Learn digital marketing, with practice after every lesson
8 lessons, about 135 minutes of reading, and 24 multiple-choice questions. Each lesson names the mistake that most often costs people the interview, because that is where the hours actually go. Part of the free coding course.
The funnel, and the numbers that describe it
BasicsDigital marketing is measured work, and the vocabulary of that measurement is the funnel: people become aware, show interest, consider, then convert. Every channel and campaign sits somewhere on it, and the mistake beginners make is judging them all by the same metric. A brand awareness campaign that produces no immediate sales has not failed if it moved people into the top of the funnel — but you have to have agreed that in advance.
Four numbers describe almost any campaign. CTR is what share of people who saw it clicked. Conversion rate is what share of those who clicked did the thing you wanted. CAC is what one acquired customer cost you. ROAS is revenue divided by ad spend. Being able to compute those from raw figures in an interview, without a calculator, is a genuinely common screening question in marketing roles.
The number that matters most is the one furthest down the funnel that you can measure reliably. Impressions and reach are the easiest to inflate and the least meaningful — a campaign reporting two million impressions and no conversions is reporting that two million people ignored it. When you present results, lead with cost per acquisition or revenue, and let the upstream numbers explain rather than headline.
Syntax
// Work these out from raw numbers — a common interview exercise.
//
// Impressions 240,000
// Clicks 4,800
// Signups 384
// Customers 48
// Spend ₹96,000
// Revenue ₹384,000
//
// CTR = clicks / impressions = 4800/240000 = 2.0%
// Signup conv. = signups / clicks = 384/4800 = 8.0%
// Customer conv. = customers / signups = 48/384 = 12.5%
// CAC = spend / customers = 96000/48 = ₹2,000
// ROAS = revenue / spend = 384000/96000 = 4.0x
//
// The question that follows: is ₹2,000 acceptable? Only if a customer is
// worth more than that over their lifetime. CAC alone means nothing.Key points
- CTR, conversion rate, CAC and ROAS describe nearly any campaign. Be able to compute them from raw numbers without a calculator.
- CAC is meaningless without lifetime value. ₹2,000 to acquire a customer is excellent or ruinous depending entirely on what they are worth.
- Lead with the metric furthest down the funnel that you can measure honestly. Impressions headline a report when there is nothing better to say.
Practice challenge
A course business gets 40,000 monthly visitors, 800 email signups and 60 sales. Compute the two conversion rates, say which stage is the weaker one, and name the channel type that fits each funnel stage.
Visitor to lead: 2%
Lead to sale: 7.5%
Weaker: visitor to lead - the traffic is not converting into contacts
Top of funnel: SEO, social, content. Bottom: email, retargeting, branded searchVisitor to lead: ______%
Lead to sale: ______%
Weaker stage: ______
Top of funnel: ______ Bottom: ______Show a hint
- Work out each rate against the stage immediately above it, not against total traffic
- 2% signup is roughly normal; the judgement is about which one has the most room
Check yourself
1. How is CAC calculated?
Show answer
B. Spend divided by customers acquired
2. Why is CAC alone insufficient?
Show answer
B. It means nothing without knowing what a customer is worth over time
3. What does a campaign with huge impressions and no conversions actually show?
Show answer
B. That a great many people ignored it
SEO that survives an algorithm update
Working levelSearch engine optimisation has three parts. Technical SEO is whether a search engine can crawl, render and index your pages at all. On-page is whether the content answers the query someone actually typed. Off-page is whether anyone else considers you credible, which mostly means links. Work in that order — no amount of content strategy rescues a site that returns errors to crawlers.
The single most important shift to understand is that search intent beats keyword density. Someone searching 'best crm for small business' wants a comparison, not a product page; someone searching 'what is crm' wants a definition. Matching the format the query implies matters more than how many times the phrase appears, and this is precisely what the 2024-2026 updates rewarded and what keyword-stuffed pages lost to.
The other durable change is that thin, mass-produced pages are actively penalised rather than merely ignored. Generating a thousand near-identical location pages used to work and now drags the whole domain down, including the pages that were fine. Depth on fewer pages is the strategy that survives updates, and 'we consolidated four thin pages into one good one and traffic rose' is a genuinely strong thing to be able to say in an interview.
