The Test Bench

Enter the password to open the dashboard.

Weekly command center

This week on the bench

—

01 — Targets

Weekly targets

Sourced is hand-tallied. The other four pull live from the scorecard below.

Testing Budget

$100 maximum initial spend per product campaign
Five static ads inside each product campaign
The $100 is shared between the five statics — not $100 per static
Stop the product at $100 total spend if it produces no promising signal
Pause individual statics only after meaningful delivery and clear underperformance
Mark ads receiving little or no delivery "untested," not losers
Reserve the remaining $50 for the strongest-performing product
If all three products perform, all three may advance

Product 1Up to $100

Product 2Up to $100

Product 3Up to $100

Best performer / reserve$50

Weekly hard ceiling$350

03 — KPI diagnostic guide

Where is the funnel breaking?

Use these patterns to identify the most likely bottleneck. They are diagnostic hypotheses, not automatic verdicts.

Results below approximately 1,000–2,000 impressions per creative are directional and may be noisy.

Funnel diagnosis

Weak CTR

Creative, hook or avatar problem

Strong CTR, but weak ATC

Intent mismatch, weak offer, poor product page, or product problem

ATC, but weak Checkout rate

Check price reveal, shipping cost, missing trust signals, or cart friction

Checkout activity, but weak purchase rate

Payment failure, unexpected costs, checkout friction, or insufficient trust

Economics verdict

CPA above break-even, no improvement

Currently losing money — kill

CPA below break-even but above target CPA

Continue optimizing

Below target CPA

Candidate to advance

04 — Evidence

Product scorecard

One saved entry = one product. Keep editing the same entry as it moves through gates — don't create new rows for individual statics, hooks, avatars or campaigns.

Validated winner = verdict set to Validated winner, with at least 20 purchases, CPA at or below target CPA, and positive contribution profit. Anything short of that stays Pending, Revise, Continue, Kill or Advance — not a winner yet.

Break-even CPA and target CPA auto-fill from your economics inputs — override either by hand if needed. CTR, CPC, funnel rates, CPA, ROAS and contribution profit always auto-calculate from your aggregated totals. Marking a product Advance or Validated winner freezes its test-period metrics — spend added afterward (e.g. while scaling) won't change its saved numbers or feed the probability panel.

Date	Product	Gate	Status	Spend	Purch.	CPA	Contrib.	Verdict

05 — Odds

Probability panel

These are mathematical scenarios based on an assumed independent hit rate — not a revenue forecast.

You choose the assumed hit rate; nothing here claims a universal product success rate. Formula: P(≥1 winner) = 1 − (1 − hit rate)^products tested.

Your estimated hit rate (%)

Products tested / week

Weeks

Before you test anything

What counts as testable

Almost any product can work with the right angle — but some products make the angle do all the work, and some do half of it for you. Screen for these six before you spend a single dollar on ads.

01
Legible in one frame. A viewer should understand what it is and roughly what it does within the first still frame, with the sound off. If it needs a voiceover to be understood before it can be desired, it's fighting you before the story even starts.
02
Names a pain, not a category. "Posture corrector" is a category. "Three weeks of WFH gave me a permanent hunch" is a pain. The product should map cleanly onto one sentence a real person would text a friend.
03
Impulse price band. Roughly $18–$55 retail, with landed product cost ideally below 25–30% of the selling price — enough margin to test, cheap enough to not require a consideration phase.
04
Sourceable this week. If you can't get a sample or a supplier link in under a week, it fails the volume game before it starts. Speed to first test beats a better product you can source next month.
05
Margin survives inefficient early data. Testing CPA commonly runs less efficient than mature CPA — but scaling isn't guaranteed to improve it, so don't bank on that. What actually matters: your contribution margin needs enough room to keep gathering conversion data while operating above your eventual target CPA. For a quick screen, modeling against 2–3x your break-even-derived target is a reasonable planning assumption — just don't mistake it for a promise.
06
Rides a wave or resists one. Either it's visibly trending right now (Method A), or the pain is evergreen and undramatized (Method B). Products in the dead middle — not trending, not urgent — are the ones that quietly burn a month.

Sourcing

Two ways in

Both are legitimate. Pick based on how much production time you have this week, not on which feels more "original" — originality is an angle problem, not a sourcing problem.

Method A

Trend-Rider

Pull this week's rising products from a TikTok Shop data tool (Kalodata, TrendTrack) — filter for steep unit/revenue growth in the last 7–14 days, not just high volume.
Pull every live ad already running for it via an ad-spy tool (Minea, PiPiADS) — read the comments on the top three, not just the ad copy.
Find the angle nobody's running yet, or the avatar being underserved in the comments.
Cut a cheap VO ad in CapCut using your own hook, same product, different story.

Best when: you have < 2 hrs for production and want proof of demand before you write a single word.

