A portfolio system for what we shoot — formats, talent, and locations run with intent, so we stop remaking the same winner and start compounding new ones.
The rule that runs it: protect the proven core, systematically go grab the untested bets. Every week ships a fixed slate of known winners plus a couple of real experiments — the experiments are the thing that keeps the account fresh.
Proven — data-backedIn test / to grabUntested — blind spot
Live account read · 23 Sep 2026
The account is geo-segmented — a campaign per market (Regional NSW, Sydney, Melbourne, Canberra, Young, Brisbane) plus a fresh Central Coast test just launched. Creative right now leans on the NSW 0% loan and Jan-1 rebate urgency angles, and winners are already being re-flipped across formats — the "Another Quote" sketchbook is running re-cut as a talking head. The per-market structure means landmark-local and per-city cutdowns drop straight in, and Central Coast is a clean fresh-market test bed.
01
The engine
Every week's ad slate is split into two jobs: exploit what wins, explore what's next. Roughly a third of output is net-new by design — that's the structural fix for "same thing over and over."
1proven
2proven
3proven
4proven
5test
6test
3–4 proven — known-converting formats/angles2 test — new format, talent or place
Maps to the retainer's ~6 paid ads/week. Organic sits alongside on the same principle.
Graduate
Hits the CPL range → into the CoreA test that lands in our proven CPL range earns a permanent rotation slot. We grade the creative on CPL; the spend calls stay with Ana & Vanessa.
Rest
CPL drifting up → rest & re-angleA Core format whose CPL starts climbing gets parked and refreshed with a new angle or hook — rested, not retired.
Shelve
CPL well off → shelved + loggedA test whose CPL stays outside the proven range gets recorded in a "tried it" bank, so we don't accidentally remake it later.
The cheapest net-new: re-flip a proven angle into a fresh format. A winning sketchbook becomes a talking head, a talking head becomes a b-roll voiceover — same proven message, new shell. Already happening in-account ("Another Quote"), and it feeds the test slate without a full new bet.
02
Proven core
What the account keeps scaling — the top performers on CPL and attributed lead volume. Two aces carry it; iPhone Notes is the reliable third. Shares are of the 3–4 proven slots each week.
Ace · Format
Sketchbook
Adam — hands + voice, overhead paper draw
$56–98CPL band
~850leads driven
The account's longest-running winner. same_house_different_quotes ($56 CPL / 451 leads) and another_quote (346 leads, still live 6 months in) are the workhorses — durable and efficient.
~40%of proven slateStill running
Ace · Format
Talking head — pain / quote-decode
Adam — to-camera, 4-beat urgency structure
$45–96CPL band
~650leads driven
Our most efficient angle family. stop_thinking_rebate ($45 CPL / 229 leads), four_numbers_two_hidden and 2014_system_still_works (still live) all perform — the pain / quote-decode structure is a reliable engine.
~40%of proven slateStill running
Support · Format
iPhone Notes
Adam — to-camera walking through a Notes file
$124CPL
118leads driven
Reliable at scale — bygs_iphone_notes ($124 CPL) was a heavy-rotation earner, and the checklist variant is live now. A rung above the aces on cost, so keep it in rotation without over-leaning.
~20%of proven slateRotate in
03
Test slate — the go-grab backlog
The 2 experiments/week. These are the things we don't have real data on yet and need to go shoot. Prioritised — each carries a hypothesis and a clear win condition, so a result graduates or kills cleanly.
Tony — talking heads at volume
P1
A real installer's face carries trust and authority differently to Adam. Barely tested (one legacy TH, tiny sample).
Grab
Batch on the monthly Canberra on-site day — Q&A, screw-up stories, "how many companies / why STAG". The "Hey Adam / Hey Tony" Q&A shell is already in rotation.
Win
CPL holds up against our proven talking heads → graduate as a second face.
