Reading the player
Game and player analytics: what the data says, what it doesn't, and how it changes design
BSc (Hons) Game Design & Development
What does your favourite game know about you?
Not your name. Your habits: when you play, where you die, what you buy, when you quit.
By the end of today you can…
Explain what analytics is for
where telemetry fits in the analytics value chain, and why it needs qualitative data beside it.
Read the core metrics
engagement, retention and churn, progression, economy, social and technical.
Turn a number into a design change
evidence, hypothesis, change, measurement, and an A/B test to check it.
Collect data responsibly
with privacy, consent and minimisation, and without turning prediction into manipulation.
Six stops
Why analytics?
Value chain, telemetry, what and why
Engagement & retention
DAU, D1/D7/D30, churn, lifecycle
Progression & funnels
Where players stop, and why
Economy, social, technical
Currencies, communities, frame rates
Data to decisions
A/B tests, prediction, tools, ethics
Practical
Pip's Peak: read the climb
Weeks 2, 3 and 7, in one line each
You recorded flow scores, processed them in a class sheet and analysed the result. That is the analytics pipeline, by hand.
"Retention on its own isn't the goal: a slot machine has great retention." Today we measure retention, so keep that in mind.
Playtesting shows you a handful of players closely. Analytics shows you thousands, from a distance.
Analytics tells you what players do, at scale. Playtesting tells you why. A designer needs both.
Why
analytics?
From gut feeling to evidence
Every design is a hypothesis
"This level is a fair step up." Is it? Data checks the hypotheses built into every design.
Data-driven changes improve retention, engagement and satisfaction, on purpose rather than by luck.
When the team disagrees, a shared number ends a lot of arguments.
"Without data, you're just another person with an opinion." Often attributed to W. Edwards Deming.
And with data but no judgement, you're a person with a spreadsheet.
GDnD wiki · Game AnalyticsThe game analytics value chain
An event is a sentence about play
// one event, sent when a player falls { "event": "player_death", "player": "P0417", // pseudonymous "time": "2026-09-03T19:42:07Z", "level": 4, "x": 212.5, "y": 48.0, "cause": "wind", "attempt": 3, "fps": 27 }
Every event answers the same four questions, plus the details you'll need later.
Decide what you want to know, then which events answer it. "Log everything" gives you a haystack.
tutorial_step, level_start, level_fail, purchase: consistent names make queries easy.
What players do, and why they do it
Measurable, numerical, often system-generated. Captures what players do.
Dashboards · telemetry · logs
Descriptive, from the player's experience. Explains why it happens.
Surveys · think-aloud · observation · sentiment analysis
| Telemetry says | 40% of players quit after Level 2. |
| Playtesters say | "The puzzle feels unfair." "The controls are confusing." |
| Triangulate | One interaction in Level 2 reads as breaking the rules. Redesign that, then measure again. |
Engagement
& retention
Do they play, and do they come back?
How much, how often, how long
Average duration of a play session.
How often players return per day or week.
Total accumulated time.
Daily, weekly and monthly active users.
DAU ÷ MAU. Around 20% or more suggests a habit (Sellers).
% of players who return after a set time: D1, D7, D30.
Long sessions aren't automatically good. Ask what players were doing.
Who stopped playing?
1,000 players start. 250 stop playing within a week. Churn = 25%; weekly retention = 75%.
64–72% churn by Day 1. 94–97% by Day 30, so D30 retention is often just 2.5–5%.
D1 40%+ · D7 20%+ · D30 10%+
Most players decide in the first minutes. The first-time user experience (FTUE) is a design priority, not an afterthought.
GDnD wiki · Games as a ServiceRetention by cohort
Club members stay 3× longer. Should every player be pushed into a club? Careful: the club only unlocks after Level 3.
Five stages, three retention checkpoints
Compare players by days since install (cohorts), not by calendar date.
GDnD wiki · Player LifecycleChurn spikes after Level 3. Now what?
Where exactly?
Event logs and heatmaps: which room, which jump, which screen.
What changed?
Death rate, difficulty, UI confusion, unclear objectives, an unskippable cutscene.
Why?
Cross-reference with surveys and playtests (Week 7).
Did it work?
Tutorial, checkpoint, pacing or reward timing; then A/B test.
The usual suspects: onboarding bottlenecks, difficulty spikes and reward droughts.
