Steps

Why one advertising test at a time beats five queued up

A five-deep test queue sounded efficient. It avoided the harder question of what to learn next. The seven-field brief that forces the question first.

What a ten-minute check caught that the ad review didn't

A blind check before launch once caught a word that had come apart into characters that weren't letters, and links that had dropped tracking.

One optional manual build step will eventually get skipped

One command rebuilds the whole asset with no human decision in the middle. The test: a stranger who's never seen it gets byte-identical output.

Collect the market's own words before writing a single line

A quota, a harvest list, and a 15% misfit threshold for building a verbatim buyer-language corpus before drafting any copy for an advertisement.

Most businesses fill 3 cells of their awareness grid, no more

Five awareness stages, times four or five buyer types, makes a grid. Two of five buyer types can go unaddressed for months before anyone notices.

Build the hero as if it's the only screen anyone reads

66% of visitors never got a quarter of the way down one page. Build the first screen, and the proof under it, assuming most never scroll past it.

A short writing sample gives AI a confident, wrong voice

A small writing sample doesn't give a model less of the picture, it gives a confident, wrong one. Collect tens of thousands of words first.

Why can't I tell which ad paid for a sale, with UTMs on

Writing campaign tags onto the checkout session isn't enough; that metadata never reaches the payment record. The one-line fix most businesses miss.

A win carries almost no information. Collect the failures first

Every receipt needs five fields: the number, the denominator and window, the context, what changed, and the era. Four fields and it's an anecdote.

Generate thirty names before building a method around one

A name has to survive being repeated badly by a stranger six months later. Thirty candidates, three filter tests, one trademark search.

What should happen after a buyer, before you sell them more

A buyer who never opens what they bought won't book a call. Three follow-up chains depend on whether they opened it, plus the rule overriding all three.

Design a price ladder where every rung removes effort

A price ladder with four rungs, real dated prices, and a rule: stop at two paid steps past the entry point, or refunds eat the gain.

Revenue dropped. Is it the creative, the algorithm, or what

The instinct is to name a cause fast. Work the chain in order instead: instrument, traffic, page, checkout, order value, delivery. Stop at the link moved.

A disciplined writing process defaults to forgettable order

Nothing self-disrupts. Schedule where a chapter is allowed to break its own shape at the outline stage, or every chapter defaults to the same safe pattern.

A word quota forces a decision a hard deadline doesn't

Draft one chapter at a time to a word count, not an hour count, then freeze it before reading back. Parallel drafts cost nothing if the brief matches.

How do I set a kill threshold for an underperforming ad

A static cost-per-purchase threshold goes stale the moment order value moves. The weekly-recomputed formula, and what changed when the target moved.

A rung selling to nobody is usually duplicated, not mispriced

A $97 product sold to zero buyers next to a $27 product that worked. The price was fine. The job was the same job twice.

A self-score from a sample overstated every line, one direction

An editor scored its own sample at 9.0/9.5/6.5. Six blind readers scoring the whole manuscript came back at 8.5/7.5/4.0. The gap ran one way, every time.

Why do my analytics show sessions that were never a real person

About 30% of raw sessions in one test were never a person: prefetchers, scanners, bots. Filter at read time, never collection, with a four-signal check.

A sentence nobody argues with isn't a point of view, it's a fact

An industry-irritating sentence needs a real opponent to be worth building a book around. Here's how to pressure-test whether yours is one.

Do I need my own event log if I already have a pixel

A platform's pixel grades its own work and gets blocked by browsers. The twelve-event, four-event-to-start log that answers questions a pixel can't.

Generate a dozen offers, score your favourite against a rubric

Score a dozen candidate offers blind against a fifteen-dimension rubric dated before you looked. Your favourite scoring a six is the finding.

Does rewording a sales page actually sell more books

A clearer landing page sold fewer books in one real test. The maths behind why, and the sample size a page test actually needs.

Why does the platform say I made money when my bank says I didn't

An ad set scored 0.06 return on platform reporting. The bank's ledger showed 1.55, A$263.94 across 7 settled sales on A$120.70 spend, before a wrong kill.

A 10.7% click-through rate and zero sales: why clicks lied

15,368 impressions, 1,649 clicks, A$564 spent, zero purchases. The click rate was roughly four times normal and meant nothing without a purchase count.

A retail listing can breach an exclusivity clause you forgot

Paid, free, direct outreach and retail each cost differently and need reading differently. Retail also needs its own exclusivity check before it goes live.

