History of performance marketing: Hopkins to Advantage+

A hundred years of ads you can count, from coupon-keyed press ads to campaigns that pick their own audience. The machines kept winning, and so did one human job.

A 1924 Canadian magazine advertisement for Lifebuoy soap
Photo: Lifebuoy soap, Public domain, via Wikimedia Commons

The history of performance marketing is the history of advertising that is paid for, and judged by, a counted response. It runs from Claude Hopkins testing coupon-keyed ads in the 1900s and writing Scientific Advertising (1923), through pay-per-click search (GoTo.com in 1998, Google AdWords in 2000) and Facebook's social ad products (2007), to Apple's tracking limits in 2021 and the fully automated campaigns Google and Meta sell today.

Key takeaways
  • 1923: Claude Hopkins' Scientific Advertising wrote down the test-and-count method every ad platform still runs.
  • 27 Oct 1994: HotWired went live with paid banners from about a dozen brands, and the click became countable.
  • 24 Oct 2000: Google AdWords launched, refining GoTo.com's 1998 pay-per-click auction.
  • $10 billion: Meta's own estimate of the 2022 revenue hit from Apple's App Tracking Transparency.
  • 59% of Indian ad spend was digital in 2025 (dentsu-e4m), up from 12% in 2016.
  • Brand search ads showed no measurable short-term lift in eBay's field experiment: reported returns are not real returns.

Two forces repeat across that century. Every time measurement improved, money moved toward the measurable channel. And every time the channel got more measurable, it also got more automated, until the machine was choosing the audience, the bid and now the creative.

What humans kept is narrower than it looks, and more valuable. Here is the timeline, with dates and the numbers behind each turn.

1923Scientific Advertising published
24 Oct 2000Google AdWords launches
$10 billionMeta's estimate of the 2022 revenue hit from Apple's ATT
59%digital share of Indian ad spend, 2025 (dentsu-e4m)

Who invented performance marketing?

No one person invented it, but Claude C. Hopkins (1866 to 1932) wrote its rulebook. Working at the Chicago agency Lord & Thomas, which hired him in 1907 at a reported $185,000 a year, he ran the same playbook again and again:

  • ads with keyed coupons, so each insertion's replies could be counted;
  • headlines and offers tested against each other;
  • only what paid was kept and scaled.

The book that wrote it down

His 1923 book Scientific Advertising argued that advertising should be judged like salesmanship, by sales per dollar spent, and that the coupon made that measurable. His earlier work shows the method: he promoted Bissell carpet sweepers as Christmas gifts with a mailing of 5,000 letters, and he is credited with popularising toothbrushing through Pepsodent.

Strip the period language away and the loop is the one every Meta ads manager runs today: make variants, attribute responses, cut losers, scale winners. The tools changed a hundred times; the loop never did.

One loop, a hundred years1. Make variants2. Count response3. Cut losers4. Scale winners1923: hand-written headlines,coupon codes per paper2026: AI-made variants,pixel + model attributionSources: Hopkins, Scientific Advertising (1923); Meta and Google product notes
The test-and-count loop Hopkins described is the same loop an automated campaign runs. What changed is who makes the variants and how a response gets credited.
Engraving of the Western Daily Press printing press in 1858
A newspaper press in 1858. Keyed coupon ads rode on mass-printed papers: a code on each coupon told the advertiser which paper and which ad produced the reply.Photo: Samuel Loxton (artist), Public domain, via Wikimedia Commons

How did direct response move from mail to the internet?

For most of the 20th century, measured advertising meant mail order, coupons and reply cards, which tracked response by code or address. The internet replaced the coupon with the click.

The story

On 27 October 1994, Wired magazine's web affiliate HotWired went live with paid banners from a dozen brands, AT&T, MCI, Volvo, Club Med, Sprint, IBM and Zima among them. Each 476 by 56 pixel banner weighed about 10 KB. HotWired charged $10,000 a month for a section placement, which, its staff recalled to Digiday, was what a page in the print magazine cost. Co-founder Andrew Anker said they made no traffic commitments at all: advertisers bought a slot, not a number. Brian Behlendorf threw the switch on launch day. Within months the measuring began. Andy Batkin of Interactive Marketing told Digiday his early Yahoo banners drew click rates of 15%, and that the rate soon started to drop. The click had become a priced, logged unit, and the race to sell it by the click had started.

