The archive
Articles
135 articles, free to read. Filter by topic or search for a word in a title.
What an online business actually is
Somebody gives you money and you give them something back. Everything else is detail, and most of the detail is a way to feel productive without finding out whether anyone will pay you.
How to choose something you can sell
Starting from passion gives you something nobody needs. Starting from market size gives you a worse version of what three funded companies already sell. There is a third option.
How to find out whether anyone wants it
Asking people whether they like your idea does not answer the question. People are kind. Here is what to look at instead, and why competitors are the best news you will get.
Customer conversations that tell you the truth
The moment you describe what you are building, the conversation becomes a performance. Ask about the last time the problem happened instead, and everything you hear is a fact.
How to price your first offer
Almost everyone prices too low, often by a factor of three or four, and the reasons are psychological rather than commercial. A useful first price makes you slightly uncomfortable to say.
How to write an offer people understand
An offer that exists only in conversation changes shape every time you describe it. One page, six parts, and the paragraph about what you are not doing is the valuable one.
What your first page actually needs
You need a page, not a site. Six elements, one action, and an honest answer to the problem of having no testimonials yet because you have no customers yet.
Where your first ten visitors come from
Not search, not ads, not an audience. Ten specific people who reply beats a thousand strangers who bounce, because a thousand strangers tell you nothing about why.
What an objection actually means
Four objections cover almost everything you will hear, and each one usually means something other than what it says. The useful response is a question, not an argument.
Reading the result and deciding what comes next
Three outcomes, each with a clear next move. The expensive mistake is not failing. It is carrying on without noticing which outcome you are in.
How search actually works
Crawl, index, rank. Three jobs in that order, and a surprising number of SEO problems are one of the first two failing while you optimise for the third.
How to find keywords you can actually win
Volume is not value. Three questions decide whether a phrase is worth writing for, and the one everybody skips is whether you can plausibly rank at all.
Search intent, and why it decides everything
You can write the best page on the internet and still not rank, because the page is the wrong kind of thing. The results already tell you which kind wins.
How to write a page that ranks
One page, one job. Answer the question near the top, structure the rest around the follow-ups, and stop counting words. Titles and descriptions included.
Internal links are the cheapest SEO you have
Entirely within your control, free, and effective as soon as the page is recrawled. Also the part of SEO most people never deliberately touch.
What to do when rankings drop
Diagnose before you change anything, or you will have two problems. Four questions in order, from cheapest to check to most expensive.
When search ads are worth it
Work backwards from what a customer is worth, not forwards from what a click costs. If that sum does not work, nothing else in the account will save it.
How the auction actually works
Highest bid wins is wrong in a way that costs money. Relevance and the landing page are the cheap levers, and the bid is the expensive one.
An account structure that stays manageable
One rule: group keywords so that one advert and one page could serve all of them. Match types included, because choosing them is structural.
Negative keywords are where the money leaks
You do not buy keywords, you buy searches that resemble them. The least interesting hour in the account is usually the most profitable one.
The ad and the landing page are one job
Somebody typed something. Everything they see next should visibly be about that thing. Where the ad and the page disagree, you pay for the gap.
Reading the data, and killing what does not work
Until a sale can be traced back to a search, every other number is decoration. Then set the cost ceiling before you are attached to the keyword.
Why Meta is not search
On search somebody tells you what they want. Here nobody asked, and you are creating demand in the three seconds before a thumb moves. Everything follows from that.
The offer comes first
Most campaigns fail before anything is built, because the thing on offer was not interesting enough to interrupt anyone for. The usual fix is a smaller ask.
Creative: finding the angle, then making it
Aspect ratios and hooks are the easy part. The expensive part is the argument the advert makes, and five executions of one argument all die together.
Audiences, and how little they matter now
Stacked interests and lookalikes were a proxy for information the system now has directly. Splitting budget across audiences keeps all of them guessing.
A testing framework, and creative fatigue
Most people do not test, they fiddle. Change one thing, in order of effect, and write the decision rule down before you are fond of the answer.
Scaling, and knowing when to stop
Performance at a small budget is not a rate you can multiply. It is the cheapest slice of the audience. Decide the stopping rule before you are attached.
What sells as a digital product
Zero marginal cost cuts both ways. Nobody pays for information, so it is worth knowing which of the four things you are actually selling.
Validate it before you build it
People will say an idea sounds useful and then not buy it. The only signal that settles the question involves somebody parting with money first.
Scoping version one, and building it
Once somebody has paid, the risk stops being whether anyone wants it and becomes whether you ever finish. Version one should embarrass you slightly.
Pricing digital goods
There is no cost to anchor to, so any price is defensible and none is obvious. Price is read as information before it is read as cost.
