“Did it work?” The 100-year-old question marketing still can’t answer

Our new documentary traces the history of marketing measurement, from TV meters to click-tracking to AI agents that move budgets thousands of times a second. Every era thought it had found the wasted half of its ad spend. Every era was wrong in a way it couldn’t yet see.

“Half the money I spend on advertising is wasted. The trouble is, I don’t know which half.”

Every marketer knows the line. Almost nobody can say who actually said it. It’s pinned on two men, the Philadelphia merchant John Wanamaker and the British soap magnate William Lever, and there’s no hard proof either one did.

The attribution is murky, but the problem is real. For more than a century, marketing has been hunting for that missing half. Each generation built new tools, found new numbers, and declared the problem solved. That’s the story we set out to tell in Did It Work?, and this post walks through its main turns.

Truth at low resolution

For most of the 20th century, measuring marketing meant a few thousand households with a meter wired to the television, interviewers with clipboards on front doorsteps, and statisticians modelling last year’s sales to estimate what the advertising had done.

It was slow, expensive, and imprecise. You learned roughly what worked months after it had already worked.

It did measure the thing that mattered most, though: whether people remembered you, trusted you, and chose you when it counted. It measured the brand. The picture was blurry, but it was the whole picture.

That trade-off of completeness over precision was about to flip.

2005: everything becomes countable

In 2005, Google made its analytics product free. Almost overnight, every click, every visit, and every customer had a number attached to it.

Marketers no longer had to wait months or rely on samples of a few thousand people. They could watch activity happen in real time, down to the individual.

One simple rule quietly took over the industry: whatever someone clicked last, right before they bought, got all the credit. Last-click attribution was easy to understand and easy to report, and it came with the internet at no extra cost.

The money followed the numbers. Budgets drained out of brand-building and into “performance,” the work you could count to two decimal places. Social platforms, programmatic buying, and the whole ad-tech machine were optimised for the click. It looked like the wasted half had finally been found.

The precision that lied

There was one problem. The numbers were lying.

Last-click gave all the credit to the final step, usually a search ad or a retargeting ad, even when the customer had already made up their mind. It’s like crediting the rooster for the sunrise. The rooster crows, the sun comes up, and the rooster takes the bow.

A few landmark cases made the gap impossible to ignore.

eBay switched its brand search ads off. In research published in 2013, eBay ran the experiment nobody selling ads wanted to see: it turned off its paid brand-search ads and watched what happened. Sales barely moved. Nearly everyone who had clicked those ads would have bought anyway, arriving through organic results instead. eBay had been paying to reach customers it already had.

P&G cut $200 million in digital spend. Procter & Gamble, one of the largest advertisers in the world, pulled roughly $200 million out of digital advertising. Sales didn’t drop.

JPMorgan Chase cut its ad footprint by 99%. The bank went from showing ads on about 400,000 websites to around 5,000. The results were essentially identical.

Part of the explanation was structural. The companies selling the ads were also the ones grading them, and they tended to give themselves very good marks.

For a decade, the industry had measured activity (clicks, impressions, visits) and called it impact. Meanwhile the thing it had stopped measuring, the brand, was quietly starving. Marketers could see everything except whether any of it was working.

The ground disappears

Then the foundation the whole system stood on began to give way.

Europe’s GDPR raised the bar for collecting personal data. In 2021, Apple began asking iPhone users whether apps could track them across other apps and websites, and most people said no. Browsers restricted third-party cookies, the small files that had made cross-site tracking possible.

The perfect, personal, real-time measurement that had defined digital marketing for fifteen years stopped working as advertised.

History rhymes

So the industry did something unexpected. It went backwards.

It revived the statistical models from the panel era and rebuilt them for modern data. Big platforms even released marketing mix modeling tools as open source. And it started running real experiments again: incrementality tests that hold advertising back in a few places and measure what actually changes compared with a control group.

The methods the industry had set aside in 2005 came back because they answer the one question clicks never could: what actually made a difference?

The machines take the wheel

That brings us to now, and to something genuinely new.

Measurement is increasingly not for humans at all. More and more, nobody is reading the dashboard. Algorithms don’t just measure the marketing; they decide it. They set budgets, pick audiences, write ads, and move money, thousands of times a second.

The marketer of the past asked “did it work?” and adjusted next quarter. An AI agent asks the same question a thousand times a minute and acts on the answer before you’ve finished reading it.

Here’s the uncomfortable part. If the number is wrong, if we’re still measuring activity and calling it impact, the machine has no way of knowing. It just optimises, faster and faster, toward the wrong target, with no one in the room to catch it.

The last-click mistake cost the industry a decade. An agent running on the same mistake can compound it at machine speed.

What’s next?

A hundred years ago, a merchant admitted he didn’t know which half of his money was wasted. Today, our machines answer that question instantly and confidently, millions of times a day.

The question was never really about the numbers. It’s about whether we can trust what they tell us. We’ve never been more certain, and we’ve never had less time to check.

That’s why the input matters more than ever. When AI is making the decisions, it needs to be optimising against causal truth (what the marketing actually caused) rather than what a platform says it touched. Incrementality testing, marketing mix modeling, and calibrated attribution are how you give it that truth.

Watch the full documentary, “Did It Work?” →

Explore Lifesight’s marketing measurement platform →