Measure the voice with numbers, then test it by covering the name

'Direct but warm' isn't a usable voice spec because nobody can check it. 'Median sentence length seventeen words' is, and here's the test that confirms it.

How much should I risk testing an idea before I kill it

A partner-built product demoed well and topped out at about 50% accuracy in production. It was killed on a number written down before anyone cared.

What's still running after 90 days is the signal, not today's

Every ad platform publishes what's currently live, and almost nobody checks. One $4.99 product ran live for 482 days; that longevity is the real signal.

Target the week they're having, not the job title on their badge

Job-title targeting finds people who resemble your customer. It doesn't find customers. Here's the date rule that tells the difference.

The old cover stayed live for weeks and nobody did anything wrong

A cover lives in more places than anyone would guess before counting. One canonical file, one named owner, walks the register the day anything changes.

When is a one-click upsell unauthorised, not convenient

A one-click charge is fine for a small amount and unauthorised at a few hundred dollars. The consent threshold that decides which checkout a buyer sees.

How much should an order bump cost, and who should set the price

Published order-bump take rates run 10% to 60% with no dataset behind either number. The pricing rule that matters more: who owns the price.

If a chapter can't fill one grid row, it isn't doing a job yet

Build the outline as a grid before any prose exists. A chapter drifted from its own promise gets caught by re-summarising it, not by rereading it.

Run the Conversions API with the pixel, or instead of it

Run server-side events alongside the pixel, never instead of it. The dedup window and Event Match Quality target that decide if sales double-count.

0.07 inches of spine misalignment cost an entire print run

A spine cut even slightly off centre reads as cheap before a buyer opens the book. It depends on the exact printer and stock, recalculated every change.

My average order sat below break-even. What actually fixed it

A $16.57 average order sat against a $20-40 floor for months. Three repricings later it held near A$31, and the fix wasn't traffic, it was the price.

You'll never finish an asset you can't lose on paper

Write an abandonment scene for three readers before drafting: the one you want, the one you'll get most of, and the one you don't want at all.

Three funnel additions that looked obviously right and weren't

An exit-intent offer, a fancier statistics engine, and an always-on retargeting layer were all satisfying to build. One made the funnel worse.

Done is when the measures clear, not after a set number of passes

A three-pass cap would have stopped a real revision run on its worst score, with no fourth round left. Here's the stop rule that replaced the pass count.

A script found in a second what a careful reader missed once

53 leaked internal tags across 24 lines, found by a script, after one slipped past a careful reader. Six gates, each failing the build, not just warning.

Why save a customer's card at the first sale

Order bumps and one-click upsells both depend on a saved card from sale one. Skip it and every later offer has to ask for payment details again.

An adjustable floor isn't a floor, it's a preference

Five scored lines, anchored against two real named books, floor written down before anyone reads a page. The exact rubric and why enjoyment gates lower.

Should a paid-only funnel have a free version too

A free front door ran alongside a paid one for months and produced conversations the paid door never did. What a second, free door costs to run.

Set the day-one break-even floor before you blame traffic

A funnel can run for months on the wrong fix. One real failure: two months spent fixing traffic when the problem was a $16.57 order against a $20-40 floor.

Write your never-say list before a deadline tempts the claim

A real never-say list has sixteen banned claims, settled before a word of drafting, each one tested on substantiation rather than comfort.

Do I have a spike problem or a floor problem

A spike problem wants one very large month. A floor problem wants to never have a small one. A spike machine for a floor problem stays anxious.

Is ad platform reporting always wrong in the same direction

Three distortions understate what a campaign did. One, broad targeting stealing credit, overstates it. A 663-experiment study found a 62-115 point error.

Should a download link ever expire

A permanent download link gets forwarded until a paid product is free with extra steps. The token-expiry window that stops that without annoying a buyer.

Does it matter if a customer opens what they bought

A buyer who never opens what they bought won't book a call. Completion on a low-priced book runs an estimated 20-30%, measured, never borrowed.

Should one ad campaign do all the testing and all the scaling

A campaign asked to test and scale at once does the second job badly while looking busy at the first. Splitting into two, judged separately, is the fix.

Work out what a customer, and a book buyer, is really worth

Most people price off the first invoice. Revenue per buyer uses lifetime value times close rate instead, and the close rate most people never measure.

Find the sentence you've learned not to say out loud

Most experts have one true sentence they'd never put on their website. Naming it, and finding who'd argue against it, is what gives an asset a spine.