Digital Marketing & Advertising: Where Attention Is Bought Now

You type a few words into a search box. Before the page finishes loading, a handful of advertisers have already bid for the right to sit above the results, a ranking has been computed, and the winner's price has been fixed. All of it takes less time than a blink. Later you open a news site, and while its text arrives, an exchange somewhere offers your visit, with a profile attached, to a room of bidders. By evening, an email, a post and an ad for the same shoes have each claimed that you bought them.

Digital marketing is the machinery behind those moments: how attention is bought, how it is earned for free, how the results are counted, and how the rules about your data keep changing. This piece takes the free-to-start ideas from the library's Social Media Marketing course and goes further, into the money and the evidence.

How ads are bought: auctions in a blink

Almost all online advertising is sold by auction, because the thing for sale (this person, on this page, right now) exists once and has no list price. Auctions have a famous design problem: if you pay what you bid, you should bid less than the thing is worth to you, and everyone ends up second-guessing everyone. In 1961 William Vickrey described the fix. In a second-price auction the highest bidder wins but pays only the second-highest bid. Now there is no point shading: bidding exactly what it is worth to you is the best you can do, whatever the others bid.

Search ads use a variant. Several ads share the page, ranked by Ad Rank, roughly bid times a quality score (how likely people are to click, how relevant the ad and the landing page are). You pay only when someone clicks, and the textbook rule is that you pay the least you could have bid and still held your position:

This "generalised second price" rule is not quite Vickrey's: it no longer makes honesty the strictly best bid, but it stays stable and easy to explain, and a high-quality ad can beat a higher bidder. That is why a relevant ad can pay less than a careless one for the same spot. Try changing the bid and quality of one advertiser below.

Display and video ads, the banners and clips around other people's content, use programmatic buying: when a page loads, the site's ad server offers the slot to many bidders in a fraction of a second, and the highest bid wins. For years these used second-price rules. Then, in a change Google announced on 6 March 2019 and rolled out over the following months, Google's Ad Manager moved its display and video marketplace to a first-price auction, where the winner pays what it bid, and dropped the "last look" advantage that Google's own exchange had enjoyed. Buyers who now have to guess how low they can go use "bid shading", which brings the price down to what the winner needed.

The pattern to carry away: in second price, your best bid equals your value, so bidding is easy. In first price, your best bid is your value times k/(k+1), where k is the number of rivals: the more rivals, the closer to your full value. Moving to first price made auctions simpler to explain to a publisher, harder for buyers to play, and a bit more of a guessing game.

Earning attention: search optimisation, social, content and email

Paid ads are only the visible part. Search engine optimisation (SEO) is the work of earning a place in the unpaid results. In plain words, a search engine crawls the web, builds an index of what it finds, and then ranks pages for a query by how relevant and trustworthy they look. The original idea behind Google's ranking, described by Sergey Brin and Larry Page in 1998, was that a link from another page is a vote, and votes from pages that are themselves well-linked count for more. The engines now use many more signals and change them often, so the durable advice is dull: make pages that answer the question people actually typed, load fast, work on a phone, can be crawled, and earn links from sites people trust. Tricks that aim at the ranking system rather than the reader are a race you rent, not own.

Social media marketing is attention earned through posting, community and creators, and rented through paid placements. The rules are set by each platform's feed, so a strategy built on one feed is a strategy built on someone else's decisions. The library's course covers the how-to; the lesson here is strategic. Content marketing (useful articles, videos, tools) builds an asset that keeps bringing people in; email is the channel you own, a list you can reach without paying a platform each time.

Email carries a measurement lesson too. Since 2021, Apple's Mail Privacy Protection loads message images in the background for users who turn it on, so "opens" stopped meaning that a person read it. Clicks, replies and purchases are the numbers worth reading. And lists must be built with permission: the EU's rules and many countries' laws require opt-in, while US law sets different baseline rules for commercial email, so check the law where your readers live.

