Module 6 · Mobile / F2P (2012–2018)

The core loop and retention

A F2P game lives not off a sale but off retention: you monetize a player over months, not once. So the master metric is not the install and not even the first purchase but retention, and it is engineered by the "action → reward → progress" loop plus the pacing mechanics that gate that loop.

~17 min🎲 design + 📊 metrics

The gist in 30 seconds

In F2P there is no sale — there is a relationship: a player's value is spread over months, so everything rests on retention (the share who come back). The engine is the core loop: a short "action → reward → progress → again" cycle that is both pleasant in itself and a reason to return. It is measured as a retention curve at the D1 / D7 / D30 checkpoints; the reality of 2024 is harsh — the median is ≈ D1 23% · D7 4% · D30 <1%, where 40/20/10 used to count as "good", and with traffic costs rising the target is now D1 50%+. The loop is deliberately gated by pacing mechanics (energy/lives, cooldowns, appointment mechanics): out of energy → wait or pay — that is the monetization hook. Sessions are short for mobile (20–30 min), and a short loop = more sessions = more monetization points. And the math is brutal: LTV ∝ the area under the retention curve, so a tiny gain in retention doubles the revenue. It is a leaky bucket: you pour traffic in the top (UA) and it drains out the bottom (churn) — retention decides how big the holes are.

The mechanism: the loop, the curve and the hook

F2P inverts the model: a game used to be a product (sell a copy, move on to the next), and now a game is a service (retain the player and monetize over time). Three connected mechanisms.

The core loop — what the player repeats

The core loop is the short sequence of actions a player runs over and over. An example (Clash of Clans): log in → attack somebody's base (3–5 min) → wait for troops to respawn (2–4 h) → collect resources → start an upgrade (hours) → log out → come back in 2–4 h to repeat. The loop has to deliver both pleasure (that "30 seconds of fun") and a reason to return (unfinished progress). It is the game loop raised to the level of habit design: trigger → action → variable reward → investment (Nir Eyal's hook model).

The retention curve — the whole business model on one chart

Retention $R (t)$ is the share of installers still playing on day $t$ . It falls fast and then flattens (a "core" forms). The canonical three checkpoints:

Checkpoint	Median 2024	Top 25%	"Good" (the old target)
D1 (came back the next day)	~23%	31–33% (iOS)	40% → now the target is 50%+
D7 (a week later)	~4%	7–8%	20%
D30 (a month later)	<1%	~3%	10%

You can see the chasm between the dream (40/20/10) and reality (23/4/<1): 75% of projects have under 3% left at D28. D1 is the north star: if the player did not come back the next day, the rest of the funnel does not matter.

Why retention = the business

The link to money is direct. A player's "lifetime" is the area under the retention curve; for a geometric tail (a daily return rate $r$ ) it has a closed form:

L \approx \frac{1}{1 - r} and LTV = ARPDAU \cdot L

The nonlinearity is cruel: $r = 0.9$ → $L = 10$ days; raise retention to $r = 0.95$ → $L = 20$ — lifetime and LTV have doubled off 5 points of retention. And since UA only pays off when $LTV > CPI$ (F2P economics), that shift turns loss-making traffic buying into profitable traffic buying. That is why retention is the master metric and the install is not.

Pacing mechanics and the monetization hook

The loop is deliberately gated to create a wait that can be sold:

Energy / lives: play N times and you are out — wait or pay. Candy Crush — 5 lives, +1 every 30 min; Clash — troops take hours to respawn.
Cooldowns: an action on recharge (LoL abilities 5–60 s; building upgrades — hours to days).
Appointment mechanics: come back at a specific time for a reward (daily logins, a shop rotating every 24 h) — they build habit through a scheduled trigger.

The hook: out of energy → I want to keep going → I pay gems for a refill. The wait is designed, not accidental — that is the monetization engine of F2P. Sessions are short (20–30 min on mobile): a short loop → 10+ sessions a day → more touchpoints and a stronger habit.

🕹 Games to play — and what to notice

You can feel the pacing gates and the hook in a couple of minutes — look for the "wait or pay" wall and the "speed it up for currency" button.

Candy Crush Saga 2012 · the lives system

The reference implementation of energy-as-a-gate: 5 lives, lose a level and you lose a life, run out and you hit a wall. Recovery is 1 life / 30 min, or pay, or ask friends. A "level in 2–5 min" loop is perfect for a short mobile session.