Female voice + hands — sketchbook
P1
Every sketchbook to date is Adam's hands/voice. A female voice may broaden demo relatability and soften the "salesy bloke" read on our best format.
Grab
Same proven sketchbook scripts, new voice + hands. Low-lift on a format we already know converts.
Win
CPL lands in the Adam-sketchbook range → new rotation lane.
Landmark-local shoots
P2
A recognisable local landmark in the opening frame signals "we're local" and may lift conversion in that market. Never tested.
Grab
Per-city cutdown openers with the local landmark in shot — the account is already split by market, so they drop straight into the right campaign. Central Coast (just launched) is the clean first test.
Win
Beats the generic master's CPL in that market → adopt as a geo layer.
Sam — founder talking heads
P2
The owner's face is a distinct trust signal ("the boss himself"). Untested on paid.
Grab
Short founder-voice angles — why STAG exists, standards, the guarantee behind the install.
Win
CPL holds up against our proven talking heads → third face in rotation.
Skillshow — installer POV
P3
Watching the actual work done right is proof no talking head can fake. Proven format elsewhere, never run at STAG.
Grab
On-site: a clean install shot as process, voiced over the pain/standard it solves.
Win
CPL lands in the proven range → new authority format.
Reworked testimonial / Us-vs-Them
P3
Testimonials and statics have underperformed our proven formats so far — worth one proper rework before writing either off.
Grab
Phone-call-style testimonial + a tight Us-vs-Them static comparison (no competitor naming).
Win
CPL lifts into the proven range → keep; else shelve and log.
04
The three knobs
Every ad is a combination of format × talent × location. Here's where each knob stands — and where the blind spots are. The test slate above points straight at the grey.
Format
what shape the ad takes
SketchbookProven
Talking head — pain/decodeProven
iPhone NotesProven
B-roll + VO (national)In rotation
"Hey Adam" Q&AIn rotation
Us-vs-Them / staticWeak — rework
TestimonialWeak — rework
Skillshow — installer POVUntested
Talent
who fronts it
Adam — hands + on-cameraProven
Tony — installerBarely tested
Sam — founderUntested
Female voice + handsUntested
Real customersWeak
AI-generatedExplore
Location
where it's shot
Studio / neutralProven default
Jobsite / on-siteLightly tested
Landmark-localUntested
Home / lifestyleUntested
05
Target mix
Where we're steering the monthly output once tests report back — deliberately diversified across talent and format so no single lane can fatigue the account. About 40% of this mix is untested today; the go-grab backlog is the bridge to it.
20%
20%
20%
10%
10%
10%
10%
Adam — talking heads Proven20%
Sketchbook — Adam Proven20%
Sketchbook — female voice Test20%
Tony — talking heads Test10%
Sam — talking heads Test10%
iPhone Notes / other proven Proven10%
New formats / testimonial Test10%
Location overlay (applies across the mix): target ~25–30% of on-camera shoots framed with a recognisable local landmark for geo-conversion — deployed as per-city cutdowns, and only after the landmark test validates the lift.
06
Rules of the system
Real people lead
Real faces, hands and sites are the trust engine — that authenticity is what converts. AI-generated creative stays a deliberate, clearly-distinct test lane worth exploring, kept alongside the real-people core rather than replacing it.
Graduate
A test whose CPL lands in our proven range earns a permanent Core rotation slot. CPL is the yardstick — it's the shared metric we grade every creative on.
Rest & re-angle
A Core format whose CPL starts drifting up gets parked and refreshed with a new angle or hook — rested, not retired.
Shelve + log
A test whose CPL stays well off is shelved and recorded in a "tried it" bank, so it's never accidentally remade later.
Claim discipline
No "cheapest" or price-match, no brand-shaming, no unverifiable rebate/landlord claims, never frame sub-contractors as a red flag.
Stay in lane
This is creative rotation — format, talent, angle, place, graded on CPL. Spend, placement, targeting and budget stay with Ana & Vanessa; we read the numbers, we don't set them.