads in their own ad set inside Launch with a minimum spend. It is "a target range, not a guarantee," it works on ad sets only (so you'd still need one ad per ad set), and Meta discourages it. (VERIFIED-DOC)
  4. Push delivery to this ad. Reserve a share of the budget for one new ad in an existing ad set. It's in beta and may not be available. (PRACTITIONER, Loomer Jun 2026)
  5. Cost cap tests (Hott, Pryor). Spend only goes where Meta can hit the cap, which lowers waste but often under-delivers on small budgets. It is also incompatible with Creative test. (PRACTITIONER)
  6. Dynamic creative / flexible format / 3-2-2. One ad holding up to 10 assets. Good for finding a winning combination, but it reports in aggregate and is not a clean per-ad read. Meta suggests it for reducing ad volume. (VERIFIED-DOC 2720085414702598; PRACTITIONER Charley T)
  7. Separate campaign per test. Same as one ad per ad set, with more overlap and admin. No advantage at this budget. (UNVERIFIED)
  8. Rotating one test ad at a time. Spend is guaranteed and the read is clean, but it's slow (about 2 weeks per ad), and Meta warns that "informal testing" by switching things on and off is unreliable. It is reasonable at NZ$40/day. (VERIFIED-DOC caveat; UNVERIFIED suitability)

5. What changes at NZ$40–60/day with a CPL of NZ$16–60

Sources

Meta (VERIFIED-DOC; help pages undated, retrieved 18 Nov 2026)

Practitioners and secondary sources