🎮 Play: lose 5 times in a row and hit the lives wall — notice the "wait 30 min / pay / ask a friend" screen. That is not a progression bug, it is the monetization point, built into the core loop.

Clash of Clans 2012 · timers and appointments

A loop built on timers: an attack takes minutes, a building upgrade hours to days, troops respawn slowly. That stretches a session across a whole day and creates the "come back in the evening to check" appointment mechanic that daily returns rest on.

🎮 Play: start an upgrade and notice the "finish instantly for gems" button. Count how many times a day you feel pulled to come back and check timers — that is a designed return rhythm and a direct driver of D1/D7.

Hearthstone / any game with dailies appointments through quests

Daily quests ("play 3 matches", "deal damage") give you a reason to log in today specifically and a direction instead of aimless grinding. A classic appointment trigger, and it lifts D7/D30.

🎮 Play: log in and look at the dailies — notice how gently they dictate "come back tomorrow, don't lose your streak". Compare with a hypercasual game (a 30-second loop, zero gates): there retention is held by frequency rather than timers.

Deep end · metrics: the retention curve, the "leaky bucket" and why D1 decidesskippable

The F2P growth model is a leaky bucket: the inflow of installs (UA) minus the outflow (churn). At equilibrium DAU settles where inflow = outflow, and the level is determined entirely by the shape of $R (t)$ .

The shape of the curve

Retention is not an exponential with a single λ: the early days fall steeply (the curious, who did not click with it), and then the curve flattens — a "core" of loyal players forms with a nearly constant daily return rate $r$ . It is often modeled with a mixture or a power law, but for a lifetime estimate you take the tail $r$ and the closed form $L \approx 1 / (1 - r)$ .

Why D1 is the north star

D1 is the earliest and most predictive signal: it multiplies the whole subsequent curve (someone who did not come back the next day will not be there at D7 either). Raising D1 = raising the whole curve = multiplying lifetime and LTV. Which is why product teams are obsessed with the first few minutes (onboarding, time-to-fun, a first reward in the first seconds): half the churn happens in D0–D1.

Cohorts, not averages

Retention is always read by cohort (grouped by install date/source), otherwise fresh inflow masks the churn of the old. "Average DAU" lies; "D7 for the March 1st cohort" does not. This is the direct bridge to analytics.

Deep end · design: session length, the hook model and the ethical lineskippable

Loop length is chosen for the player's context: mobile — 20–30 min (a commute, a queue), console — 60–90 (an evening). A short loop gives more sessions/day and more monetization points, but too short is boring and too long does not fit into a break.

The hook model (Eyal)

Habit is built by the loop trigger → action → variable reward → investment. A variable reward (as in gacha) hooks harder than a fixed one; investment (you upgraded the base, you completed the collection) raises the cost of leaving. Appointment mechanics are an external, scheduled trigger.

Where the line runs

The same levers that raise retention (FOMO, streaks, variable rewards, a designed wait) slide easily into dark patterns and the exploitation of vulnerable people. Engineering honesty means distinguishing "I helped the player return to something they themselves value" from "I manufactured anxiety/dependency for a metric". The same argument returns in recommenders (below, 🔁).

Analogy

A F2P game is a leaky bucket you keep pouring water into (installs from UA). The value of the bucket is decided not by how fast you pour (traffic buying) but by the size of the holes (churn). Retention is patching the holes: a bucket with tiny holes (high retention) keeps the water and accumulates it, while a sieve (low retention) can never be filled no matter how much you pour. The core loop and pacing mechanics are the walls of the bucket: the loop gives a reason to stay, the gates decide when you come back and where you pay.

Why it matters

F2P changes the master metric: not the install, not the first purchase, but retention — because you are monetizing a relationship that lasts months. The core loop and pacing mechanics engineer that retention, and the math of lifetime makes tiny gains in retention enormous in LTV. Lose retention and neither traffic nor monetization will save you (the leaky bucket). The same logic — optimize the long retention loop rather than a one-off conversion — governs any service product you build.

🔁 Beyond games — where this transfers

The lesson is retention as the master metric of a service: an engagement loop, a retention curve and churn instead of a one-off conversion.

ML / AI (your domain): retention/engagement is the objective function any recommender optimizes (DAU, dwell time, returns as a reward signal). The "leaky bucket" = churn prediction / survival analysis (predicting who will leave is a canonical ML task; that same area under the curve = expected lifetime). The hook model with its variable reward is exactly the RL/dopamine dynamic that ML feeds exploit — and right here is the feedback loop and engagement maximization problem: optimizing raw engagement leads to the same dark patterns, and knowing where NOT to optimize retention directly is part of professional ethics (long-term value / user wellbeing as the real goal). Cohort curves are the standard way to evaluate the rollout of any model or feature.