Method B

Avatar Miner

Pick one category you know is durable — not necessarily trending this week, just provably always-on (sleep, posture, skin texture, desk ergonomics).
List 8–10 avatars who feel that pain differently: the night-shift nurse, the new dad, the hybrid commuter, the competitive gamer.
Write one static ad per avatar, same product, same offer — only the opening line and the visual change.
Run all of them at once as a static batch (see Gate 1) and let the data pick the avatar for you.

Best when: you have a product already in hand and want to find who it's actually for, instead of guessing.

Foundation

Know your break-even CPA first

Every threshold in the loop below is derived from this one number, not from a flat percentage of retail — a $50 product with $8 in costs can tolerate a much higher CPA than a $50 product costing $25 to fulfill. Calculate it before Gate 0, not after Gate 3.

Break-even CPA = AOV − product cost − shipping − fulfillment − fees − refund reserve

Target CPA ~70–80% of break-even

Max tolerable = break-even CPA

Scale threshold at/below target, margin holding

Repeat-purchase products (a fragrance, a consumable) can tolerate more than first-order math allows. Once you have real repeat-rate data, recalculate break-even against a 60–90 day LTV instead of a single order — until then, use first-order math and don't guess at a repeat rate you haven't measured.

The core procedure

The weekly loop

Five gates. Every product idea, no exceptions, moves through them in order — and most die at Gate 0 or Gate 1, which is the point. The numbers below are a starting calibration, not law; see the note at the bottom.

01Mon–Tue

Source & shortlist

Run Method A or B to land on one product and 4–6 candidate angles or avatars. Write the hook line for each before you build anything — if you can't write a one-line hook, the angle isn't ready.

02Gate 0
20 min

Feasibility & risk screen

A fast, hard-stop check before you spend a dollar on ads. Any single red flag below is enough to reject the product outright, no matter how good the angle looks — this is a screen, not an audit, so keep it to twenty minutes.

Compliance IP risk, medical/body claims, platform ad restrictions

Fulfillment supplier reliability, delivery time, breakage, variant/sizing load

Exposure return/chargeback risk, customer-service burden

Creative can you actually produce original assets for it?

Tracking Purchase/ATC/Checkout events fire, UTMs set, mobile checkout tested, page speed checked, ads labeled by product·avatar·angle·variation

Skip this cluster and a broken pixel or a slow page will make a genuinely good product look dead by Gate 2 — it's the single most common false negative in this whole loop.

03Gate 1
Statics

Angle test

Build 4–6 cheap static ads, one per angle/avatar, same product and offer. This is the cheapest, fastest signal you'll ever get — don't skip straight to video.

Exposure 750–1,500 impressions / concept

≈ Spend roughly $20–45 at a $25–35 CPM

Judge on outbound CTR + CPC, batch-relative

Keep top 1–2 angles only

750 impressions is enough to spot the clear losers — it won't reliably rank a close call between your top two, so treat it as a filter, not a leaderboard. If your platform's automatic delivery won't split spend evenly across concepts (common with Meta's CBO), use its dedicated split-testing tool (Meta Experiments, TikTok Split Test) to force comparable exposure. And read outbound/link clicks, not overall engagement clicks — Meta's "CTR (all)" inflates this with profile taps that never left the platform.

04Gate 2
Motion

Story test

Turn only the surviving angle(s) into the full VO / brainrot-song story ad. This is where the real production time goes — spend it only on what already proved it could stop a thumb.

Spend greater of ~1–2× break-even CPA, or the traffic floor

Traffic floor 75–100 landing sessions minimum

30 sessions early warning only — not a decision

Judge on ATC rate, checkout-initiation, cost/ATC

Stop-loss ≈2–3× break-even CPA with zero purchases and weak shopping activity → kill

150+ sessions before you trust an add-to-cart read with real confidence. 75–100 is enough to catch an obviously broken page or offer, which is most of what Gate 2 is actually checking for. The stop-loss is a spending ceiling, not the target-CPA math in Gate 3 — it exists so a product with decent-looking engagement but zero purchases can't quietly drain a week's budget. Exception: if checkout behavior is strong but purchases still aren't landing, that's a page/offer problem, not a product problem — allow exactly one revision (see the diagnosis below) before killing.

05Gate 3
CAC

Confirm the economics

Let spend run until purchases accumulate enough to trust the number — not before. A CPA from three sales is noise, not signal, and a CPA from ten isn't final truth either.

Target CPA ≤ your break-even-derived target above

Kill bar CPA above break-even, no improving trend

10–15 purchases provisional validation, not final

25–30+ increasingly trustworthy; 50+ = solid ground to scale on

Confidence ladder: 1–3 purchases is anecdotal, 4–9 promising but unstable, 10–15 provisional, 25–30+ increasingly trustworthy, 50+ a real foundation. Ten purchases means "keep going," not "you're right."