Progression
& funnels
Read the numbers as a story of engagement
How players move through the game
| Metric | What it reveals | If it looks wrong |
|---|---|---|
| Level completion rate | % of players finishing each level | Sharp drop: difficulty spike, unclear goal, poor checkpoints |
| Checkpoint frequency | How often players reach milestones | Too few: fatigue. Too many: no tension |
| Failure / retry counts | Balance and frustration | High: unclear tutorial. Low: no challenge, a mastery plateau |
| Item / skill unlocks | Reward pacing | Long gaps disengage: aim for early reward density |
| Achievement rate | Which goals players pursue | Rare: obscure or grindy. Universal: no perceived value |
The funnel
Heatmaps show where; timelines show when
Deaths, paths, clicks, plotted on the level.
Behaviour over time: when in a session problems happen.
Live KPIs for the whole team, from designers to QA.
Level 4 keeps 74% of starters. Level 3 kept 95%.
What could cause the drop? Be specific.
Which data would support or rule out each one?
What would you ask a playtester?
Also known: deaths per attempt went from 0.24 on Level 3 to 0.64 on Level 4.
Economy, social
& technical
Three more places players leave from
Follow the currency
| Earn / spend ratio | Is the economy balanced, inflating or starved? |
| Sinks (drains) | Somewhere to spend; no soft-lock from scarcity |
| Item popularity | Perceived value of what's for sale |
| Conversion rate | % of players who ever pay |
| ARPU · ARPPU | Revenue per user · per paying user |
10,000 players. 150 of them pay. Revenue £4,500.
A tiny group pays for everyone. That's why who you design spending for is an ethical question.
The health of the community
Long waits signal a population imbalance and make the game feel empty. Fixes: flexible rules, cross-play, bots, something to do while waiting.
Messages, trades, assists, conflicts. Pro-social tools (pings, emotes, assist rewards) raise the good ones.
Spikes signal frustration, imbalance or poor communication tools. Data feeds moderation and reputation systems.
A fall in participation often comes before churn: a leading indicator.
Bad performance looks like bad design
Frame-rate stability
Drops break timing and feel.
Loading & latency
Waiting is the first thing new players do.
Crash & error rate
Every crash is a chance to quit.
Responsiveness
Lag reads as "unfair" (Week 3).
Pip's Peak: players on devices under 30 fps keep only 2% at D30, against 10% overall. Is the game too hard, or is it just running badly?
From data
to decisions
Hypothesis, change, measure, repeat
The loop, and the A/B test
Randomly split new players. A gets the current version, B gets one change. Compare one metric chosen in advance.
One change at a time · random split · enough players · same time window · decide what "better" means first.
A test that raises D7 by adding more daily nags may also raise uninstalls and complaints. Measure the side effects.
Who is about to leave?
Churn (who stops) · spending (who buys) · retention trends
Decision trees, random forests, SVMs, neural networks, hidden Markov models
Help them past Level 4, or send them a discount? Same prediction; very different ethics.
Two you can use this year
- Sessions, active users, retention, revenue
- Custom events: level complete, deaths, positions
- Funnels through gameplay sequences
- Remote Config and A/B tests without a new build
- Predefined dashboards: sessions, retention, progression, resources, performance
- Custom events for specific actions
- Heatmaps (add-on), errors and performance
- SDKs for Unity, Unreal, Godot, C++
Or start smaller: write your own events to a CSV and open it in a spreadsheet.
Analytics that changed the game
Death heatmaps of multiplayer maps showed spots that were too strong. The maps were changed before launch.
Level difficulty tuned from how often players fail and retry each level.
Report and chat data used to design interventions against toxic play.
In each case the data found the problem. Designers still had to decide what to do.
Collect less. Explain more. Use it for players.
- Telemetry linked to a player ID is usually personal data
- You need a lawful basis and a clear privacy notice
- Data minimisation: only what answers a question
- Extra duties for children (ICO Children's code)
- Predicting who will pay and targeting them is Week 3's dark patterns, automated
- Optimising D7 alone can reward compulsion over fun
- Pseudonymise, aggregate, delete what you don't need
The test: would players be comfortable seeing what you collect, and what you do with it?
Five things to remember
Analytics is the what; playtesting is the why. Triangulate before you redesign.
