The method

How a high-volume ads team tests creative, and what a ₹5 lakh-a-month brand can copy.

HookVerdict encodes a testing discipline drawn from public accounts of Wispr Flow's paid ads programme and Meta's own documentation. Here is what we took, what we left out, and why.

What the Wispr Flow programme showed

In October 2026 Hector Serrano, marketing hire #2 at Wispr Flow, posted a video on how he built the company's paid ads programme and how he would rebuild it in 30 days (post on X). We couldn't transcribe the video, so we rely on a written summary by Ira Bodnar (thread on X) and on RocketShip HQ's analysis of the company's ads (rocketshiphq.com).

Volume: the summary describes a target of 500–700 new creator ads a month. RocketShip counted about 50 new ads a week, with a median ad lifespan of 16 days.
Diversity: each creator was assigned a gap, such as an under-served customer type, a format or a country, and told not to copy winning ads.
Iteration: RocketShip's critique was that winners were rarely built on. It recommended keeping a winning ad's shell and changing one variable at a time.
Rhythm: a weekly review that produces a scorecard and a backlog of tests.

One caution: Wispr Flow is venture-funded and may be optimising for growth rather than return on ad spend, as Laura Roeder pointed out (post on X). A bootstrapped D2C brand needs the same discipline with a profit constraint, which is why HookVerdict judges every test against your breakeven.

What changes at ₹1–20 lakh a month

Hundreds of ads a month only works when each one can get enough spend to show a result. At smaller budgets, fewer but more different concepts beat many small variations. HookVerdict sizes your weekly test count from your budget and target CPA, so every test can reach a verdict.

In India the gap is often language and city tier rather than country. Most Indian internet users read content in Indian languages (IAMAI-Kantar via Business Today), yet most D2C creative is still written in English.

The rules HookVerdict enforces

All thresholds are defaults you can change per brand.

Every variant carries a hypothesis.: A test without a hypothesis can only tell you which ad won, not why. With one, a loss is still a learning.
New concepts must differ in angle, persona, format or language.: Meta's Andromeda retrieval system picks from the ads you give it. Near-identical ads compete with each other instead of reaching new people.
Iterate winners by changing one variable.: Change the hook, the persona, the offer, the format, the language or the layout, never several at once. Then you know what moved the number.
No verdict before 2× target CPA and 3 days.: Small samples swing wildly. Meta's own learning phase needs about 50 optimisation events in a week to settle.
Kill zero-result ads at 2× target CPA.: Waiting longer rarely rescues an ad that has spent twice what a customer is worth with nothing to show.
Call winners on probability, not on one good day.: A winner needs cost per result at or below target and at least an 80% probability that its true CPA is below target.
Judge on contribution, not platform ROAS.: COD refusals, returns and GST can turn a reported 3× into a loss. Your breakeven sets the bar.
Scale in capped steps.: Budget rises by at most your daily limit (20% by default), inside a monthly cap you set.
A person approves anything that goes live.: Automation can pause and adjust within limits if you allow it. Launching a new ad always needs a human.

What we don't claim

You'll see specific “similarity percentages” for Meta's Andromeda system quoted online. Meta hasn't published them, so we don't use them. Andromeda itself is documented by Meta (Meta Engineering), and diversity of concepts is the practical takeaway.

Try the rules on your own numbers with the breakeven ROAS calculator and the creative fatigue checker.