Digiday: An oral history of the first banner ad (2014)

The 44% banner and the slide since

AT&T's banner, often called the first, was reported to have drawn a 44% click-through rate. That figure is the high-water mark the whole industry has been sliding from, and it is secondhand: no original campaign data has been published.

Digiday's oral history calls "first banner ad" a misnomer, since about a dozen brands ran that week. Either way, the banner introduced the idea that every impression could be logged and every click priced, and within four years someone priced the click directly.

YearMilestoneWhat became measurable or automated
1907Hopkins joins Lord & ThomasKeyed coupons count replies per ad
1923Scientific Advertising publishedTesting written down as a method
1994HotWired banner ads go live (27 Oct)Impressions and clicks logged
1998GoTo.com sells search positions by auctionAdvertiser pays per click
2000Google AdWords launches (24 Oct)Self-serve keyword ads
2003Yahoo buys Overture (formerly GoTo.com) for about $1.63 billionPay-per-click proven as a business
2007Facebook launches Beacon (6 Nov)Targeting by social identity; off-site purchase tracking
2021Apple ATT arrives with iOS 14.5; Performance Max opens to all advertisers (2 Nov)Tracking shrinks; automation absorbs targeting
2025Meta's plan to automate ad creation and targeting by end-2026 (reported Jun)Creative itself generated by the platform
Who held the lever?1923Hopkins: human tests copy1994Banners: clicks logged1998GoTo: human bids per click2000AdWords: quality-weighted auction2007Facebook: targeting by identity2021ATT + PMax: machine targets2026Meta goal: machine makes adsHuman sets copy and bidsMachine sets audience and creative
Each turn in the timeline moved one decision from the marketer to the platform. Dates from Google, Meta, Yahoo and Digiday sources listed below.

How did pay-per-click start?

Pay-per-click started with GoTo.com, Bill Gross's search engine of 1998 (some histories say 1997), which let advertisers bid at auction, in cents per click, to appear for specific search terms. It was ridiculed for mixing ads with results, then copied by everyone.

Google's refinement

Google launched AdWords on 24 October 2000, and the model it refined, an auction weighted by how relevant and clickable the ad was, became the most profitable advertising product ever built.

GoTo, renamed Overture, became profitable in 2001 and was bought by Yahoo in 2003 for an aggregate price of about $1.63 billion, according to Yahoo's SEC filing.

The shift that matters for marketers: for the first time the buyer told the medium exactly what intent it wanted, as a keyword, and paid only when that intent clicked.

ModelWho sets the priceWhat the buyer pays for
HotWired banner, 1994Publisher, flat $10,000 a monthA placement, no traffic promise
GoTo.com, 1998Highest bidder per keywordA click
AdWords, 2000 onwardAuction weighted by relevance and click rateA click on a quality-weighted position
Performance Max / Advantage+Platform model, within a budget and goalA conversion the platform predicts and attributes
Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field ExperimentThomas Blake, Chris Nosko, Steven Tadelis · Econometrica 83 (NBER w20171) · 2015

eBay switched off parts of its paid search buying in controlled experiments to measure what the ads really caused. Brand-keyword ads showed no measurable short-term benefit, because people searching "eBay" reached eBay anyway through the organic link. On non-brand keywords, new and infrequent users responded to ads, but frequent users, who would have bought regardless, received most of the spend, so average returns were negative. The paper's broader point is that click-based attribution counts sales that would have happened anyway. The limit: one very well-known brand, whose results may overstate the problem for unknown brands.

Read the paper (free)

What did Facebook add to performance marketing?

Facebook added targeting by who people are rather than what they type. Its most ambitious early ad product, Beacon, launched on 6 November 2007 with 44 partner sites and posted users' purchases on those outside sites into their friends' feeds.