The sales page, checkout and delivery
Two things stand between interest and money. The first gets all the attention and the second quietly costs more sales.
An audience, and the launch sequence
A launch is not a marketing event, it is the day you find out whether you built an audience. Most disappointing ones were decided weeks earlier.
After launch: support, refunds, piracy and version two
Where the phrase passive income stops being accurate. Refund generously, ignore most piracy, and build version two from what buyers struggled with.
What these tools are actually good at
Very good at turning one kind of text into another, useful for a first draft, and unreliable about any fact you did not supply. The failures cluster somewhere specific.
Writing with AI without producing slop
Google does not penalise AI writing; it penalises pages nobody needed. The reason most of it reads badly is simpler than the tooling, and so is the fix.
Using AI to find things out
It is good at telling you what the question is and bad at answering it. A reference you cannot open does not exist, and primary sources are easier to reach than people think.
AI in customer messages, carefully
Sort the queue before automating any of it. Drafting is not sending, and the policy you have not written down is the one the tool will invent for you.
What is worth automating, with the arithmetic
Hours saved are often hours moved. Frequency decides it, not difficulty, and instinct points at the rare annoying jobs that lose on the numbers.
How to actually check what it gives you
Reading it again tests whether it is coherent, which it was always going to be. Mark every checkable claim, then verify each one outside the tool.
Choosing between the tools
The least important question, asked first. What matters is where your data goes, whether it cites what it looked at, and a ten-task test set you keep.
When not to use it at all
When checking costs more than doing, when the writing was the thinking, when somebody is owed a person, and when you cannot judge the answer.
What survives the arithmetic
Plenty of things sell briskly and lose money on every order. Six physical properties set the ceiling before the product is even chosen, and advertising cost decides the rest.
Choosing a platform without regretting it
Compare total cost at five times your current volume, not at today's. Then ask the leaving question, because a platform you cannot export from is a tenancy.
The margin after everything
Gross margin makes shops look viable. Contribution per order decides whether they are, and a fixed processing fee quietly takes a fifth of the margin on cheap items.
What card payments actually cost
One small percentage is quoted and several arrive. The rate depends on the card your customer happens to hold, so work out your blended rate from the statement.
Sales tax when you sell across borders
Tax is charged where the customer is. The EU gives many countries for one registration; the United States gives one country needing up to fifty.
Returns, and the rule you cannot opt out of
Three genuinely different legal structures. A mandatory fourteen-day return in Berlin, no obligation in Brisbane, and none federally in Boston.
The seven questions a product page must answer
In order, and most pages fail at three, four and six. Write the description for somebody who cannot pick the thing up, which is the whole discipline.
Photographing your own products properly
A window, a sheet of white card and the ceiling light off will beat equipment. Consistency across the catalogue matters more than any single photograph.
Shipping without giving away the margin
Carriers charge on the size of the box as well as the weight. Three pricing models, who each suits, and why free shipping is a price rise with better manners.
Why decided buyers leave at checkout
A basket is somebody who already decided. Delivery cost appearing late and forced registration do most of the damage, and both are a page edit.
A shop that is quick on a phone
The three published thresholds, why the slowest quarter of visitors decides your score, and why layout shift is the one that costs orders directly.
Marketplace or your own shop
An arithmetic question with a sequencing answer: can you acquire a customer for less than the commission? For most people in year one, not yet.
A blog that earns without you
Passive does not mean no work. It means the work and the payment come apart, and almost all of the work sits at the front, in the part that pays nothing.
Choosing a subject for a review site
Five conditions, two of them close to mandatory. Then the per-visitor arithmetic, which is why enthusiasm for a cheap category is an expensive preference.
What SEO actually means
Three jobs, not one, and most sites fail at a different one than they think. What changed recently, and the one thing that did not.
What AI changed about earning online
Production got cheap. Demand, distribution and trust did not move at all, and cheap production made attention scarcer rather than more plentiful.
Review sites against dropshipping
One buys attention, the other earns it. Which is better depends on whether you have money or patience, and most people have one and not the other.
The quarter that decides the year
Two things happen at once and get discussed as one: people buy more, and advertising gets dearer. For a shop that is a cost; for a publisher it is income.
What a brand actually is
What somebody expects before you have said anything. Not the logo, not the colours, not the mission statement, and all four of those are the cheap part.
Positioning, in one sentence
Four slots, and the one everybody skips is what you are instead of. Put a competitor's name at the front: if it is still true, you wrote a category description.
Naming a business, and the checks that matter
A trade mark does not protect a word. It protects the word for the goods and services you listed, which is the part people get wrong and the part with consequences.
A visual identity without a designer
Six decisions made once. Consistency beats quality here, because a reader cannot assess your taste and detects carelessness instantly.
