The marksman plan
Decision
Keep the partner story. Test whether a more believable buying argument improves purchases, not whether we can make the story more upsetting.
Core desire: be her partner, not another person she has to get out of bed. Observable payoff: her alarm returns to the time she actually needs to wake. Strongest audience insight: she cannot switch off responsibility just because he says he will handle it.
Status: strategy + illustrative sample for Ayden's audit. Not approved ad copy, not a full-native-SOP PASS, not a campaign launch. Five image CONCEPTS are specified; no images generated or purchased.
Evidence boundaries
- User reports the original as Dawn's best native. No new performance export was pulled for this task.
- User-supplied analysis describes 486 unique comments/replies in a Sept 7 export, with comments spanning March 9–May 5 and all marked hidden. This is supplied archived evidence, not a independently rechecked live comment census. Likes indicate engagement, not purchase causation. No objection-frequency rankings can be calculated from the excerpt.
- Separate handwritten notes were not included. Their incorporation remains open; this plan uses the supplied commentary and the preceding editorial audit.
- Original story is a supplied creative control, not a verified customer history. Sample dialogue and chronology are visibly labeled illustrative. No new achieved Dawn results, clinician, scientific discovery or customer testimonial is invented.
- Older competitive research flags the generic partner lane as crowded. That is archival context, not a current saturation finding. This test deliberately retains the user-reported winner's territory and differentiates through responsibility, restrained payoff and buying credibility.
What to preserve
1. Early alarm belongs to the girlfriend, not the person who needs waking.
2. He answers without remembering. She hears apparent agreement and has to check again.
3. The quiet Sunday confrontation, rather than a bigger invented catastrophe.
4. He tries to release her from the job; she resumes it by Wednesday.
5. Her alarm time as the emotional resolution; voluntary touch instead of another waking attempt.
6. Partner POV. Parent purchase intent is a reason to keep the story transferable, not to insert a teen into this test.
Repair map
| Finding | Action in native | Other work required |
|---|
| Shaking failed, so why wrist vibration? | Ask the question plainly. Explain escalating, wearer-controlled alarm behavior. Do not claim it is proven superior to human shaking. | Exact-product observation comparing cue detection, dismissal and staying out of bed. Do not manufacture a distinction from feature copy. |
| Fitbit already failed | Do not promise immunity to habituation or universal waking. Escalation is a design feature, not a guaranteed rescue after another device failed. | Verify model-specific comparative motor claim. A motor-strength ratio is not a waking-success ratio. |
| I turn it off and sleep again | Define success as getting and staying out of bed without partner help. | Demonstrate real dismissal behavior. Do not imply sit-up detection, forced standing or automatic prevention of returning to sleep. |
| Why not my watch? | Focus on a dedicated, app-free alarm at the verified price. Do not attack working alternatives. | PDP comparison: who benefits from a dedicated band; who already has an adequate solution. |
| Selfishness / learned incompetence | Narrator owns setup, charging and follow-through. Acknowledge partner cost without diagnosing every sleeper or making her cruel. | No debate campaign aimed at commenters who reject the premise. |
| Fake-story seams | Remove false hour totals, no-screen/no-button claims and unexplained room entrances. Keep time motif without calculating stolen lifespan. | Verify testimonial provenance before presenting any account as real. |
| Product failures / partner still disturbed | No absolute promise that only the wearer wakes. | Separate support cases from trolling, verify SKU/order with consent, collect honest outcomes. Never hide legitimate criticism just to improve apparent sentiment. |
| Refund distrust | No risk-free/no-questions-asked trial language in the sample. | Resolve the current offer-policy conflict before paid launch. |
| Setup / wrist comfort / ankle fit | Keep setup to the supported fact: set on band, no app. | Exact-SKU short setup video and clear sizing; no ankle-fit promise without verification. |
| Other sleep conditions | No universal ADHD mechanism or claim the band treats the cause. | Helpful FAQ directing persistent severe waking difficulty toward clinical assessment. |
| Length complaints | Improve continuity and paragraph flow first. | Length is a later isolated test, not the first hypothesis. |
Live product and policy check
Verified through live Shopify product JSON: Black, Blue, Pink and Orange variants each list $39.99. Live product description supports escalating wrist vibration and no app, but also contains claims that are not independently substantiated by that listing.