06Gate 4
Fri–Sun

Expand, scale, or archive

Winners don't get a flat daily budget bump — they get a standing check. Losers get logged and closed out the same day.

Scale rule raise spend only while the rolling 3-day blended CPA, CVR, and margin stay within target

Typical step +20–30%, only when that check passes

Next batch start a fresh avatar/hook before the current one fatigues

Every idea gets a log row — win or lose

A stable $100/day campaign doesn't always behave the same way at $130/day — the rolling check is what catches that before it costs you a week's spend.

Compounding the loop

The test log

This is the one artifact that makes the loop worth repeating for weeks instead of restarting from zero each time. Every batch — win, lose, or ambiguous — gets a row. After a few weeks you're not just finding winning products, you're finding winning avatars and hooks that transfer across products.

Date	Product	Avatar / angle	Hook	Gate	Spend	Metric	vs. bar	Verdict
08.17	Posture strap	Hybrid desk worker, text neck	"3 weeks WFH gave me a permanent hunch"	Gate 2	$110	CPA $18.40	at target	Scale
08.24	Ring-light clip	Dorm student, video calls	"POV: you finally look awake on 8am calls"	Gate 1	$32	CTR 1.9%	+0.6 vs. median	Iterate
08.31	Cold-brew maker	Night-shift nurse, pre-shift routine	"I make tomorrow's coffee before I sleep tonight"	Gate 1	$28	CTR 0.7%	−0.4 vs. median	Kill

— three illustrative rows, not real data. The live Product Scorecard on the Command Center tab replaces this sheet.

The table above is a compressed read of the log for this page. The Command Center's scorecard captures the full set below in one flat row per test:

Core — once per product

Selling priceLanded costBreak-even CPATarget CPA

Per batch — every test

DateAvatarHookGateSpendImpressionsOutbound CTRCPCLanding viewsATCCheckoutsPurchasesCPAVerdict

Take the emotion out

What actually failed — then what to do

A bad static only proves one combination of avatar, hook, visual, offer, and product didn't work. Diagnose which layer broke before you decide the product itself is dead.

→
One angle is weak, others aren't. Kill the angle. The product and the rest of the avatar pool are still alive.
→
Every distinct avatar is weak. Kill the product. You've tested the concept itself, not just one execution of it.
→
Strong clicks, no shopping activity. Don't touch the creative yet — investigate the landing page, the offer, and whether the product disappoints on arrival.
→
Shopping activity, no purchases. Investigate price, trust signals, shipping cost/time, and checkout friction before you go back to the ad.

Then apply the verdict

Decide the trigger before you launch, not while you're staring at the dashboard wanting it to work.

Kill

The angle (or product) didn't clear the bar and you've already used your one iterate pass on it.

Below bar, no revision left → archive same day

Iterate

Close to the bar, or one specific element (hook, thumbnail, first line) looks fixable.

Within ~30% of the bar → one revision, same product

Scale

At or under your break-even-derived target, with a real sample size behind it — not a lucky hour.

At/under target CPA → raise spend only while the Gate 4 rolling check holds

What to run it on

Tool stack

Trend & competitor data

Kalodata / TrendTrack

TikTok Shop product, creator, and video data — use it to spot what's already moving, and to check whether your "original" idea already has a live ad before you build the story.

Ad discovery

Minea / PiPiADS

Search live and past ads across platforms by product, niche, or keyword — this is where Method A starts.

Swipe file & briefs

Foreplay

Organizes and tags the ads you save so a production brief takes minutes — it's a filing system, not an analytics tool, so don't expect performance data out of it.

Cheap production

CapCut + an AI voice tool

Keep one reusable template project per format (static, VO story, brainrot song) so a Gate 1 batch takes under an hour to build, not an afternoon.

Landing page

One reusable product-page template

Re-skin the same fast-loading page per test instead of building a new store each time — infrastructure time is time not spent testing.

The log

The Product Scorecard

The single most important tool on this list — now live, on the Command Center tab. The compounding value of this whole system lives there, not in any individual test.

Guardrails

Rules that keep it a volume game

Idea-to-launch cap: no more than 3–4 hours of your own time from picking a product to a live Gate 1 test. If it's taking longer, the product or the angle is too complicated — that's a signal too, not just friction.
One iterate pass, then move: a mediocre Gate 1 batch gets exactly one round of new hooks. If the second round doesn't clear the bar either, the product is dead — don't marry an angle that only "sort of" worked twice.
Weekly quota, not results-based: commit to starting a fixed number of new Gate 1 batches every week — three is a sustainable start — regardless of how last week went. The quota, not the hope of a hit, is what makes this a volume game.
Review the log, not the product: every two weeks, read back across every row for patterns — which avatars, pain points, or hook shapes keep clearing the bar even when the specific product died. That pattern outlives any one winner.