A brand voice you can actually use
Nobody ever wrote a better sentence because a document told them to be authentic. Pairs of sentences, not adjectives, and what you do with bad news.
Trust signals that do something
The ones that work are the ones a liar would find inconvenient to fake. What to do when you have no reviews yet, and why inventing them is unlawful in the EU.
Brand or performance, when money is short
Not a philosophical argument. Performance to find out whether it sells, then brand to stop acquisition getting more expensive every year.
Staying consistent without a rulebook
Three things have to match and the rest can flex. Most visible inconsistency is not tone, it is a dormant profile with a two-year-old price list.
Rebranding without breaking things
Write down what will be different for a customer afterwards. If the answer is that it will look different, the problem is somewhere else and this is expensive.
Measuring brand without inventing numbers
What genuinely cannot be measured at this size, the six things that can be observed, and the one question worth asking customers instead.
Email permission, by jurisdiction
Four regimes that differ in kind, not detail. One needs no consent at all, one makes you prove it. Build to the strictest and you are compliant everywhere.
Getting delivered at all
SPF, DKIM and DMARC, and what Gmail and Yahoo now require of bulk senders. A spam rate of 0.1 per cent is three complaints in three thousand.
The first hundred subscribers
Nobody subscribes to a box. They subscribe to one specific thing, described exactly and delivered immediately, and the first hundred come from places that do not scale.
The welcome sequence
The highest attention you will ever have, usually spent on a message saying thanks for subscribing. Four messages, each doing a different job.
What to send on a Tuesday
Pick a repeatable format once rather than deciding every time. Four that work, one that never does, and the forward test before you send.
Subject lines without the tricks
In the United States a subject line that misrepresents the contents is unlawful, not merely cheap. And open rate stopped measuring what you think it measures.
Segmenting a small list
Three or four groups kept by hand. The split that matters most is customers against everybody else, and permission itself can differ between segments.
Automated sequences that do not misfire
Four that reliably earn, and the rules that stop a machine sending slightly wrong things forever. Starting with: stop when the thing happens.
Measuring email now that opens are broken
Privacy features pre-load images, so open rate is inflated by an unknown amount. Three numbers still mean something, and one of them can end your sending.
Removing people from your list
It feels like going backwards and is the most effective deliverability work available. A sunset policy, and why an unsubscribe is cheaper than a complaint.
The five numbers worth watching
One test before anything goes on a report: what would I do differently if this halved. Most of the usual metrics fail it, including bounce rate and time on page.
Analytics and the consent problem
Consent-first and opt-out jurisdictions need different behaviour. In the EEA the exemption is narrow and the regulator declines to settle it.
Setting it up so it answers something
Define the outcome before installing anything. A thank-you page is not a conversion, and your own records are the only source that is actually true.
Reading the traffic sources honestly
Direct is not brand recognition, it is everything the tool could not identify. And the useful view is landing pages sorted by outcome, not by visits.
Conversion rate, computed properly
Divided by the wrong denominator and compared with somebody else's. Per path, against your own history, and with the two-conversion test applied.
The free tool most sites ignore
It needs no consent banner and answers what analytics cannot: what people searched, what you appear for but nobody clicks, and the queries you nearly rank for.
Why none of the numbers agree
They count different things over different windows and see different people. The disagreement cannot be fixed, so pick one source as the record and measure the offset.
Why your site probably cannot be tested
Detecting a small difference takes thousands of visitors per variation. What to do instead, and the one question that stops you acting on noise.
A report you will actually read
Three horizons kept separate, five lines in the weekly one, and the action written beside each number so the report is a list of conditions rather than figures.
From a number to a decision
A figure tells you where to look, never what to do. Narrow it to a page, a source and a device, then write down what you expect before changing anything.
Who is actually worth your time
Four conditions you discover rather than argue. Most conversations that go nowhere fail on cost or timing, and a fast no is the second-best outcome available.
The first conversation
Twenty minutes of asking, not five. Ask about the last time it went wrong, and the question almost nobody asks: did you look for a solution then, and what stopped you.
A proposal somebody can approve
Their problem in their words comes first. If that paragraph is right the rest is a formality, and no number in it should appear for the first time.
Saying the number without flinching
Say it early, then stop talking. When they say expensive, find out which of three things they mean, and reduce scope rather than price.
Objections, and which to ignore
The stated one is often not the real one. One question finds it, and four objections should not be overcome at all.
Following up without pestering
Most sales are lost to silence, not to a decision. Four messages over a month, then stop, and the fourth is the one that gets answered.
A pipeline that fits on one page
Five columns, no probability percentages, fifteen minutes a week. Where conversations pile up is a diagnosis of a different stage.