Syntax
<!-- Technical first: can it be crawled, and does it say what it is? -->
<title>CRM for Small Business: 6 Compared | Acme</title> <!-- ~60 chars -->
<meta name="description" content="Six CRMs compared on price, setup time and
what each is actually good at. No affiliate rankings."> <!-- ~155 -->
<link rel="canonical" href="https://acme.com/crm-comparison/">
<!-- Structured data so the result can be enriched in the SERP -->
<script type="application/ld+json">
{"@context":"https://schema.org","@type":"FAQPage","mainEntity":[
{"@type":"Question","name":"Which CRM is cheapest for two users?",
"acceptedAnswer":{"@type":"Answer","text":"..."}}]}
</script>
<!-- Intent check before writing anything:
"best crm for small business" -> comparison, table, pricing
"what is a crm" -> definition, plain language
"crm pricing" -> numbers, no preamble
Matching the FORMAT the query implies matters more than the phrase count. -->Key points
- Technical, then on-page, then off-page. Crawlability problems make everything downstream irrelevant.
- Match the format the query implies — comparison, definition, how-to. Intent beats keyword repetition, and the updates since 2024 have enforced that.
- Thin mass-produced pages now drag down the whole domain, not just themselves. Consolidating beats adding.
Practice challenge
A campaign spent 50,000 rupees, produced 25,000 clicks at a 6% CTR and 40 customers. Average customer value is 3,000 rupees. Compute CPC, CAC and ROAS, then state whether to keep spending.
CPC: 2 rupees
CAC: 1,250 rupees
ROAS: 2.4x (120,000 revenue on 50,000 spend)
Decision: keep spending - each customer costs 1,250 and is worth 3,000, so the campaign is profitable even before repeat purchasesCPC: ______
CAC: ______
ROAS: ______
Decision: ______Show a hint
- CAC is spend divided by CUSTOMERS, not by clicks
- A high CTR tells you the ad is appealing, not that the campaign makes money
Check yourself
1. What order should SEO work be done in?
Show answer
B. Technical, on-page, off-page
2. What does 'search intent' mean in practice?
Show answer
B. The format the query implies — comparison, definition, how-to
3. What is the effect of thin mass-produced pages now?
Show answer
B. They drag down the ranking of the whole domain
Attribution, and why the numbers disagree
AdvancedAttribution is deciding which touchpoint gets credit for a conversion, and it is the hardest honest problem in marketing. A customer sees an Instagram ad, searches your brand a week later, clicks a Google ad, and buys. Which channel earned that sale? Last-click says Google, first-click says Instagram, and both are defensible and both are wrong in isolation.
This is why platform-reported numbers never add up. Every ad platform claims conversions it touched, so Meta claims the sale and Google claims the same sale, and the sum of platform-reported conversions routinely exceeds your actual order count. A marketer who reports the platform numbers as fact, without reconciling against the order table, will eventually be asked why the totals do not match — and there is no good answer to that in a meeting.
The practical stance is to pick a model, state it explicitly, and compare against a source of truth you control. Your own analytics and your order database are that source. The strongest teams also run holdout tests — switch a channel off in one region and measure what actually changes — because incrementality is the only way to answer the real question, which is not who gets credit but whether the spend caused anything at all.
Syntax
// Same journey, four models, four different answers.
//
// Day 1 Instagram ad (impression, no click)
// Day 3 Google search ad (click)
// Day 7 Email (click)
// Day 7 Purchase ₹4,000
//
// Last-click Email 100%
// First-click Instagram 100%
// Linear each touch 33%
// Time-decay Email most, Instagram least
//
// None is 'correct'. State which one you used, every time.
//
// The question that actually matters is incrementality:
// Turn the channel off in one region for four weeks.
// Did total conversions fall? By how much?
// That measures CAUSE. Attribution only allocates credit.Key points
- Every platform claims conversions it touched, so platform-reported totals overlap and exceed reality. Reconcile against your own order data.
- There is no correct attribution model. Choose one, state it whenever you report, and keep it consistent so trends stay comparable.
- Holdout tests measure incrementality — whether the spend caused anything — which is the question attribution cannot answer.
Practice challenge
Organic traffic fell 35% in a week. Rankings are unchanged for the top ten keywords. List three causes that fit those two facts together, and name the tool you would check first.