Retention and churn are the headline, but segment them: by cohort, device and progress.
Read funnels as stories. Fix the earliest big drop first; the gain carries through.
Correlation isn't cause. Test changes with one-variable A/B tests and a metric chosen first.
Collect less, explain more, and use prediction to help players, not to target them.
Practical:
read the climb
Metrics in a game you know, then a launch week of telemetry
Metrics in a game you know
- Pick a game: League of Legends, Fortnite, Hades, Stardew Valley, FIFA… or your own
- 3 in-game metrics (e.g. KDA, gold, creep score) and 2 performance metrics (frame rate, load times)
- How is each tracked? Which design decision would it inform?
- Choose one tool or method: an analytics platform, server logs, heatmaps, surveys
- How does it work? What does it give developers?
Runs completed, boons collected, time per run → is a weapon or boon underused? Are runs getting shorter as players improve?
Pip's Peak: launch week
A mobile climbing game. A five-step tutorial, ten levels from Mossy Steps to The Summit, gems to spend on skins, a climbing club after Level 3.
2,000 players, followed for 30 days. Made up for teaching: no real players.
| Sheet | One row per… |
|---|---|
| Read me | column meanings |
| Players | player: device, tutorial step, highest level, D1/D7/D30, club, gems, spend |
| Attempts | player per level: attempts, deaths, completed, minutes |
Download CRE342_W8_PipsPeak_Telemetry.xlsx from the module site. Pivot tables are your friend.
Find three findings
| Look at | Ask | Hint |
|---|---|---|
| The funnel | Where are the two biggest drops? One is Level 4 (the worked example). Find the other. | Players: tutorial_step_reached |
| Retention | D1, D7, D30 overall; then by device and by club. Is the club a cause? | Average of returned_d7 by device_fps |
| Difficulty | Completion % and deaths per attempt for each level | Attempts: sum by level |
| Economy | Gems earned vs spent by highest level reached. What's missing? | Ratio earned ÷ spent |
| Money | Conversion, ARPU, ARPPU | spend_gbp > 0 |
For each finding: evidence → hypothesis → design change → how you'd measure it → one playtest question.
Level 4, "Gusty Gap"
| Evidence | 1,198 players started Level 4; 881 finished it (73.5%, against 92–98% for every other level). Deaths per attempt: 0.64, against 0.24 on Level 3. |
| Hypothesis | Wind is new in Level 4 and never taught: players don't realise the gust is pushing them, so falls feel random. |
| Change | Add a short, safe wind section before the gap, with visible wind lines and a sound cue (Week 3 feedback; Week 4 affordance). |
| Measure | A/B test on new installs: Level 4 completion is the success metric, chosen first. Watch D7 and Level 5 too. |
| Playtest question | "Why did you fall there?" If players say "the wind", the fix is right; if "the jump button", it isn't. |
One chart, one finding, one change
The chart
One clear chart. Label the axes; say what one bar means.
The finding
One sentence, with a number in it.
The change
What you'd change in the game, specifically.
The test
How you'd know it worked.
“In my own game, I'd track …, because it would tell me ….”
One sentence each, before you leave.
Clark, O. (2014) Games as a Service: How Free to Play Design Can Make Better Games. Burlington, MA: Focal Press.
El-Nasr, M. S., Drachen, A. and Canossa, A. (eds) (2013) Game Analytics: Maximizing the Value of Player Data. London: Springer.
Information Commissioner's Office (2020) Age Appropriate Design: A Code of Practice for Online Services. Wilmslow: ICO.
Lin, J. (2013) 'The science behind shaping player behavior in online games' [conference talk]. Game Developers Conference 2013.
Sellers, M. (2017) Advanced Game Design: A Systems Approach. Boston: Addison-Wesley.
Su, Y., Backlund, P. and Engström, H. (2021) 'Comprehensive review and classification of game analytics', Service Oriented Computing and Applications, 15, pp. 141–156.
Thompson, C. (2007) 'Halo 3: how Microsoft Labs invented a new science of play', Wired, 15(9).
Yannakakis, G. N. and Togelius, J. (2018) Artificial Intelligence and Games. Cham: Springer.
Game Analytics · Telemetry · Player Lifecycle
Games as a Service · F2P Monetisation Design · Internal Economy
Progression and Power Curves · Playtesting · Player Modelling