You Died. The Game Took Notes.

Imagine you are one move away from finishing a level. You have already spent ten minutes on it. The board looks beatable. You know exactly what you would do next.

Unfortunately, you are out of moves.

Fortunately, the game has prepared a small financial presentation.

Buy a few more moves. Wait for another life. Maybe watch an ad and get another chance. The details vary, but the offer arrives when the thing being sold feels unusually useful.

That timing is the interesting part. You opened the app to play a game. Now you are deciding how much the next attempt is worth.

I wanted to understand how much of this comes from ordinary game design, how much comes from psychology, and how much is an optimization system quietly doing its job. The best example I found involves a cartoon jetpack and a model that takes into account how you died.

But first, Pokémon. Because the business gets much easier once you care about something.

Pokémon made your friends part of the game

The original Pokémon games gave you several reasons to keep playing at once. Catch something new. Raise something you already have. Build a better team. Beat the next Gym. Fill another slot in the Pokédex.

Those activities fed each other. Catching a creature expanded your options in battle. Battling made your team stronger. Progress opened more of the world. The collection had a purpose beyond making a number go up.

Then there was the other kid with a different cartridge.

Nintendo's description of Pokémon Red makes the arrangement explicit. Red and Blue did not each contain every Pokémon. Completing the collection required trading.

Your friend's game contained something yours did not. Their participation increased what you could do with your own copy. That is a pretty effective reason to talk about a game at lunch.

It also gave the collection a personal dimension. You had a team, favorites, and creatures you had spent time raising. An unofficial English translation of a 1996 developer interview describes the thinking behind nicknames, original-trainer records, and making creatures feel alive as they moved between Game Boys. Attachment was part of the design.

The early success was not a perfectly executed spreadsheet, either. In Nintendo's retrospective interview, the creators describe a quiet launch followed by growing word of mouth, magazine coverage, and excitement around the mysterious Mew. An April 1996 giveaway offering Mew to 20 winners attracted roughly 78,000 entries. Mew's appearance through a bug had not been planned.

Sometimes the growth strategy is an undocumented Pokémon.

The cards and television series added more ways to encounter the same characters. You could care about Pikachu before understanding a single battle mechanic.

My read is that Pokémon's appeal came from these things working together. Collection, mastery, attachment, friends, and a world that followed you outside the cartridge. Reducing all of that to a reward loop misses most of what made the reward worth wanting.

The unfinished thing does a lot of work

A nearly complete collection has a different feel from an empty one. So does a level with one obstacle left.

There is a useful experiment here, and it involves the significantly less exciting world of car washes. Researchers gave customers either an eight-stamp loyalty card or a ten-stamp card with two stamps already filled. Both groups needed eight more purchases to earn the reward.

The work was identical. The sense of progress was not. In the 2006 experiment, 34% of customers with the head start redeemed the reward, compared with 19% of customers starting from an empty card.

That does not tell us what a Pokémon progress bar does to sales. It does demonstrate that how progress is presented can change behavior, even when the remaining effort stays the same.

Now put a purchase next to the unfinished task.

King's explanation of Candy Crush lives describes five lives, losing a life after failing a level, and the option to purchase extra moves. A depleted life replenishes after 30 minutes. Players can also buy lives or ask friends for help.

Time, money, and social assistance become different routes back to the same activity.

Extra moves are especially interesting because they offer a chance to finish the attempt you are already invested in. The product is not just five abstract moves. It is five moves on this board, after this much effort, while the solution still feels within reach.

That is a much better sales pitch than putting a pack of moves in a shop and hoping you develop an interest in inventory management.

The offer knows how you died

Here is where the story moves from interpreting game design to a documented engineering implementation.

In a February 2022 technical account, Google described working with Halfbrick on Jetpack Joyride. The game offered digital rewards in exchange for watching a video ad. Previously, the reward was selected randomly.

The new model used information about the player's skill and current session, including how they died and their chosen powerups. Halfbrick collected 129 training signals, and the trained model ran on the device. Google reported a 36% increase in interactions with the rewarded-video offer.

Google's reported 36% lift was relative, measured in offer interactions rather than revenue, and the public account leaves some experimental details out.

The interesting mechanism is clear, though. Instead of merely deciding when to interrupt, the game could select a reward that made the interruption more attractive.

You see something useful after a failed run. The system sees another response to an offer it can measure.

A separate Halfbrick case study describes personalizing ad frequency and the time players wait to unlock levels. It reports 16% higher average revenue per daily active user, with no reported change in retention, engagement, or ratings. That is a different intervention and a different metric.

This is also why claims about games secretly making your next level impossible need better evidence than a frustrating afternoon. The Jetpack example documents selecting rewards. It does not establish individualized difficulty manipulation.

