Are there still bosses willing to spend this kind of money?

I

Advertising has only two problems to solve.

First, whether people are willing to keep watching. An ad's rival has never been the competition; it's the remote, the mouse, the thumb that could swipe up at any moment. No one is obliged to watch you.

Second, once they've watched it, whether you've given them a message, a value, a belief, an identity bigger than the product itself.

The first is the floor, the second the ceiling. Get only the first right and the audience laughs and forgets; the brand banks nothing. That's sensationalism. Get only the second right and no one sits through the message, however grand—that's a public service announcement. Plenty of Taiwan's award winners died there.

Spots that do both: you won't see more than a handful a year.

First, watch the commercials we made back then, then read on:satsumacreative.tw/cf

II

For a product like online games, the second problem is especially hard.

Most of what's on the market comes from the same batch of licensed titles, and the differences in gameplay can neither be shown nor proven in thirty seconds. You say the graphics are gorgeous; so does the game next door. And "fun" is something you only find out after you buy. Players have been burned too many times; a claim like that carries roughly zero persuasive weight, like a realtor calling a place cozy.

Worse, games are a social good. Playing an online game alone means nothing; it only means something when there are people to curse with, fight alongside, argue about in class. So what the ad has to manufacture is "everybody knows"—I know you know, you know I know, so tomorrow someone will bring it up at the office.

The product can't supply that on its own. So you borrow.

There are always symbols in society already fully charged—suppressed, ruled unfit for polite company, untouched on television because no one dared. Plug the product into them and the current runs through the brand.

A nun shoulders a sniper rifle, reciting the Lord's Prayer. Two high-school girls kiss in a TV commercial. A folk-religion god of war paired with the line "play, and be blessed." Japanese-style epic visuals over dialogue thick with Taiwanese slang.

What these share is obvious: in the Taiwan of that time, all of them had been judged unfit for the living-room TV. Religious solemnity mustn't be profaned, same-sex intimacy mustn't be seen, folk religion was superstition to the educated class, Taiwanese wasn't refined enough. An ad can't afford to generate its own message, so it goes and wins thirty seconds of legitimacy for these things instead, trading their voltage for its own attention.

It's arbitrage.

III

For arbitrage to work, there has to be a time lag.

Taiwanese audiences, through Japanese dramas, anime, and magazines, consumed the symbols circulating outside in real time, with almost no gap. But the people making the decisions weren't the same crowd—the client's marketing director, the media buyer, the network's censors, the regulator were the audience's parents' generation.

Audience: year zero. Client: ten years behind. Regulator: five more.

The lag lived in the decision-makers, not in the audience. You put something on the table the audience had long since understood; the client found it daring, the regulator found it over the line, the audience felt someone had finally done it. Much of the so-called boldness of Taiwanese advertising in those years was just feeding on that gap.

Set Japan alongside it and there's one more difference. Japan's market is big enough to segment: a commercial can speak to a small group, so its symbols can be fine-grained, ambiguous, understated, can go a full sixty seconds without mentioning the product. Taiwan's market is too small; any ad that goes on TV has to hit every demographic at once, so the symbol has to be coarse, loud, readable in a second.

The Taiwanese version is always blunter than the foreign original. That's the denominator, not taste.

IV

Borrowed current has two poles. That's the risk clause in the arbitrage.

A symbol suppressed for that many years has accumulated both humiliation and power; both poles were always there. When you borrow, you only get to choose which symbol. Which pole the audience latches onto the moment it lands, the market decides.

One spot illustrates this. The first half is a pink room, plush toys, a full wall of pink curtains, a girl sitting on an electric rocking horse, crying. The man in frame is an overexposed, nearly faceless back—not a character, just the act of leaving. At second nine she transforms: hair flies up, white light bursts. After the transformation the rocking horse becomes a real white horse, the toys become a polearm, and those men shrink into set dressing—headbands on, pumping fists on the sidelines.

Fake mount swapped for real mount. The shift in position is done with props.

In the structure, that rocking horse stood for powerlessness. In circulation it was read as rocking. Same prop, two poles; the market chose the latter, and kept choosing it for over a decade.

A twenty-second spot with the transformation at second nine: eight seconds of setup, eleven of spectacle, and the weak half placed up front, so by the time the audience's attention actually catches, what they're seeing is already the spectacle. A reversal needs a "before" to work, but reposts, parodies, and news clips only carry off the ending; setup doesn't travel. Before and after also share the same palette, so the audience never receives the signal "this is another world"—the reversal's legibility was undercut by its own design.

And one more layer, more fundamental: that spot carried a message and a talking point at the same time, and the talking point cost almost nothing to decode while the message cost a great deal. Both competed for the same attention. The fast one won, ten to one.

I was on set for all of these. There was a price, and I wasn't the one who paid it; I'm in no position to close the case on behalf of those who did.

V

That entire method rested on one condition—a single spot could reach all of Taiwan on the same night.