Cause 1: an AI Overview or featured snippet now answers the query, so impressions hold and clicks fall
Cause 2: tracking broke - an analytics tag lost on a deploy
Cause 3: seasonality or reduced search demand for those terms
Check first: Search Console - compare impressions against clicksCause 1: ______
Cause 2: ______
Cause 3: ______
Check first: ______Show a hint
- Rankings unchanged plus traffic down points at clicks, not positions
- One of the three causes means the traffic never actually left
Check yourself
1. Why do platform-reported conversions exceed real orders?
Show answer
B. Each platform claims any conversion it touched, so the same sale is counted repeatedly
2. Which attribution model is correct?
Show answer
C. None — choose one, state it, and stay consistent
3. What does a holdout test measure that attribution cannot?
Show answer
B. Incrementality — whether the spend actually caused conversions
Report a channel to the person paying for it
Job-readyThe monthly report is where marketing careers are made and lost. The people reading it control the budget, do not share your vocabulary, and are deciding one thing: whether to spend more here or somewhere else. A report full of impressions and click-through rates does not answer that, so the answer defaults to no.
Translate everything into money and decisions. Not "CTR improved 0.4 points" but "we now acquire a customer for ₹1,250 against a customer worth ₹3,000, so every additional ₹1 lakh of spend should return about ₹2.4 lakh — the constraint is inventory, not budget." Same data, and the second version is a recommendation somebody can approve.
Report what did not work, in your own words, before anyone asks. A channel you paused and why, a test that failed, a number that dropped. Marketers who only report wins get read sceptically, and rightly — every channel has a bad month, so a report that never has one is a report that is hiding something. Owning the bad month is what makes the good month believable.
Syntax
Paid search — August, one page
THE ASK Increase budget from 5L to 8L/month
WHY CAC ₹1,250 vs customer value ₹3,000 (2.4x)
400 customers at 5L spend (5,00,000 / 400 = 1,250); unit economics hold at 8L
based on July's headroom test at 6.5L (CAC ₹1,310)
WHAT WORKED Branded + high-intent terms: CAC ₹740
WHAT DIDN'T Broad match "analytics software": ₹4,100 CAC, PAUSED
on the 12th. That is ₹48,000 I would not spend again.
RISK CAC drifts up as we exhaust high-intent volume.
Reviewed at 6.5L, hard stop if CAC passes ₹2,000.
NOT MEASURED Assisted conversions from organic. Last-click only,
so paid is probably slightly over-credited here.Key points
- Convert every metric into money before it reaches the report. CAC against customer value is a decision; CTR is trivia to a budget holder.
- Name what failed and what it cost. It is the thing that makes the rest of the report credible.
- State what your measurement cannot see. Last-click attribution over-credits paid, and saying so first is far better than being caught by it.
Practice challenge
Paid search spent ₹5,00,000 this month and produced 400 customers. A customer is worth ₹3,000. One campaign group ran at a ₹4,100 CAC before you paused it on the 12th, having spent ₹48,000 on it. Write the four lines of a report asking to raise the budget to ₹8,00,000.
Ask: raise paid search from ₹5L to ₹8L per month.
Why: CAC is ₹1,250 (5,00,000 / 400) against a customer worth ₹3,000 — a 2.4x return, ₹12,00,000 of value on ₹5,00,000 of spend. Every additional lakh should return roughly ₹2.4 lakh while the economics hold.
Didn't work: the broad-match group at ₹4,100 CAC — above customer value, so paused on the 12th. That is ₹48,000 I would not spend again.
Risk: CAC drifts up as high-intent volume is exhausted. Review at ₹6.5L; hard stop if blended CAC passes ₹3,000, which is the point the channel stops paying for itself.The ask: ______
Why, in money: ______
What didn't work: ______
Risk and stop condition: ______Show a hint
- Do the division before writing anything: spend divided by customers, then compare to what a customer is worth
- The stop condition should be the number at which the channel stops being profitable, not an arbitrary ceiling
Check yourself
1. What makes a marketing report actionable to a budget holder?
Show answer
B. Metrics converted into cost and revenue
2. Why report what failed?
Show answer
B. It is what makes the successes believable
3. Why state attribution limits yourself?
Show answer
B. Because last-click over-credits paid and being caught is worse
Who you are actually selling to
BasicsEvery channel decision downstream depends on one thing being settled first: who this is for and why they would switch. Marketing to 'everyone' produces messaging that is true of every competitor and therefore persuades nobody. The useful unit is the segment that shares a problem — not a demographic, a problem — because people buy to resolve a situation, and two buyers of the same age in the same city may be in entirely different situations.
