Game analytics glossary

Game analytics has a habit of dressing simple ideas in intimidating acronyms. Here's every term you're likely to meet in PlayGenus — or in any analytics conversation — in plain English.

Core metrics

Retention

Did the player come back? Measured in days: D1, D7, D30. The single most important health metric in free-to-play games, because everything else — revenue, virality, LTV — sits downstream of players actually returning.

Day-N retention (D1, D7, D30)

The percentage of a cohort that opened the game exactly N days after installing. D1 is the next day, D7 is one week later, D30 is one month later. "D7 of 12%" means 12 out of every 100 installers came back on day seven.

Churn

The opposite of retention — the player left and didn't come back. Usually defined against a window: a player who hasn't played for 14 days is considered churned in most mobile contexts.

Churn rate

The percentage of players who stop playing within a time period. High churn means players are leaving fast; where and why they churn is the actionable part.

DAU / MAU

Daily / Monthly Active Users — how many unique people play per day or per month. The DAU/MAU ratio is a rough stickiness measure: 0.25 means the average monthly player shows up about a quarter of days.

Session

One continuous play period — the player opens the game, plays, and closes it. Session count and session length together describe how people engage, not just whether they do.

ARPDAU

Average Revenue Per Daily Active User — total daily revenue divided by DAU. It looks tiny ($0.12 is respectable in many genres) until you multiply by thousands of players and 365 days.

LTV (Lifetime Value)

The total revenue a single player is predicted to generate over their entire time in your game. LTV against acquisition cost is the fundamental economics of free-to-play: if LTV is above CPI, you can buy growth profitably.

Cohorts

Cohort

A group of players who all installed on the same day (or week). You track them as a unit, because comparing "everyone who installed Tuesday" against "everyone who installed after the update" is how you isolate what actually changed.

Cohort matrix

A table where rows are cohorts (install dates) and columns are retention days (D0, D1, D7…). The standard retention tool in mobile gaming — one glance shows whether newer cohorts retain better or worse than older ones.

Predicted retention

Retention values for days that haven't happened yet, estimated with decay curves or ML survival models. It's how you can judge a two-week-old cohort's D30 without waiting two more weeks.

Decay rate

How fast players drop off day over day. A decay of 0.88 means each day retains 88% of the previous day's players. Simple, useful, and the baseline that fancier models have to beat.

Survival model

A statistical model predicting how long each player will keep playing, based on their early behaviour. More accurate than a decay curve because it looks at what players do, not just when they installed.

User acquisition

UA (User Acquisition)

Spending money on ads to get new players. The discipline is entirely about arithmetic: what an install costs versus what that player eventually returns.

CPI (Cost Per Install)

Total campaign spend divided by installs it produced. "$2.50 CPI" means each new player cost $2.50 to acquire.

ROAS (Return On Ad Spend)

Revenue from a campaign's players divided by the campaign's spend, as a percentage. 100% is breakeven; everything above is profit. Usually tracked as a curve over days-since-install rather than a single number.

Breakeven day

The predicted day a campaign crosses 100% ROAS. "BE D18" means the campaign is expected to have paid for itself 18 days after its installs land.

Organic

Players who found your game without paid ads — store search, word of mouth, a streamer having a good day. Effectively free, which makes the organic share of installs a big lever on blended economics.

MMP (Mobile Measurement Partner)

Attribution tools like AppsFlyer, Adjust, or Singular that work out which ad brought each install. Without one, campaign-level revenue attribution is guesswork.

Install quality

Not all installs are equal. A high-quality install is a player who retains and spends; a low-quality one churns on day zero. Two campaigns with the same CPI can have wildly different economics because of this.

Player archetypes

Whale

A heavy spender — frequent premium purchases, often competitive. Typically 1–5% of players but half or more of revenue. Losing one whale can matter more than losing a hundred casual players, which is why churn prediction pays special attention to them.

Dolphin

A moderate spender — the occasional starter pack or cosmetic. Individually modest, collectively a sustainable revenue base.

Grinder

Plays a lot, spends little — earns everything through gameplay. Valuable for ad revenue, liveliness, and ecosystem health; a game of only spenders is a mall, not a world.

Casual

Light player — short sessions, rarely returns after the first week, almost never spends. Most of your installs, statistically.

Clustering

The ML technique (K-means and friends) that groups players into natural behavioural segments based on spend, sessions, and social actions — how archetypes get assigned without a human labelling anyone.

Game design & economy

Funnel

A sequence of steps — install → tutorial → first purchase — where each step loses some players. Funnel analysis shows exactly where you lose people, which beats knowing only that you did.

Drop-off rate

The percentage of players who reach a point (say, level 14) and never get past it. A spike in drop-off at one spot is a design problem wearing a data costume.

Frustration score

A composite of fail rate, retry attempts, and churn at a specific level or moment. High frustration with high retries is a challenge; high frustration with high churn is a wall.

Sinks & sources

Sources are how players earn currency (rewards, quests); sinks are how they spend it (upgrades, items, continues). A balanced economy keeps both in tension.

Inflation

Players accumulating currency faster than they can spend it, devaluing the economy. The data signature: rising balances, falling spend-per-earn, and sinks nobody visits.

Prediction & explainability

Genre benchmarks

Typical metric values for games of the same type — match-3, RPG, strategy. They answer the question every founder actually asks first: "is this number normal?"

Genre delta

How your metric compares to the genre's typical value. "+2% vs genre" means you're two percentage points above it.

Percentile

Where you rank among comparable games. "Top 25%" means 75% of games in your genre score lower than you on that metric.

SHAP values

A method for explaining ML predictions: for each prediction, every input feature gets a score for how much it pushed the result up or down. It's how "73% churn risk" becomes "73% churn risk, mostly because session length halved and the last three levels were all failures" — the difference between a number and something you can act on.

Anomaly detection

Automatically flagging when a metric moves outside its normal range — revenue spiking, D1 collapsing after a release. The point isn't the flag; it's noticing on the day it happens instead of in next month's review.