Product / growth: retention > acquisition (the leaky bucket is a law of SaaS growth); D1/D7/D30, cohort analysis, the "aha moment" in onboarding; LTV > CAC is the same thing as LTV > CPI.

Behavioral design: the hook model, triggers, variable rewards — and the red line between a useful habit and exploitation.

The principle: in a service, value is created by repeat returns, not by first contact. Measure and engineer the retention loop — but distinguish helping a user from exploiting their vulnerabilities.

🔧 Run it and poke at it — on your home machine

What to play is above (🕹). This part is about taking the loop apart and doing its math.

🔧 Poke at it (deconstruction) ~40 min

Take any F2P game and draw its core loop with arrows (action→reward→progress→gate). Find every pacing gate (energy/cooldown/appointment) and, for each, the matching monetization button. Then estimate the lifetime: assume a daily return rate

r

and compute

L \approx 1 / (1 - r)

for

r

= 0.85, 0.9, 0.95 — you will see the nonlinear jump in lifetime.

🧪 Test it (with producer eyes) ~15 min

For an imaginary game: which is the earliest onboarding moment that would raise D1? (time-to-fun, a first reward in the first 30 seconds). Then an ethical audit: which of your pacing mechanics "help the player return to something valuable" and which "manufacture anxiety for a metric"?

Checklist: drew the core loop with every gate; matched each gate to a monetization point; computed L(r) and saw the nonlinearity; found the D1 lever in onboarding; sorted the mechanics ethically.

Connections

foundation

F2P economics — the money side (LTV/ARPDAU/whales); this lesson is the engagement side that produces that revenue.

LTV = ARPDAU \cdot L

ties them together.

foundation

Distribution — UA pours water into the bucket; retention decides whether the bucket holds it. Acquisition without retention is a sieve.

Gacha and the battle pass — how to monetize a retained loop with variable rewards and seasonal FOMO.

Analytics — how to measure retention honestly (cohorts, funnels, A/B).

Questions worth asking

Why is D1 more important than D7 and D30, if the money comes from long-term players?

Because D1 multiplies the whole curve: it is the earliest predictor and the cheapest lever. If they did not come back the next day they almost certainly will not be there at D7/D30, so the entire long tail is zeroed out at the start. Raising D1 by 5 points shifts the whole curve up and therefore scales lifetime and LTV more than fiddling with D30 ever could. On top of that, half the churn happens in D0–D1 (onboarding, time-to-fun) — that is where the leverage is.

Why retention as the master metric rather than revenue or installs?

Installs are water you bought; without retention it drains away and UA burns money for nothing (the leaky bucket). Revenue is a consequence:

LTV = ARPDAU \cdot L

, and

L

is precisely the area under the retention curve. Retention sits causally upstream of both installs (does UA pay back) and revenue (lifetime). That is why people are obsessed with it: it is the lever that moves everything else.

Is a designed wait (energy/timers) manipulation?

It depends on what it does to the player. Pacing in itself is neutral: it protects against burnout and gives structure (like cooldowns in design). It becomes manipulation when it manufactures anxiety or dependency and sells the relief (aggressive timers plus a paid skip plus FOMO), especially aimed at vulnerable people. The honest line: "it helps them return to what the player values" versus "it creates pain in order to sell the painkiller". Same levers, different intent and effect.

Why does the retention curve flatten instead of falling to zero?

Because the population is heterogeneous: the early days drain off the "curious" who did not click with it (a steep drop), but a core of loyal players remains with a nearly constant daily return rate. That core churns slowly and steadily, so the tail of the curve is close to geometric with a high

r

rather than a rapidly decaying exponential. It is precisely that core's lifetime (

L \approx 1 / (1 - r)

) that feeds the business.

Can a great game have bad retention — and the other way round?

Yes, and it is a common trap. A single-player game that is "great" by quality can have terrible retention, because it gives no reason to come back daily (you finish it and leave) — and it should not be playing the F2P metrics game at all. Conversely, a game that is mediocre in depth but has a perfectly tuned loop and appointment mechanics holds D30 better. Retention measures not "is the game good" but "is the return loop designed" — that is about the service model, not artistic value. Confusing the two means dragging F2P metrics somewhere they do not belong.