From Beacon to the pixel

Beacon was a privacy disaster: a class action followed and it was shut in September 2009. But the idea underneath it, connecting an ad view on Facebook to an action somewhere else, became the pixel and then the Conversions API, and powered a decade of direct-to-consumer brands.

Search captured demand that already existed; social ads could create demand among strangers, and measure it. That is the engine behind the D2C wave covered in CAC payback and LTV for Indian D2C.

Entrance sign of the Facebook headquarters complex at 1 Hacker Way, Menlo Park, California
Facebook's Menlo Park headquarters. Beacon's failure in 2007 led to the pixel, and the pixel to a decade of measurable social advertising.Photo: LPS.1, CC0, via Wikimedia Commons

What did iOS 14.5 do to performance marketing?

Apple's App Tracking Transparency, which arrived with iOS 14.5 in 2021, made apps ask permission before tracking users across other apps and sites, and it broke much of the measurement social ads depended on. On its February 2022 earnings call, Meta's finance chief estimated the 2022 revenue headwind at about $10 billion.

Who paid for it

The research since shows who paid:

  • advertisers lost the signal that told the algorithm which clicks became purchases, so purchase-optimised campaigns got worse;
  • spend shifted partly to Google;
  • small e-commerce firms that had built their business on Meta took the hardest hit.
Evaluating the Impact of Privacy Regulation on E-Commerce Firms: Evidence from Apple's App Tracking TransparencyGuy Aridor, Yeon-Koo Che, Brett Hollenbeck, Maximilian Kaiser, Daniel McCarthy · Management Science (in press) · 2025

The authors combined weekly campaign data from more than a thousand e-commerce firms advertising on Meta, Google and TikTok with sales data, and compared firms by how dependent they were on Meta before ATT. Purchase-optimised Meta campaigns lost about 37% of their click-through rate after ATT. Firms most dependent on Meta saw revenue growth decline 37% relative to firms with more diverse ad budgets, and Meta's ad market share fell 4.4%. Small firms, which drew about 60% of their traffic from Meta against 30% for large firms, bore most of the loss. The limit: firms that use a particular analytics provider, mostly US-based.

Read the research brief (free)

Why did ad platforms move to full automation?

Because with less user-level data, the platforms' own models became the only party that could still see enough to optimise, and they asked advertisers to hand over the levers. Google rolled Performance Max out to all advertisers on 2 November 2021, one campaign buying across YouTube, Display, Search, Discover, Gmail and Maps. Meta built its Advantage+ suite on the same logic.

Meta's 2026 target

In June 2025, Reuters, citing the Wall Street Journal, reported that Meta aimed to let brands fully create and target ads with AI by the end of 2026. The advertiser supplies a product image and a budget, and the system does the rest:

  1. writes the copy;
  2. makes the images and video;
  3. picks the audience;
  4. suggests the budget split.
He tested ten headlines by hand over months; the platform now generates and tests thousands in a day.

The arc from Hopkins is complete. How that selection works is explained in how Meta Andromeda and Advantage+ decide which ad you see, and why automated campaigns can fill a CRM with junk in why Performance Max brings junk leads.

How fast did Indian ad money move to measurable media?

Very fast. The dentsu-e4m Digital Advertising Report 2026 says digital rose from 12% of Indian ad spend in 2016 to 59% (₹71,621 crore) in 2025, and projects 70% by 2027. The Pitch Madison Advertising Report 2026, using its narrower legacy definition, has digital at 15% in 2016 and 46% in 2025.

Digital share of Indian advertising spend
See the data as a table
dentsu-e4m (%)Pitch Madison (legacy definition) (%)
20161215
202031
202442
20255946
2027 (projected)70

Source: dentsu-e4m Digital Advertising Report 2026; Pitch Madison Advertising Report 2026 (via exchange4media) · The two reports define digital differently, so the levels differ. Both show the same direction.

PMAR adds the detail that matters here: performance formats, meaning social, search and e-commerce ads, made up about 59% of digital spend in 2025 (₹31,532 crore of its core digital figure), and close to 70% once performance-driven video is included.