The funnel: where the people go

The funnel is the oldest picture in marketing: many people notice, fewer click, fewer still sign up, fewer buy. Each step has a conversion rate, and the chain multiplies. That multiplication hides a point most people miss. A ten-percent improvement at any stage gives the same ten percent more buyers. But the stages cost different things. An ad that earns more clicks brings more visitors and a bigger bill; a landing page that converts more of the same visitors costs nothing extra. The metric to watch is cost per acquisition (CPA): spend divided by buyers, compared with what a buyer is worth.

At the defaults, 100,000 views give 2,000 clicks costing $1,600, 100 sign-ups and 20 buyers: a CPA of $80 against $90 of profit per buyer, a $200 gain. Raise the click-through rate by 10% and you get 22 buyers, but spend $1,760, so CPA is still $80. Raise the sign-up rate by 10% and you get 22 buyers for the same $1,600: CPA falls to $72.73. Fixing the page beats winning more clicks, until the page is as good as it will get.

Attribution: why "last click" lies

A customer sees a creator's video, then a banner, then an email, then searches your brand name and buys. Which touch gets the credit? The default in many tools is last click: the final touch gets everything. It is easy to compute and systematically unfair. The search at the end happens because the earlier touches created the interest; paid brand search, as we shall see, often just collects people who were coming anyway. Other rules share the credit differently: first click rewards whatever introduced the customer; linear splits equally; time decay gives more to touches near the sale; position-based gives the first and last the most. Each is a rule of thumb. None is a measurement of cause.

Incrementality: the question attribution cannot answer

Attribution tells you who was present before a sale. Incrementality asks what the ad added: how many of those sales would have happened anyway? The only reliable way to find out is an experiment: show the ad to a random group, hold it back from a matched control, and compare outcomes. That is the same logic as the A/B test in Marketing Research & Analytics (34/59), applied to advertising.

The best-known example is eBay's. A team of economists (Thomas Blake, Chris Nosko and Steven Tadelis) worked with eBay on field experiments, published in Econometrica in 2015. In one, eBay stopped paid ads on queries for its own brand for a random subset of users; the free link still appeared. They found that ads on brand keywords had no measurable short-term benefit, because people who lost the paid link almost always clicked the free one. For non-brand keywords the story was more nuanced: new and infrequent users responded to ads, but most of the spend went to frequent users whose purchases the ads did not seem to influence, so average returns were negative. Their conclusion: the returns to paid search were a fraction of what non-experimental estimates suggested, partly because people who search for something are exactly the people likely to buy it.

Experiments are not free. Randall Lewis and Justin Rao (2015, Quarterly Journal of Economics) examined 25 large experiments run with major US retailers and brokerages, most reaching millions of customers, and found that measuring advertising's return precisely is hard even then: ad effects are small compared with the natural swings in people's purchases, so the uncertainty around the answer stays wide. A separate study at Facebook (Gordon and colleagues, 2019) compared large randomised experiments with the observational methods advertisers commonly use, and found that the observational methods often did not reproduce the experimental answer. Many ad platforms now sell "lift studies" and geographic holdout tests for this reason.

At the defaults, the dashboard credits the ad with 200 sales and $10,000 of revenue against $2,000 of spend: a return of 5 to 1. If 90% of those clickers would have bought through the free link anyway, the ad caused only 20 sales and $1,000 of revenue, a return of 0.5 to 1. The same ad, the same click data, a tenfold difference. The second half shows the price of finding that out: detecting a 5% lift on a 2% buying rate needs roughly 300,000 people per group.