Positioning is the sentence that follows: for this segment, we are the thing that does X, unlike the alternative they would otherwise pick. That last clause is the part usually skipped and it is where the work is, because it forces you to name the real alternative — which is frequently a spreadsheet, an agency, or doing nothing at all, rather than the competitor you benchmark against. A position that does not name a trade-off is not a position; if you claim to be the cheapest and the most complete and the easiest, the market reads it as marketing and discounts all three.
The way to keep this from being an internal opinion is to source the language from customers and then test it. Talk to people who recently bought and to people who nearly did, and write down the words they use for the problem, because those are the words that convert. Then test the message before scaling spend: two landing pages with the same offer and different framing, enough traffic to tell them apart, and the winner decided by the metric that matters. That sequence — segment, position, customer language, test — is what stops a campaign being expensive guessing.
Syntax
// WHO, exactly. A demographic is not a segment; a problem is.
// weak: "SMBs in India, 25-45"
// useful: "ops leads at 20-100 person D2C brands who reconcile
// three courier dashboards by hand every Monday"
// POSITIONING -- the 'unlike' clause is the whole job
// For <segment>
// we are the <category> that <specific outcome>
// unlike <the real alternative> <- often a spreadsheet,
// an agency, or doing nothing
//
// "Cheapest AND most complete AND easiest" reads as marketing
// and gets all three discounted. Name the trade-off.
// USE THEIR WORDS, NOT YOURS
// interview 5 recent buyers + 5 who nearly bought
// ask: what were you doing before? what nearly stopped you?
// the phrases they repeat are the phrases that convert
// ("I just want one number I can trust" beats "unified analytics")
// THEN TEST BEFORE YOU SCALE SPEND
// same offer, two framings, split traffic
// A "Reconcile three courier dashboards in one screen"
// B "Save 6 hours every Monday"
// decide on the metric that matters (qualified signups),
// not the one that moves first (clicks)
//
// Sequence: segment -> position -> their language -> test -> scale.
// Skipping to spend is what makes a campaign expensive guessing.Key points
- Segment by shared problem, not demographics. Two buyers of the same age and city can be in completely different situations, and situation is what drives the purchase.
- The 'unlike' clause is the position. Name the real alternative — often a spreadsheet or doing nothing — and accept a trade-off, or the claim reads as noise.
- Take the wording from customer interviews and test it before scaling spend. The phrase buyers repeat outperforms the phrase the team is proud of.
Practice challenge
A tool reconciles courier invoices for D2C brands. The team wrote: 'For SMBs aged 25-45, we are the cheapest, most complete and easiest logistics platform.' Rewrite the segment as a problem, write the positioning with an explicit 'unlike' clause, and say why the three superlatives weaken it.
Segment: ops leads at 20-100 person D2C brands who reconcile three courier dashboards by hand every Monday
Positioning: For ops leads at small D2C brands, we are the reconciliation tool that matches courier invoices to orders automatically, unlike the spreadsheet process they run today
Why superlatives fail: claiming cheapest and most complete and easiest names no trade-off, so it reads as marketing language and all three claims get discounted. A position requires giving something up.Segment: ______
Positioning: For ______, we are the ______ that ______, unlike ______
Why three superlatives fail: ______Show a hint
- A demographic is not a segment - a shared problem is
- The real alternative is usually a spreadsheet or doing nothing, not a competitor
Check yourself
1. What makes a useful segment?
Show answer
B. A shared problem or situation that drives the purchase
2. Why does the 'unlike' clause matter in positioning?
Show answer
B. It forces you to name the real alternative and accept a trade-off
3. Where should campaign language come from?
Show answer
B. Interviews with recent and near buyers
Running paid without lighting money on fire
Working levelPaid media buys attention immediately, which is its advantage and its danger, because it will spend the budget whether or not the campaign works. Search and social differ in one fundamental way that governs everything else: search captures demand that already exists — someone typed the query — while social creates demand among people who were not looking. That means search intent is usually higher and volume is capped by how many people search, while social has effectively unlimited reach and needs the creative to do the persuading.
