Retention, and what it does to your ad budget
Churn is your biggest lever on acquisition. Every point changes what you can afford per player.
Monthly churn
30%
About 3.3 months per subscriber
Subscriber value now
£20
Monthly plan, at current churn
At 20% churn
£30
5 months, 50 percent more to spend
Best source at day 90
Discord
64% still subscribed
Retention by acquisition source
Share of paying subscribers still active. This is the number that decides which channel actually deserves budget.
| Source | Day 7 | Day 30 | Day 60 | Day 90 | On annual |
|---|---|---|---|---|---|
| Discord | 96% | 83% | 71% | 64% | 44% |
| 93% | 74% | 58% | 49% | 31% | |
| Meta, face creative | 91% | 72% | 55% | 46% | 28% |
| Meta, broad | 88% | 64% | 42% | 31% | 16% |
| X | 89% | 61% | 44% | 34% | 14% |
Discord keeps 64 percent of its subscribers to day 90 while Meta, broad keeps 31 percent. Two subscribers acquired on the same day are worth roughly double each other depending only on where they came from.
At risk right now
Saves fire at the moment, not as a winback after they have gone.
291
Kept in the last 30 days
Finished a term, nothing started since
Within 24 hours of the final month
412
players
Fresh scenario suggested, plus a Discord invite
No session in 10 days
Before the next renewal date
688
players
Nudge with an unplayed era or premium mode
Abandoned mid-career, losing term
After the second abandoned session
204
players
Difficulty guidance and a scenario restart offer
Renews in 3 days, low engagement
72 hours pre-renewal
156
players
Surface what they have not tried yet on premium
The Discord is already the strongest retention surface you have, so the triggers point players into it rather than trying to build a new habit somewhere else.
Where subscribers actually leave
Churn clusters at moments rather than spreading evenly, which makes it fixable.
First career finished
38%
They completed a 48-month term and had nothing queued next
Month 2 renewal
24%
Charged again before playing a second scenario
Stalled mid-career
19%
Last session ended inside a losing term
Everything else
19%
No single dominant pattern
A player who finishes a 48 month term with nothing queued next is a solvable problem. A trigger into a fresh scenario or into your Discord at that exact moment is worth more than any bid adjustment.
Audiences pushed back to Meta
This is where retention data turns into cheaper acquisition.
Retained subscribers, 90 days
4,820 players
Players still subscribed at day 90. What Meta should be matching against.
Annual plan buyers
1,640 players
Worth roughly three times a monthly subscriber on day one.
Churned inside 30 days
3,110 players
Suppressed so spend stops chasing the pattern that already failed.
Signed up, never played
9,740 players
Free accounts burning inference with no path to a subscription.
Built from players who stayed and paid rather than players who clicked. Political ads sit in a restricted category on Meta, so which audience tools are available gets confirmed at setup. The underlying data is the same either way.
What a high-value subscriber looks like
Political ads run in a restricted category where audience targeting is limited, so creative and the offer are the levers. This profile is the brief.
Comes from Discord or your own subreddit
Community-native players convert 2-3x better. Make creative that looks like the community, not like an ad.
Buys annual, not monthly
44% of Discord subscribers take annual. Lead with the annual framing for community-sourced traffic.
Plays multiple scenarios in week one
Breadth in the first week predicts staying. Show scenario variety in creative rather than a single run.
Started on a face-led creative
Face creative brings the cheapest subscribers and they retain better. Weight production toward it.
Stays past the second renewal
The month-2 charge is the cliff. Offer and onboarding work harder there than any ad does.
Concept view on sample data. The churn moments shown are a hypothesis based on how the game is structured, a free tier capped at month 12 and a full career running 48 months. Real cohort data either confirms it or points somewhere better.