A shared topic is a collective state; it requires that everyone know everyone else saw it too. In the era of terrestrial plus cable, buying enough airtime for one commercial could manufacture that state. That's why advertising could carry a message: a message needs time to ferment, needs someone to bring it up the next day, needs to pass through a round of different mouths before it sets. Concentrated media provided that fermentation tank.

VI

The first step in the window closing: results became attributable.

Once mobile games took off and social ad platforms matured, every dollar spent showed how many installs, registrations, retained users, paying players. For the first time in a hundred years, advertising could prove itself.

Once things are attributable, budget flows entirely toward what can be attributed. And a message can't be attributed—its effect is that three months later someone still remembers your name. There's no field for that; if it won't fit in the report, it doesn't exist.

So in pitch meetings, "what is this spot saying" got replaced by "what's this spot's cost per acquisition."

Advertising retreated to numbers: number one in Taiwan, a million players, top download. That's the most primitive spec-sheet advertising, only the specs are now social proof, and herd pressure stands in for persuasion.

The industry's roles shifted with it. Agencies went from pitching to order-taking, creative went from proposition to asset, one film became thirty thumbnail variants for testing. No one did anything wrong; everyone was answering to their own report.

VII

Second step: distribution moved from the networks to the algorithm.

Three things happened at once.

Concentration vanished. Everyone's feed is different; no spot can reach everyone simultaneously. The infrastructure for a shared topic was torn out.

The yardstick was swapped. Whether a film gets seen depends on its three-second completion rate, not on what it says. The machine reads signal strength; it can't read a message. The judge of an ad used to be the audience's memory three months out; now it's the machine's verdict in three seconds.

So the talking point beats the message across the board. A message has to unfold, be watched through, settle; a talking point can be clipped, reposted, and scored as high engagement in a second. Both compete for the same attention, and the fast one always wins.

Even the numbers fail here. "Number one in Taiwan" means nothing in a fragmented environment, because "all of Taiwan" is no longer a unit.

VIII

So how does someone who makes ads keep making ads.

Now there's one more variable. AI can generate a thousand lines of copy in ten seconds, pair them with a thousand thumbnails, deploy, test, and keep the best-performing one automatically. The marginal cost of copy has gone to zero, and the version the machine picks is usually genuinely better than the one a person would.

The machine tests which line converts. It never asks what the spot is saying. It finds the optimum inside the frame you hand it; where the frame comes from, it can't answer.

And now you don't even have to shoot.

Footage used to be the barrier. Casting, studio, lighting, editing, score—a spot cost at least three weeks and a serious sum, so it was an expensive bet. Because the bet was expensive, you had to get it right, which is why in the pitch someone had to answer "what is this spot saying"—being wrong was costly. Once the cost of generation goes to zero, that constraint disappears with it. Generate a thousand different films at once, push them all out, sort it out after testing. Getting it right no longer has value, because there's no longer a bet.

When will audiences find this normal? Sooner than most people think, and the criterion has nothing to do with image quality. The tipping point for a medium's acceptance is the moment it stops being discussed. Today's AI ads still get called out—"this was made with AI"—so they haven't crossed over yet. When no one mentions it anymore, normalization is complete.

This road has been walked many times. Models in print ads haven't had real skin in years; no one protests. Movie explosions, TV skies, dubbing, library music, keyed composites—all the same. Fake was never rejected; only fake you could see was. Once the seams drop below the threshold of attention, it's over.

Something else disappears with it. A high-production ad used to be a signal in itself: this company can afford this, so it's serious. Production scale was a guarantee; the audience might not articulate it, but they could see it. Once generation costs nothing, everything looks like a blockbuster, and looking like a blockbuster stops meaning anything.

Then the loop closes. Machine generates, machine deploys, machine scores, machine regenerates—no human needed anywhere in the cycle. The loop optimizes for three-second completion, and it will converge on whatever is most stimulating and most empty. What the audience ends up conditioned to accept as normal may not just be "a machine made this spot," but "every spot is optimizing against me in real time."

So only one job remains: make a proposition. Decide which symbol to borrow, whom to win those thirty seconds for, where to take the audience. This can't be A/B tested, because before it's made it doesn't exist; there's nothing to compare. Testing only ranks existing options. Someone has to bring the options into being.

But a proposition comes with a condition, and the condition is harder than the proposition itself: you have to convince the person paying.

A nun with a sniper rifle, two girls kissing on TV—at the moment of the pitch, none of this had data behind it, no precedent to cite. You can't offer a guarantee, only a judgment, and then someone has to be willing to put budget behind that judgment.

It's many times harder now, because the other side has a dashboard. Clients used to have no choice but to gamble; now they can point at the numbers and ask why you're not running the version with the higher conversion rate. Defending something unmeasurable in front of a dashboard is far harder than it was over a decade ago.

My luck was in the bosses who let me spend their money. Those spots got made because, at the end of the pitch, someone said yes. Leading the audience presupposes that someone first lets you onto the field.

The two standards from the start haven't changed—keep people watching, and give them a message, a value, a belief, an identity bigger than the product. What's changed is that the second one now has to survive a meeting first.

Are there still bosses willing to spend this kind of money?