Structure decides whether you can learn anything. Group tightly enough that each ad set has one audience and one message, because a campaign mixing five audiences reports one blended number that tells you nothing about which audience worked. But do not split so finely that no group gets enough conversions to be significant, which is the more common failure — at a 2% conversion rate, thirty conversions needs fifteen hundred clicks, and below that you are reading noise and acting on it confidently.
The discipline is in what you optimise toward and when you stop. Optimising for clicks buys clicks, which is exactly what you asked for and not what you wanted; optimise toward the conversion that has business value even though it is rarer and slower to accumulate. Set the maximum you will pay for that conversion before launching, from your actual margin and not from ambition. And kill losers on a schedule rather than by feel — a fixed review point with a pre-agreed threshold, because the alternative is the sunk-cost argument that a campaign only needs a bit longer, which is how budgets disappear.
Syntax
// SEARCH vs SOCIAL -- different jobs
// Search captures demand that EXISTS (they typed the query)
// higher intent, volume capped by search volume
// Social creates demand among people not looking
// unlimited reach, the CREATIVE does the persuading
// STRUCTURE: one audience + one message per ad set,
// but big enough to be readable
// too broad -> a blended number that explains nothing
// too narrow -> 4 conversions per group, all of it noise
//
// at 2% CVR, 30 conversions needs ~1,500 clicks per group.
// Below that you are reading randomness.
// OPTIMISE FOR WHAT YOU WANT, NOT WHAT IS EASY TO COUNT
// optimise to clicks -> you get clicks. That is all.
// optimise to purchase /
// qualified lead -> rarer, slower, correct
// SET THE CEILING FROM MARGIN, BEFORE LAUNCH
// AOV 2,000 x 30% margin = 600 gross per order
// target CAC <= 300 -> max CPA 300, decided in advance
//
// CPC 20 and CVR 2% -> CPA = 20 / 0.02 = 1,000 STOP.
// Either CVR must triple or CPC must fall by two thirds.
// KILL ON A SCHEDULE, NOT BY FEELING
// review at a fixed spend (e.g. 3x target CPA with no
// conversion) and pause. "It needs more time" is the
// sunk-cost argument, and it is how budgets vanish.Key points
- Search captures existing demand and social creates it. That difference sets the intent you can expect and how much work the creative has to do.
- Each ad set needs enough conversions to be readable — roughly thirty. Splitting finer produces confident decisions made on noise.
- Derive the maximum cost per acquisition from margin before launching, and pause at a pre-agreed spend threshold rather than when it feels wrong.
Practice challenge
Average order value is 2,000 with a 30% margin. The campaign shows CPC 20 and conversion rate 2%. Compute the cost per acquisition and gross margin per order, say whether to run it, and state what would have to change. Then say how many conversions an ad set needs before its result is readable.
CPA = 20 / 0.02 = 1,000
Gross margin per order = 2,000 x 30% = 600
Run it? No - it loses 400 on every order
What must change: conversion rate would need to roughly triple (to ~6%) at the same CPC, or CPC would need to fall by about two thirds, or AOV/margin would need to rise. Set the maximum CPA from margin before launch.
Conversions needed: about 30 per ad set; at 2% that is roughly 1,500 clicks, and below that you are reading noiseCPA = ______
Gross margin per order = ______
Run it? ______
What must change: ______
Conversions needed per ad set: ______Show a hint
- CPA is cost per click divided by conversion rate
- Compare it against gross margin, not revenue
Check yourself
1. Roughly how many conversions does an ad set need before you can trust its result?
Show answer
B. About 30
2. CPC is 20 and conversion rate is 2%. What is the cost per acquisition?
Show answer
C. 1,000
3. Why not optimise toward clicks?
Show answer
B. The platform delivers cheap clicks from people who will not buy
Email and lifecycle: the channel you own
AdvancedEmail is the only significant channel not rented from a platform. An algorithm change cannot take your list away, the marginal cost of a send is near zero, and it consistently returns more per rupee than paid channels — which is why it is usually the most underinvested thing in a marketing function. The asset is the list plus permission, and both are damaged by exactly the behaviour that produces a good week: sending more, to people who did not ask, about things they do not want.
Lifecycle means the sequence is triggered by what someone does rather than by the calendar. A welcome series when they sign up, an onboarding nudge when they have not completed setup, a cart recovery when they leave without buying, a win-back when they have not opened anything in ninety days. These outperform broadcasts by a wide margin because they arrive when the message is relevant, and once built they run without anyone deciding to send them. A weekly newsletter to the whole list is the least effective thing an email programme can do and is usually the only thing it does.