Choosing the right thing to offer can be commercially useful all by itself.

Your free game has more than one customer

It is tempting to call all of this optimizing ad clicks. That collapses several different transactions.

The player wants something inside the game. The publisher wants revenue. The advertiser wants someone who might install an app, buy something, or take another valuable action.

An ad impression means an ad was shown. A click means someone clicked or tapped the ad. Watching a rewarded video can earn an in-game reward without requiring a purchase from the advertiser. Google explicitly prohibits incentivizing clicks on advertiser links, while rewarded ads are an established format.

The distinction matters because the publisher can make money from someone who never buys anything in the game. That person still brings attention that can be sold.

More advertising is not automatically better, either. Consider some invented numbers. Ten thousand daily players seeing four ads each produce 40,000 impressions. At a publisher eCPM of $10, meaning $10 earned per thousand impressions, that is $400 a day.

Increase the average to five ads and the same arithmetic gives $500. But if the interruptions eventually drive the audience down to 7,000 players, those five ads produce $350.

These are hypothetical inputs, not market rates. They illustrate why a business might care about your return tomorrow as well as your response today.

The fifth ad was very productive right up until you uninstalled the game.

The model does not need to understand your childhood

Some of this works with ordinary rules. A developer can treat purchasers differently from people who have never spent money, or wait until someone finishes the tutorial before showing ads.

Firebase's hybrid monetization tutorial explicitly walks through creating a purchaser audience and assigning it a different ad experience.

Individual personalization goes further. In Firebase's documented system, developers supply alternative experiences and an objective to optimize. That objective can involve impressions, clicks, purchases, or revenue. The system chooses among the alternatives for users and compares results against a holdout group.

Conceptually, the loop is straightforward:

Observe what happened
Choose an offer or experience
Measure the response
Use the result to inform later choices

An A/B test compares experiences across groups. A purchaser rule assigns people to a category. A model can make an individual choice using available signals. Those are different amounts of personalization, and none requires a complete psychological biography.

The developer chooses what counts as success. The system then gets feedback about that choice.

That should sound familiar to anyone who has spent time with production metrics. A dashboard can tell you that a number improved while leaving the more interesting question unanswered. Was that the number you actually wanted to improve?

Same monsters, different checkout

The original Pokémon cartridges sold an adventure. Later mobile games attached recurring purchases to some of the same reasons people cared about that adventure.

In Pokémon GO, extra incubators let you hatch multiple eggs at once. The contents are uncertain, and hatching requires walking. Buying more capacity means more opportunities progressing simultaneously.

For a concrete example, Adventure Week in August 2024 combined specified Shiny Pokémon opportunities and an egg pool with a $2 Timed Research ticket offering incubators and encounters. Tasks had to be completed and rewards claimed before the deadline.

You have a collection goal, more ways to pursue it, and a clock.

Pokémon TCG Pocket makes the relationship even easier to see. Its 2024 launch announcement described one pack's stamina accumulating every 12 hours, allowing two free packs a day. Purchasable Poké gold can reduce the wait.

The timer is predictable. The cards are not. Paying buys more opportunities sooner.

None of that requires the player to stop enjoying the cards. Enjoyment is precisely what gives the next pack value. The commercial question is how much of the experience is organized around helping someone enjoy the collection, and how much is organized around making the next wait feel inconvenient.

Fun and pressure can fit on the same screen

Research on game motivation connects enjoyment with competence, autonomy, and social connection. Getting better, making meaningful choices, and playing with friends are real reasons to return. Dopamine research helps explain learning from unexpected rewards, but it does not turn every rare card into a measurable unit of addiction.

The same feature can also land differently for different people. In a 2022 survey of 178 players, engagement features elicited reports of motivation alongside chores, obligation, and fear of missing out. A daily challenge can be a pleasant routine. It can also become another thing you have to finish before bed.

Clinical gaming disorder involves impaired control and substantial harm to everyday functioning. Hours played and dollars spent cannot establish that on their own.

Consent is another boundary worth keeping concrete. In 2023, the FTC finalized a $245 million refund order against Epic Games to settle charges involving dark patterns and unauthorized purchases. The case concerned purchase interfaces and charges, including confusing button arrangements. It gives us something more specific to examine than whether a game was too entertaining.

For people who build software, the practical questions are familiar. What behavior are we optimizing? What choices does the interface make clear? What happens when a user wants to stop? Which costs never appear in the success metric?

Back at the almost-finished level, the extra moves may be worth the price to you. The reward for watching a video may be useful. The game may be fun.

You can recognize all of that and still notice the second system running around the first. One tracks your progress toward winning. The other measures which offers you accept along the way.

You were paying attention to the game. The game was taking notes.