Clicks to Customers

One idea carries this course. It is worth stating plainly now, because if you do not accept it the rest will read as pedantry.

Every marketing claim is a claim about what would otherwise have happened. "The campaign generated 340 leads" is not a finding. It is a count of leads that arrived while the campaign was running and were tagged by a system the platform designed. The finding would be that 340 leads arrived and some number of them would not have arrived otherwise. Nothing in an advertising dashboard, an analytics report or an attribution model computes that number, and the difference between the two is not small: when it has been measured properly, it has been off by a factor of three, or five, or two hundred.

The mechanism is easy to state and hard to feel. Advertising is delivered to people the system already predicts will act. Those people would have acted at a higher rate anyway. Then, separately, people who are doing one thing online are more likely to be doing everything else online at the same moment, so anyone who saw your ad was already having a busy afternoon. Both effects push the measured number the same way, which is up, and neither leaves a trace in the report.

Three things follow, and they shape everything in the course.

One: attributed is not incremental, and no amount of modelling closes the gap. Multi-touch attribution is not a better version of last-click. It is the same exposed-user log with the credit divided differently, and it inherits every problem the log has. The only things that measure a counterfactual are a group that did not get the thing and a comparison against them.

Two: the discipline is to name the comparison before you spend. Not afterwards, when the numbers are in and the comparison that flatters them is easy to find. A holdout you decided on in advance is a measurement. A comparison you selected after seeing the results is a story.

Three: where you cannot measure, say so. A great deal of small-business marketing cannot be measured to the standard this course describes, because the arithmetic of statistical power does not care how much you want an answer. The professional response is not to fabricate confidence and it is not to give up. It is to know which of your decisions rest on measurement and which rest on judgement, and to be able to tell a room which is which.

Four things get fixed now, before Level 0, and do not change for eight weeks.

The offer. One real product, service, programme or cause that you are responsible for. It must have a price or an equivalent commitment, a buyer who can decline, and something you are allowed to change about how it is sold. Everything in the course runs against it.

The number that would tell you. Name the single business outcome that would make you say the eight weeks were worth it. Sales, qualified enquiries, bookings, registrations, renewals. One, not a family of them. Write down what it is today and where that figure comes from.

The record. Decide now where your campaign record lives and make the first entry today, dated, saying what you currently believe your marketing is producing and why. In week eight you will read it back. Almost everyone finds the entry disagrees with what they remember believing.

The room. Name one person who will have to be convinced: a client, a director, a finance lead, a board, a spouse who is funding it. Level 5 runs against that person, and a campaign nobody has to defend has not been tested.

Seven rules. Where a rule rests on published evidence it is marked, and where it is a design choice it is marked as that, because a course about unexamined claims should not slip its own preferences past you unlabelled.

One. Everything runs against a real offer. No invented brands, no competitor teardowns, no campaign you would run if you had the budget. Design choice. A case has all its information present, nobody withholding a number and nobody whose bonus depends on the answer, and marketing skill built on that material collapses on first contact with a market that has all three.

Two. Every number carries where it came from and what it counted. Not a footnote: in the sentence. Which system produced it, over what dates, and what it treated as a conversion. Design choice. Definitions carry more variance in marketing reporting than sample sizes do, and the definition is the part nobody writes down.

Three. Attributed and incremental are different words and are never used interchangeably. Evidenced. Across fifteen randomised experiments at Facebook covering roughly 500 million user-experiment observations, half the studies had observational estimates of purchase lift off by a factor of three, and the bias varied in direction as well as size, so no correction factor exists.

Four. Before you spend, you write down what success is and what would have to be true for you to stop. Design choice. A success criterion invented after the results is not a criterion, and everyone who has ever presented a campaign has watched one being invented in the room.

Five. A test changes one thing, and its size is decided before it runs. Evidenced. At a realistic rate of successful ideas, a test run at twenty per cent power produces declared winners that are wrong more than half the time. The number of people you need is arithmetic, not ambition.

Six. Where a claim has no source you can reach, you stop repeating it. Design choice. This field runs on numbers that dissolve on contact: the five times cost of acquisition, the rule of seven, the twenty per cent drop from half a second, the forty-two per cent that failed for want of a market. Each is repeated by people who would be embarrassed to be asked where it came from.

Seven. You answer for the spend. No platform, no agency, no dashboard and no model takes on the accountability for money you signed off. Design choice, and the position behind every experimental result in this course. Where a machine produced part of the work, the machine is an input to your judgement, and the sentence you say out loud is that you reviewed it and you stand behind it.

Six components. The weights are a design choice. What each demands is not.

Offer and audience, 15 per cent. One offer sentence that a stranger can use to say who the offer is not for, three buyer accounts built from purchases that actually happened, and ten questions collected in buyers' own words from sources you did not write. Marked on whether the accounts contain a reason not to buy and an alternative the buyer was already using. An account in which the buyer has no hesitations has not been built from a real conversation.

Search and content, 15 per cent. Twenty real queries classified by intent with the ones you should not chase named, a page map with the missing pages listed, one page published or rewritten against a named query with the date recorded, and five pieces of content against a single buyer problem with each piece's intended action written before publication. Marked on the intended actions and on whether the query work is honest about what share of clicks exists.

Paid campaign planning, 15 per cent. A campaign brief carrying objective, audience, offer, creative, destination, budget, success criterion and stopping condition, plus a test design with its required sample computed before the test. Marked on whether the objective names a business outcome and whether the sample calculation was done in advance. A brief whose success criterion is reach has failed this component regardless of the campaign's performance.