Selling when you dislike selling
What is being avoided is usually a performance, not the activity. Only one part is unavoidable, and the rest can move into writing.
Being referable, then asking
Somebody recommending you is answering a question for a third party. Two of the three conditions are positioning rather than quality of work.
What the losses tell you
Ask what made the difference, once, after it is over. If most losses are to doing nothing, the work belongs at the start of the conversation.
What a small site is actually for
Decide the one action before anything is built. Six kinds of page is a complete site, and nobody arrives at your home page.
The page that has to produce an action
Seven questions in order. Most pages that underperform are missing two of them rather than being badly designed.
Mobile first, meant literally
Narrow is the harder constraint, so solve it first. Then test on a real phone on mobile data, which is the one experience nobody else has.
The speed that actually costs you
Three published thresholds measured at the 75th percentile over 28 days, and the one of them that loses you taps rather than patience.
Accessibility is mostly legibility
Six things, most of them an afternoon. Then put the mouse away for two minutes and try to finish your own form.
Navigation, labels and links
Use the plain noun. Keep it to six items. And accept that the links inside your content get the clicks the menu does not.
Forms people finish
Every field is a reason to leave. Delete the ones whose answers you never use, fix the input types, and say what happens next.
Writing for people who are not reading
The answer first, then the explanation. Cut the first sentence of every section, and remove any line a competitor could use unchanged.
When to redesign, and when not to
Write down what a customer will be able to do afterwards. If the honest answer is that it will look different, do the five cheaper things.
Running a site you can maintain
One page listing what it is made of, a backup you have restored from, and thirty minutes four times a year.
Finding the hours worth buying back
Log a week as you work. The job that eats it is the four-minute one done eleven times a day, and the best automation is often deletion.
Is it worth automating at all
Three numbers, one of which everybody omits. Do it by hand ten times first, because that is where the exceptions are.
Connecting the tools you already have
Trigger, filter, transform, action, failure. Filter early, because these platforms charge per run and the bill arrives after a busy month.
When automation breaks quietly
Only one of the five failure modes announces itself. The alert people forget is the one for nothing having happened at all.
The automation that needs no software
Most repeated work is decisional, not mechanical, and a template removes the deciding. Write it the next time you do the job.
Putting a model inside a workflow
The useful question is which of its failures you can afford. Put it where a wrong answer is visible, and design the position rather than the prompt.
Automated messages to customers
Five worth sending, all of them transactional. Then the consent regimes, and the exit condition people forget to test.
What not to automate
The value in a small business is frequently in the departures from the rule. And an approval step nobody reads is not a check.
What automation actually costs
Subscriptions that scale with volume, tier upgrades for one feature, and permanent maintenance. List the charges from your statement, not your memory.
Doing this in the right order
Delete, write down, use what you already pay for, then build one thing. One at a time, a month apart, because the rate is set by maintenance.
Do you need social media at all
Four jobs it can do, and two of them need a maintained profile rather than a programme. Write down how your last twenty customers found you.
Choosing one platform and meaning it
Four thin accounts is a presence on none. Pick by where the conversation is and which format you will actually sustain.
Posting without it eating the week
Capture during the work, publish on a schedule. The bottleneck is never ideas, it is not having an image.
What to measure, and what to ignore
Followers and likes are not outcomes. Saves, shares, profile clicks and enquiries are, and the platform's numbers will never match yours.
Disclosure, endorsements and reviews
If a reasonable person would not know a post is commercial, it has to say so. The rules apply to one person with a phone exactly as to a brand.
Handling comments and complaints
You are writing for the fifty people reading, not the complainant. And deleting it is almost always worse than answering it.
Paid social on a small budget
Warm audiences first, one offer, and a stop condition written before the first currency unit. The boost button is not advertising.
Moving an audience to something you own
Followers are rented. Offer one specific useful thing for an address, and remember that following you is not consent to email.
What to do about video
The three useful formats need no appearing on camera. Vertical, audible, steady, captioned, and film five times what you post.
A routine you can keep for years
Capture daily, batch fortnightly, review monthly, decide twice a year. About two hours a month, and still running in two years.
Stock, for a shop with one person in it
Weeks of cover against lead time is the whole method. Then the thirty-day delivery default, and a six-month rule for everything that has not moved.
Answering customers without losing the week
Three of the five messages you get are a documentation problem. Also: the dead dispute-platform link still sitting in your footer.
Subscriptions, and what most shops actually want
Repeat purchase needs no billing relationship to defend. If you do run a subscription, the arithmetic is churn and the rules are moving.
Selling through a social platform's checkout
Three arrangements, and the third costs more than the fee. What the terms let you do with the buyer, and the details your listing has to carry.