Live refund policy: customers may request a return within 60 days after placing the order; item must be undamaged and in the same condition received, in original packaging, with proof of purchase. Customers pay return shipping. Refund follows inspection and approval.
The storefront says 60 nights risk-free. Those are not equivalent terms. Do not fix this by burying qualifications inside the ad. Obtain an explicit operational policy decision and separately authorized storefront repair. No Shopify changes made here.
Current internal product card describes an LED time readout and a haptic button. Thus the original's 'No screen. No buttons.' cannot be retained. Button count also differs between internal card and live description; omit count pending exact-SKU verification.
Sources: https://dawnbands.com/products/wake-up-band-adhd.json ; https://dawnbands.com/policies/refund-policy ; https://dawnbands.com/products/wake-up-band-adhd
The five-image / multi-text unit
One revised body. Three interchangeable opening blocks. Three headlines. Five distinct single-image ads, each carrying the SAME three complete primary-text options and the SAME three titles. Each primary-text option is its hook block plus the identical body continuation. No length variants in this first unit.
| Ad | Image concept | Execution and hypothesis |
|---|
| M1 | Existing winning image anchor | Retrieve the exact partner-ad image, verify its hash and usage authorization. No recreation. Connects the rewrite to the original visual. Asset not supplied in this task. |
| M2 | Her early alarm | Genuine phone photo of an alarm list with the relevant early alarm. Consented own capture preferred. No fake social UI, fabricated customer screenshot or text overlay. Tests immediate problem recognition. |
| M3 | Waking attempt | Ordinary consented POV photo of a hand on a covered adult's shoulder. Do not imply it is the narrator or proof Dawn worked. Tests the responsibility burden; monitor whether it amplifies the tactile objection. |
| M4 | Waiting awake | Ordinary bedroom/nightstand scene taken by the awake partner, with the other side still occupied only if consented. No staged crying, stock-model expression or added labels. Tests the cost to the person who stays alert. |
| M5 | Voluntary touch | Consented candid-style hand on a partner's back during an ordinary household task, no coffee, no product. Illustrative relationship desire, not an achieved customer result. Tests warmth against the pain-led images. |
These are proposed scenes, not found assets. Use owned/licensed photos or permissioned references. Publicly visible does not mean reusable. Review actual candidates before choosing; if generation is authorized later, reference-based generation only, with no testimonial implication.
Naming: PA-PARTNER-M1 through PA-PARTNER-M5. One explicit image hash per ad. No carousel, no image swapping, no image mutation/expansion/text overlays or auto-generated text rewrites. Verify actual enabled creative features through the publishing contract before launch.
Hook blocks
Replace the first three lines of the sample with exactly one block. All feed into 'She shakes my shoulder. Says my name.'
H1 — original time contradiction / Loop
My girlfriend sets her alarm for 5:50am.
She doesn't need to be up until 7:15.
She sets it because I sleep through mine.
H2 — unequal waking burden / Loop
My girlfriend gets up early to wake me.
I sleep through my alarms, so she sets one of her own.
She doesn't need to be up yet.
H3 — responsibility relapse / Loop
I asked my girlfriend to stop waking me up.
By Wednesday she was setting her early alarm again.
I'd slept through mine.
Hooks 1 and 3 are illustrative narrator history derived from the supplied control, not verified customer facts. H3 intentionally previews the Wednesday beat; if it feels repetitive in a read-through, use H1/H2 until a distinct third opening is approved rather than silently changing the shared body.
Headline options
T1: When your partner is your alarm
T2: A wrist alarm without an app
T3: For mornings you want to handle yourself
Headlines carry relationship recognition, concrete product differentiation and desired ownership. No guaranteed outcomes. Same titles across all five ads so each image has the same text opportunity.