Conversion and retention, 15 per cent. A fifteen minute conversion audit, one obstacle actually removed with the result recorded, a thirty day post-purchase sequence with each message's job named, and a customer value calculation with invented inputs marked as invented. Marked on the record, including where removing the obstacle changed nothing. An audit that found no friction was not conducted on a real page.

Campaign analysis, 15 per cent. One holdout, geographic split or pause test on a real activity, and a written reading of a campaign report separating measured from attributed. Marked on the separation and on whether the analysis names what it could not establish.

The final campaign and the campaign room, 25 per cent. A campaign that ran, a measurement plan written before it started, one thing to continue, one to change and one to stop with the evidence for each, and a defence of the whole thing in a room that disagrees. Marked on the account, not the outcome. A campaign that worked and cannot be explained scores below one that failed and can.

Nothing is assessed on whether the campaign hit its number.

The check

Ten statements, each scored one to five, taken before Level 0 and again after Level 5: run the scorecard on this course's practice screen now and record the date. Its job is the line-by-line comparison in week eight: a statement that did not move is a finding about the course or about you, and either is worth knowing. It is a self-report instrument, so it gets the same label this course puts on every unmeasured claim: the total and its bands are a design choice, and the lines are where the information is.

Every load-bearing figure in this course carries a citation and one of three grades.

Evidenced means a meta-analysis, several independent studies, or one large study whose full text was read.

One study means exactly that, with the sample stated so you can weigh it.

Convention means no adequate test exists. This course says so at the point of use rather than teaching silence.

The structural fact about this field. Digital marketing has more measurement than almost any other commercial discipline and less evidence. The two are not the same thing. What gets measured is delivered by the companies selling the advertising, using definitions they set, and independent checks on those numbers have been rare, expensive, and consistently unflattering. Meanwhile the practitioner literature is built on figures that were never measurements: consulting projections, awards entries, vendor benchmarks drawn from that vendor's own customers, and blog posts citing each other in a closed loop. When the advertising elasticity literature was corrected for publication bias, the reported effects turned out to be exaggerated by a factor of 5.7. That is the field, and this course is designed against it rather than around it.

Claims you will not find in this course, and why.

"It costs five times more to acquire a customer than to keep one." Removed. There is no primary source. The earliest attribution is unpublished consultancy work from the late 1980s, and several firms claimed the same finding as their own. The two quantities are also not commensurable: acquisition cost is marginal and per customer, retention cost is usually a fixed budget divided across an existing base.

"A five per cent increase in retention raises profits by 25 to 95 per cent." Removed. The original 1990 wording is 25 to 85 per cent, applies to a five percentage point reduction in defections, and is a profitability projection computing what profit would be under an assumed improvement, with no cost of achieving it netted off. Level 4 gives you the measured elasticity instead.

"Net Promoter Score is the single most reliable indicator of growth." Removed. The independent replication, on 21 firms and more than 15,500 interviews, failed to confirm it, and found the older American Customer Satisfaction Index predicted growth better in two of the three cases reconstructed from the original's own published charts. Across 93 firms in 18 industries, plain top-two-box satisfaction predicts retention better than NPS does.

"Half a second of latency cost Google twenty per cent of its traffic." Removed. The source is a blog note of a spoken talk in 2006, and the two conditions differed in latency and in the number of results shown. Google's own randomised latency experiments, three years later, measured effects between 0.2 and 0.6 per cent. Level 3 uses those.

"Fewer choices convert better, as the jam study showed." Removed as a general law. The famous result rests on four purchasers in one cell, a direct replication in a German supermarket found nothing, and a meta-analysis of 50 experiments covering 5,036 people put the mean effect at 0.02 with a confidence interval straddling zero.

"A prospect needs seven touches." Removed. No primary source exists. The nearest real antecedent argues for three, and is a theoretical argument about response stages rather than a measurement of anything.

"Ninety per cent of businesses fail, and forty-two per cent of them for want of a market." Removed. US establishment survival has been stable across cohorts a decade apart, including recessionary ones, at about 79 per cent after one year and 51 per cent after five. The forty-two per cent is a coded self-attribution from a collection of failure post-mortems that founders chose to publish, which is a genre rather than a dataset.

Where the evidence does not reach. No controlled study establishes that educational content produces qualified enquiries. No study validates engagement as a causal step between exposure and purchase; the two randomised tests that exist find the effect does not travel that path. No randomised evidence exists that increasing retention spending increases profit. No published experiment measures form field count in a commercial checkout, trust badges, or call-to-action wording. And every incrementality experiment in this course was run by a company spending millions, because the arithmetic implies a firm spending a few thousand a month cannot power such a test on sales at all. That last gap is the most common situation in the field and the one with the least behind it. Each is marked where it arises, and the exercise that stands in its place asks you to take the measurement on your own work.

Three working artefacts, reproduced in their real format and annotated. They are the formats this course teaches, shown at full strength rather than as blank templates.

The offer sentence. One real offer, taken from the version its owner had been using for two years to the version a stranger could act on, with every intermediate draft kept. Annotated at the three points where a word that felt essential turned out to be doing nothing, and at the one place where naming the alternative changed who the offer was for.

The campaign brief that survived the room. A single page carrying objective, audience, offer, destination, budget, success criterion and stopping condition. Annotated for why the objective is written as a business outcome, why the stopping condition is there at all, and what the finance representative asked that the brief was written to answer.

The measurement plan, before and after. The same campaign's plan written before any money was spent, and the post-mortem written six weeks later against it. Annotated for what the plan got right, what only the plan makes visible in hindsight, and the one line everyone involved would have sworn was in the original and was not.