Privacy rules, cookies and a measured future

Almost everything above runs on identifiers: cookies, device IDs, email addresses. Governments have been regulating them for a decade, and the rules differ by country. A fair summary, dated as of the sources checked for this piece:

Where	Rule	What it does for advertisers
EU / EEA	GDPR, applying since 25 May 2018; the separate ePrivacy rules on cookies	Needs a legal basis for processing personal data; consent must be specific and unambiguous (no pre-ticked boxes); applies to outside firms that target people in the EU; fines up to €20 million or 4% of worldwide annual turnover, whichever is higher.
California, US	CCPA (passed 2018, in force 1 January 2020), later amended by the CPRA; similar state laws in Virginia and Colorado (2021)	Right to know, delete and opt out of the "sale" or sharing of data. The US has relied on state laws rather than a single national statute.
China	Personal Information Protection Law, enacted 2021	Consent and purpose rules modelled in part on the GDPR, with its own state-oversight provisions.
Brazil, India	LGPD (law of 2018, in force 2020); India's Digital Personal Data Protection Act (2023)	Consent-based regimes for personal data, with national regulators.
Elsewhere	Japan, South Korea, Singapore, Thailand and South Africa have GDPR-influenced laws; Turkey adopted one in 2016	A growing patchwork, which is why global advertisers build for the strictest region.

Platforms added their own rules. Apple's App Tracking Transparency, introduced with iOS 14.5 in 2021, made apps ask permission before tracking people across other companies' apps, which reduced the cross-app data that tracking-based ads relied on. Safari and Firefox block third-party cookies by default.

What about Chrome? This is the part that most often goes stale, so read the date. Google announced in 2020 that it would phase out third-party cookies in Chrome and replace them with its Privacy Sandbox tools, then postponed it several times. In 2025 it said Chrome would maintain its current approach of offering users a third-party cookie choice, rather than removing the cookies for everyone. On 17 October 2025, Google's Privacy Sandbox team announced it was retiring most of the Privacy Sandbox technologies, including the Topics and Protected Audience interfaces and the Attribution Reporting API, citing low adoption, while continuing work on a few such as CHIPS, FedCM and Private State Tokens. As of that last official statement I could verify (October 2025), third-party cookies were still working in Chrome by default. I could not check for anything newer. If you are reading this later, check Google's Privacy Sandbox pages and your browser before relying on it.

The practical result for a marketer: depend less on tracking individuals, more on first-party data you collected with consent, on contextual placement (the page, not the person), and on experiments that measure lift without following anyone around.

What a graduate just sees

• Ad space is sold by auction; the rule of the auction decides how bidders behave.
• In a search auction, a better ad can win a higher slot at a lower price.
• Conversion rates multiply, so fix the leakiest stage; paying for more clicks and fixing the page are different spends.
• "Last click" tells you who was there at the end, not who caused the sale.
• Clicks that came from people who were already coming are not worth buying.
• Only a holdout experiment measures incrementality, and it needs a big sample.
• An "open rate" or a "view" is a proxy; read the metric nearest the money.
• Privacy rules are set by region and by platform, and both change; build on first-party, consented data.

Where it shows up next

• Product Innovation & Go-to-Market (34/63): launch plans choose channels using funnel maths and test budgets.
• Econometrics (34/44): when an experiment is impossible, the techniques that estimate cause from messy data.
• Computer Networks (34/27) and Databases (34/28): the plumbing that serves an ad in a blink and keeps its logs.
• In the world: performance-marketing teams, agencies, publishers' ad operations, privacy officers, and the data scientists who build bidding and measurement systems.

If you want to go deeper

• 📘 Textbook. Dave Chaffey and Fiona Ellis-Chadwick, Digital Marketing (Pearson). A standard university text covering channels, strategy and measurement.
• 📘 Textbook (measurement). Ron Kohavi, Diane Tang and Ya Xu, Trustworthy Online Controlled Experiments (Cambridge University Press, 2020). Experiments at web scale.
• 📘 Gentler (and sceptical). Tim Hwang, Subprime Attention Crisis (2020). A readable argument that the online ad market is shakier than it looks.
• 🌐 Free courses. Google Skillshop (Google Ads and Analytics training) and HubSpot Academy (free marketing courses). For the evidence side, the eBay paper below is free to read.