India's ad market is now largely bought on counted outcomes, which is why a fuller history of the market sits in the history of advertising in India.

Are platform-reported results the same as real results?

No. A platform reports the conversions it can attribute to its ads; some of those would have happened without the ad. The eBay experiment found brand-search ads produced no measurable short-term lift, which means their reported return was almost all borrowed from organic demand.

Use the calculator to see how much of a reported return survives once you remove the sales that would have happened anyway. The share is yours to estimate, ideally from a holdout or geo test; building a measurement stack after cookies covers how.

Reported ROAS vs incremental ROAS

What did humans keep as the machines took over?

Humans kept the inputs the machine cannot generate from data alone:

  • the offer and the positioning;
  • the creative idea;
  • the choice of what to measure;
  • the judgement about when a reported number is lying.

Every automation step in this history removed a manual task (counting coupons, setting bids, picking audiences) and raised the value of the strategic one above it.

What Hopkins would recognise

Hopkins himself would recognise the split. His tests chose between ideas; they never produced the idea. The same holds for an automated campaign in 2026: it can rank a thousand AI-made variants, but it ranks them on the objective you give it, with the conversion signal you send it, for the offer you built.

A weak offer optimised perfectly is still a weak offer, which is why the balance between brand and performance remains a human call. It is also how I approach the work described on my work page: systems for the repetitive parts, judgement for the parts that decide the outcome.

Questions people ask

What is the difference between performance marketing and direct marketing?

Direct marketing is the older practice of asking for a response, by mail, coupon or phone, and counting it. Performance marketing is its digital form, usually bought on a cost per click, lead or sale, with the platform's algorithm doing much of the targeting.

When did Google AdWords launch?

On 24 October 2000. It was renamed Google Ads in 2018, when Google retired the AdWords and DoubleClick brands.

Did iOS 14 kill Facebook ads?

No, but it made them less precise and hurt small advertisers most. Research by Aridor and colleagues found purchase-optimised Meta campaigns lost about 37% of their click-through rate, and Meta-dependent firms saw revenue growth fall 37% relative to more diversified firms.

Is performance marketing still worth it in India?

The money says yes: digital reached 59% of Indian ad spend in 2025 by dentsu-e4m's count, and performance formats are most of it. The caution is measurement: reported returns overstate real ones whenever the ad reaches people who would have bought anyway.

Sources

  1. Wikipedia: Claude C. Hopkins (accessed Nov 2026)
  2. Digiday: An oral history of the first banner ad (2014; source for the HotWired story, $10,000 pricing and 15% Yahoo click rates)
  3. Wikipedia: Web banner (accessed Nov 2026; the 44% figure as reported)
  4. Yahoo SEC filing on the Overture acquisition (2003)
  5. Wikipedia: Google Ads (accessed Nov 2026)
  6. Wikipedia: Facebook Beacon (accessed Nov 2026)
  7. CNBC: Facebook says Apple iOS privacy change will cost $10 billion this year (2022)
  8. Blake, Nosko and Tadelis: Consumer Heterogeneity and Paid Search Effectiveness, NBER / Econometrica (2015)
  9. UCLA Anderson Review: research brief on Aridor et al., ATT and e-commerce firms (2025)
  10. Google: Performance Max campaigns launch to all advertisers (2021)
  11. exchange4media: Meta aims to fully automate advertising with AI by 2026 (2025)
  12. exchange4media: dentsu-e4m report, digital to command 70% of ad spends by 2027 (2026)
  13. exchange4media: PMAR 2026 tracks India's 10-year advertising surge (2026)
  14. exchange4media: PMAR 2026, India's ad economy crosses digital milestone (2026)
performance marketingdirect responseClaude HopkinsAdWordsApp Tracking TransparencyAdvantage+Performance Max
Sapta

Sapta (Saptarshi Sarkar) is a mechanical engineer turned marketer in Bengaluru and a solo touring rider. He writes the seven Saptaverse blogs. Work with him or read more of Marketblog.

Keep reading in Marketblog

From the other blogs