Deliverability is the part that is invisible until it is catastrophic. Mailbox providers judge you on engagement — opens, replies, and crucially complaints and hard bounces — and a list bought or scraped will destroy a sending domain's reputation quickly, after which even the people who genuinely want your email stop receiving it. Authenticate the domain with SPF, DKIM and DMARC, warm a new domain gradually, and prune unengaged addresses. Removing people who have not opened in six months feels like deleting the asset and is what protects it, because sending to dead addresses is the signal that puts you in spam.
Syntax
// THE ONLY CHANNEL YOU OWN.
// Algorithms cannot take a list away; the asset is list + permission.
// LIFECYCLE (triggered by BEHAVIOUR) beats broadcast (calendar)
// signup -> welcome series, days 0 / 2 / 5
// setup incomplete -> nudge at 24h with the single next step
// cart abandoned -> recover at 1h, 24h, 72h
// no open in 90d -> win-back, then suppress
// churned -> reactivation at a real reason to return
//
// A weekly newsletter to the whole list is the least effective
// thing an email programme can do -- and often the only thing it does.
// DELIVERABILITY -- invisible until it is catastrophic
// authenticate: SPF + DKIM + DMARC on the sending domain
// warm a new domain slowly (hundreds -> thousands over weeks)
// NEVER send to a purchased or scraped list -- one campaign can
// end a domain's reputation permanently
// complaint rate > 0.1% -> you are in trouble
// hard bounces: remove immediately, every time
// PRUNING FEELS LIKE DELETING THE ASSET. IT PROTECTS IT.
// suppress no-opens at 6 months
// 100k list at 8% open -> 40k list at 22% open
// more actual readers, better placement, lower cost
// MEASURE REVENUE PER RECIPIENT, not open rate.
// Opens are inflated by privacy proxies that pre-fetch images
// and are no longer a reliable metric on their own.Key points
- Behaviour-triggered sequences beat calendar broadcasts because they arrive when the message is relevant, and they keep running once built.
- Deliverability depends on engagement and complaint rates. A purchased list can permanently damage a sending domain, after which willing subscribers stop receiving you too.
- Measure revenue per recipient. Open rates are inflated by privacy proxies that pre-fetch images and no longer mean what they used to.
Practice challenge
A brand sends one newsletter weekly to a 100,000-address list. Open rate is 8%, complaints are rising and deliverability is dropping. Name four behaviour-triggered sequences to build, explain why suppressing non-openers helps rather than hurts, and say why open rate alone is now unreliable.
Sequences: 1 welcome series on signup 2 onboarding nudge when setup is incomplete 3 cart abandonment at 1h/24h/72h 4 win-back after 90 days with no open, then suppress
Why suppression helps: mailbox providers judge on engagement, and sending to dead addresses is a spam signal. A 40,000 list at 22% open reaches more actual readers than 100,000 at 8%, and protects placement for everyone who does want it.
Why open rate is unreliable: privacy proxies pre-fetch images and register opens that never happened
Metric instead: revenue per recipientSequences: 1 ______ 2 ______ 3 ______ 4 ______
Why suppression helps: ______
Why open rate is unreliable: ______
Metric to use instead: ______Show a hint
- Triggered by behaviour beats triggered by the calendar
- The list is an asset, and the thing that damages it is sending more
Check yourself
1. Why does lifecycle email outperform a broadcast newsletter?
Show answer
B. It is triggered by behaviour, so it arrives when it is relevant
2. What does suppressing subscribers who have not opened in six months do?
Show answer
B. Protects deliverability — sending to dead addresses is a spam signal
3. Why is open rate now unreliable on its own?
Show answer
B. Privacy proxies pre-fetch images and inflate opens
Landing pages and testing without fooling yourself
Job-readyTraffic is only half of an acquisition problem and usually the expensive half. If a landing page converts at 1% and you double it to 2%, every channel feeding it just halved its cost per acquisition, with no additional spend. The page has one job — continue the promise that brought the visitor and remove whatever is stopping them — and the most common defect is a mismatch, where the ad promises one specific thing and the page opens with a general description of the company.
