← All writing

How condoms can go viral

Experience5 min

At the bottom of our basket screen there was a row of suggestions, and it had the most ordinary name a row of suggestions can have: “Consider buying this too…”

Nobody had ever given that name a second thought. Then one day somebody put a pack of diapers in their basket, and the row underneath suggested condoms.

Put in the basket
Basket3 items
  • Sourdough loaf1
  • Red wine1
  • Flowers1
Consider buying this too…
Condoms, pack of 12
Dark chocolate
Candles
logical. correct. forgotten in a second
A schematic of the basket screen, not the real app. The row title and the four baskets that triggered the suggestion are real; the other products are made up.

The contract

In 2021 the grocery app I worked at had a very large brand contract to honor. With a condom brand. The condom brand, really. Trust me, you know it, and you have used it.

The usual way to honor a contract like that is banners. A banner on the home screen, a banner in the category, a push notification nobody asked for, and at the end a report with a large number of impressions in it. Everybody signs the report and nothing has happened in the world.

We had a better slot than a banner: the row under the basket, the one place in a shop where one product gets to meet another. A paid product in that row is still a paid product, though. Somebody has to decide which baskets it turns up under, and this time it was not the algorithm’s call. It was ours.

Correct and useless

Recommendations do work. In a randomized experiment on about a million shopping sessions at a department-store site, switching them on raised sales by 11%.1 You may have heard a bigger number, that 35% of everything the biggest online shop sells comes from its recommendations. That one has been repeated since 2013, and nobody has ever shown the method behind it.2

People who build recommenders have a word for the thing ours was missing. The word is serendipity, and it is a real metric: how surprising is a suggestion that still turns out to be right. The textbook way to fail at it is a grocery app that recommends milk.3 Accurate. Useless. You were buying milk anyway.

A sponsored placement is usually the milk of recommendations. The obvious version writes itself. Wine in the basket, here are some condoms. Flowers in the basket, here are some condoms. Correct, on message, exactly what the contract asked for, and forgotten before you reach the checkout button.

So we kept those, and then went looking for the baskets where the same suggestion would be true and funny at the same time. Diapers. The morning-after pill. A greeting card that says “Will you forgive me?” Under every one of them, the same pack, under the same title.

Wrong but harmless

There is a theory of humor that I find suspiciously useful at work. It says a joke needs two things at the same moment. Something has to be a little wrong, and it has to be obviously harmless.4 Only wrong, and you have offended somebody. Only harmless, and you have made a banner.

The suggestion is always the same pack. What it appears next to:
wrongharmlessfunny
Wrong, and harmless

Wrong, because it arrives about nine months too late. Harmless, because you are buying diapers, and you have every right to laugh about it.

Where the same paid suggestion lands, depending on what it appears next to. An illustration of benign-violation theory (McGraw and Warren); where each one sits is my opinion, not a measurement.

The wine and the flowers were the contract. The overlap was the point, and every joke in it is about the product, never about the person holding the phone.

The left-hand side of that picture is where I would not go. There is a famous story about a retailer whose model worked out that a teenage girl was pregnant before her father did. The scoring model was real. The angry father, as far as anybody has been able to check, was not.5 I still would not want to be the product manager in that story.

Why people passed it on

People took screenshots. Of a basket. I have shipped features with ten times the engineering in them that nobody ever screenshotted.

Two researchers once went through nearly seven thousand newspaper articles to find out which ones readers emailed to each other. Sad stories traveled badly. Stories that wound people up traveled well. A noticeably angrier article (one standard deviation, if you want the unit) was 34% more likely to reach the most-emailed list. A more awe-inspiring one, 30%.6 What gets passed on is whatever raises the pulse.

Amusement raises it too, and they tested that in a lab. With sausages.

Two adverts for the same sausages. Which one would you send to a friend?

Pick one to see what the experiment found.

From Berger and Milkman’s 2012 paper on what gets shared. The sausage brand is real. I have left the name out, as I do.

Humor has a price, and I should say what it is. Across decades of studies it buys attention and liking, and it costs credibility: the funnier the source, the less seriously it gets taken.7 For a bank that is a problem. For a pack of condoms in a grocery basket it is a bargain.

Consider buying this too

The part I like best is that the funniest line in the whole thing was one we never wrote. It had been sitting on the basket screen all along, three dots included, waiting for the right product to stand under it.

People have been joking around this particular product for as long as it has existed. Casanova blew his up like balloons to check them for holes. In the Depression, Americans got through a million and a half a day, on the reasoning that condoms were cheaper than children.8

That is how condoms go viral. You put the paid product where it tells the truth, and you let the customer get to the joke first.

Sources and further reading
  1. Kumar and Hosanagar, randomized experiment on product recommendations, Information Systems Research. Summary from the University of Florida
  2. The 2013 consultancy article that the “35% of sales” figure is usually traced to. No methodology is given
  3. Recommender systems: serendipity as an evaluation measure, and the grocery shop that recommends milk
  4. Benign-violation theory, McGraw and Warren
  5. KDnuggets, 2014, on what is and is not substantiated in the pregnancy-prediction story
  6. Berger and Milkman, “What Makes Online Content Viral?,” Journal of Marketing Research, 2012
  7. Eisend, “A meta-analysis of humor in advertising,” Journal of the Academy of Marketing Science, 2009
  8. History of condoms: Casanova, and Depression-era sales
  9. Wikipedia’s entry for the world’s largest condom manufacturer, on its share of global production
  10. Marketing Dive on a 2022 survey of more than 12,000 people about humor and brands. A vendor survey
  11. Kantar, “How to get humour right in advertising,” February 2022