Test sequence: what it can and cannot tell us
1. Approval and proof gates first. Resolve policy congruence and the sample's illustrative status. Do not relaunch the erroneous original as a fresh control just because it won historically. Historical performance is a benchmark, not randomized evidence.
2. If a truthful, approved control version is available, compare it with the revision using the same verified original image, one fixed H1 opening and one fixed T1 title. Match audience eligibility, placements, destination, offer, attribution and start period. Use Meta Experiments random split if available and explicitly authorized. This measures the revised copy package, not individual paragraph causality. A normal CBO is not a randomized copy test.
3. After choosing the body, run the five-image unit above with identical multi-text pools. This finds a commercially useful image/text package. Image-level CPA is interpretable as a delivered package; Meta does not serve each hook/headline combination evenly. Do not claim that an image won independently of text selection.
4. If hook attribution matters after a package wins, hold its image and title fixed and compare three single-text ads. If headline attribution matters, hold image and hook fixed and compare titles. Do not run every axis simultaneously and call it marksman learning.
5. Keep a parent/teen adaptation, length reduction and authority-heavy mechanism rewrite for later units. They change the buyer, narrative architecture or persuasion question.
Measurement and stopping rules (proposals, not approved spend)
Set allowable CPA from current contribution margin first: net order revenue minus product/fulfillment cost, fees, expected refunds, shipping subsidy and required contribution. Neither that margin nor the current control CPA is supplied here, so no dollar budget is invented.
Primary decision: purchases and CPA against the predefined allowable CPA. Track net economics after cancellations/returns mature. Secondary diagnostics: outbound CTR, landing-page views per outbound click, purchase conversion from landing-page views, CPM, and objection categories in delivered comments. CTR without purchase improvement is not a win.
Suggested guardrails to approve before launch: review a zero-purchase cell at 2x allowable CPA; hard spending review at 3x. Do not call a low-spend cell with one purchase a winner. After at least five purchases per materially delivered cell, make a provisional commercial decision, not a statistical-significance claim. Assess again after the chosen attribution lag; do not keep extending failing cells because of a high CTR. These are operating heuristics, not sample-size calculations.
If Meta starves a cell, mark it undelivered/inconclusive rather than a loser. Purchase attribution and refund economics take priority over like counts. Compare commenter objections only with appropriate exposure context; the archived all-hidden sample is not a denominator for a causal before/after sentiment claim.
Sample acceptance boundaries
The sample intentionally preserves the supplied winner's compact architecture. It is not relabeled as the full authority-dialogue, six-week-results native SOP. No fictional clinician, exclusive efficacy claim, unverified company history, guarantee promise or achieved product-results timeline has been added to fill that rubric.
The most important remaining creative gap: there is still no exact-product evidence establishing why Dawn succeeds after physical shaking or a tracker failed. The sample offers a concrete reason to try, not proof that those failures are overcome. Stronger copy cannot manufacture that missing fact.
The prospective ending is honest but less emotionally conclusive than the original kitchen scene. Review that trade-off explicitly. A verified same-customer outcome could restore a past-tense relationship payoff; until then, publish no fabricated success story.
Mechanical native pipeline: NOT RUN because no approved structured long-SOP brief or batch validator exists for this scoped control-led sample. Manual copy/claim/continuity check is separate; it is not an original-SOP PASS. Independent editorial review: pending/not claimed. Human audit: requested.
Action order
- Now: Ayden audits this sample's middle, the ownership framing, and the prospective close.
- Before creative production: incorporate missing handwritten notes; approve the copy route; retrieve original image; collect four permissioned image candidates.
- Before launch: resolve return-policy promise, exact SKU operation and testimonial status; verify destination congruence, tracking and contribution margin; approve the actual budget and Meta changes.
- Then: fixed-asset copy comparison where possible, followed by the five-image multi-text unit. No campaign or policy changes were performed for this review.