What actually moves conversion is rarely styling. It is message match between ad and headline, clarity about what happens next, proof that is specific enough to be checkable, and removing friction — every extra form field costs completions, and asking for a phone number before there is any trust costs more than it collects. It helps to state the objection you are answering: 'no card required' beside a signup button removes a specific fear, while 'trusted by thousands' removes nothing because it is what everyone says.
Testing is where teams fool themselves most confidently. A test needs a hypothesis, a single primary metric chosen in advance, and a sample size calculated before it starts — because peeking at a running test and stopping when it looks significant produces false winners at a rate far above the 5% people believe they are accepting. Run for whole weeks to cover the weekly cycle, and expect most tests to show no difference, which is a real and useful result. The failure mode to name is running many variants at once on low traffic: with twenty tests, one crosses significance by chance alone, and that is the one that gets rolled out and reported.
Syntax
// CONVERSION IS THE CHEAP HALF
// 10,000 visits, 1% -> 100 signups
// same traffic, 2% -> 200 signups (every channel's CAC halves)
// WHAT MOVES IT (roughly in order)
// 1. MESSAGE MATCH -- ad said "reconcile courier dashboards";
// page must say that, not "We are a logistics platform"
// 2. ONE clear next action, above the fold
// 3. SPECIFIC, checkable proof
// weak: "Trusted by thousands" (everyone says it)
// strong: "Cut Monday reconciliation from 6 hours to 20
// minutes" -- Ops lead, 40-person D2C brand
// 4. LESS FRICTION -- every field costs completions.
// Asking for a phone number before trust exists costs
// more than it collects.
// 5. ANSWER THE OBJECTION: "No card required."
// TESTING WITHOUT FOOLING YOURSELF
// BEFORE starting:
// hypothesis "naming the time saved in the headline
// will raise signup rate"
// ONE primary metric (qualified signups, not clicks)
// sample size CALCULATED, not guessed
// baseline 2%, detect +0.5% -> ~6,000 per variant
//
// PEEKING AND STOPPING AT SIGNIFICANCE INFLATES FALSE
// WINNERS WELL ABOVE 5%. Decide the end in advance.
//
// Run whole weeks -- weekday and weekend behave differently.
// Most tests show no difference. That is a real result.
//
// 20 variants at once -> ~1 wins by chance alone,
// and that is the one that gets shipped and presented.Key points
- Doubling conversion halves the cost per acquisition of every channel at once, without extra spend. It is usually cheaper than buying more traffic.
- Message match between ad and headline outranks design. A specific, checkable proof point beats a superlative that every competitor also claims.
- Fix the sample size and the primary metric before starting. Peeking and stopping at significance produces false winners far more often than the 5% people assume.
Practice challenge
A team runs 20 landing page variants at once on modest traffic, checks daily, and ships whichever is ahead when one crosses significance. Name the three things wrong, write what must be fixed before starting a test, and say what result they should expect most of the time.
Wrong 1: 20 variants means about one crosses significance by chance alone, and that is the one they ship
Wrong 2: peeking daily and stopping at significance inflates false winners far above the assumed 5%
Wrong 3: modest traffic split 20 ways gives no variant enough conversions to be readable
Fix before starting: one hypothesis, one primary metric chosen in advance, a calculated sample size, and a fixed end point - run whole weeks so weekday and weekend are both covered
Expected most of the time: no significant difference, which is a real and useful resultWrong 1: ______
Wrong 2: ______
Wrong 3: ______
Fix before starting: ______
Expected result most of the time: ______Show a hint
- Running many variants at once is a multiple-comparisons problem
- Deciding when to stop after seeing the data is the core error
Check yourself
1. A page converts at 1%. Doubling it to 2% does what to cost per acquisition?
Show answer
B. Halves it across every channel feeding the page
2. Why is peeking at a running test and stopping at significance a problem?
Show answer
B. It inflates false winners well above the assumed 5%
3. You run 20 variants on modest traffic. What should you expect?
Show answer
B. About one crosses significance by chance alone
Common questions
Do I need any background to start Digital Marketing?
No. This track begins at its own beginning and assumes nothing, and the first lesson explains what the thing is before showing you any syntax.
How long does the Digital Marketing track take?
About 135 minutes of reading across 8 lessons, plus the practice challenges and 24 multiple-choice questions, which is where the time actually goes.
Is it free?
Yes, and there is no account. Everything runs in your browser.
More: all 15 tracks · what employers actually ask for · the full syllabus