A-Game Labs / Community & Retention

Just because the music is still playing doesn't mean the ship isn't sinking

Near the end of Titanic, the band keeps playing. The musicians know. Everyone on deck knows. They play anyway, to comfort people while the water rises. I think about that scene when I look at communities on edge.

By Julio VargasCommunity Strategy02 September 20266 min read

The music tells you nothing about what is happening below deck.

Most founders I speak to cannot tell me whether their community is healthy.

They can tell me its size. A few hundred members, maybe a few thousand. They can tell me about the guy who posts memes, and the one who complains about everything, and that most days there is some conversation happening.

None of that answers the question.

Players will not tell you they are drifting away from your community, and from your game with it. What you see instead is smaller. They post a little less. They stop answering each other. The regulars thin out, and the new arrivals never say anything at all.

You may or may not notice. Either way you keep playing your music, and you decide the problem is content, or visibility, or an influencer campaign. Then one day the server is quiet, and the ship has been going down for months.

Posting rates, who answers whom, how long the regulars last: all of it is measurable, and most of it weeks in advance. So here are the five signals I actually look at, what each one costs when it goes unread, and how to check yours this week.

First, the two numbers that will mislead you

▼ Member count

It only ever goes up. A server can add members every week while nobody who joined last month is still reading. Growth and health are different measurements, and only one of them is on your dashboard.

▼ Message volume

It is usually a handful of regulars talking to each other. Healthy-looking totals routinely hide the fact that no new member has spoken in weeks.

There are five signals that can tell you whether your community is thriving or not. All of them are either a ratio or a trend, not a count. Fancy metrics will flatter you without telling you what is actually happening.

The five signals

Signal 01

Do new members post in their first 24 hours?

Whether somebody posts once on their first day predicts almost everything downstream. A member who lurks through week one rarely comes back in month two.

Check it this week

Take your last 20 joins. Count how many ever posted anything. If it is under a quarter, your problem is onboarding, not content.

▼ Concord

Nobody built the room. No creator programme, no community push, and a first trailer that landed roughly two months before release. So there was no first-day ratio to measure, because there were barely any first days. You cannot activate members you never recruited, and by launch day it is far too late to start. Servers off in 14 days, Firewalk closed, 174 jobs.

▲ Deep Rock Galactic

Ghost Ship's developers answer in the server as routine, not as a campaign. A new member who posts once and gets a reply from an actual developer has a reason to post twice. That is what builds a first-day ratio, and it costs attention rather than money.

The decisionIf the ratio is low, fix onboarding before you spend anything on acquiring new members. Give them a reason to say something in their first few minutes, and check whether your welcome invites a reply or just talks at them.

Signal 02

Do members talk to each other, or only to you?

Members answering members is the difference between a community and a support channel. It is also the part that keeps working when you cannot.

Check it this week

Take the last 50 messages in your busiest channel. Count how many are members talking to each other, and how many are aimed at you or answered by you. If most of the room routes through staff, you are running a help desk with a chat window.

▼ Good intentions, no strategy

No famous name for this one, because nobody publishes a postmortem admitting they talked too much. It is the pattern I run into most, and it comes from good intentions rather than neglect. You end up running community on instinct instead of a plan: one enthusiastic founder or CM answers every question inside five minutes, so members stop bothering to answer each other. The graphs look healthy the whole time. Then if that person gets busy with something else, the community goes quiet, then silent. There was no community. It was a support queue.

▲ Final Fantasy XIV

Square Enix built the peer conversation instead of hoping for one. New players carry a sprout icon and get invited into the Novice Network, a channel where they ask other players rather than the studio. Veterans who answer wear a mentor tag, and both sides earn bonus experience for playing together. The help happens between members by design, and nobody from Square Enix is in the room.

The decisionIf most messages route through you, start leaving questions alone for an hour. Someone else will answer. When they do, say so in public and give them something for it: a role, early access, whatever you actually have.

Signal 03

When players are unhappy, do they hear back from you?

People are never all happy. What matters is whether you know what they are unhappy about, and whether they get an answer. Unanswered frustration does not disperse. It hardens, and then it settles on your storefront.

Check it this week

Write down the top three things your players are annoyed about. Then go and read the server. If your list was wrong, you have been guessing. Then find the last thread where people were angry and count the hours before anyone from the studio replied in it.

▼ Battlefield 2042

The count of hours ran to about six weeks. DICE went quiet through the holidays after a rough launch, the exact window when the complaining was loudest. When EA's comms director finally replied in January 2022, he called fan expectations "brutal," then deleted the post and apologised. The silence had already done the damage, and the reply confirmed it.

▲ Helldivers 2

The count ran to hours. When a PSN policy pulled the game from 177 countries, Arrowhead's CEO was in the conversation the same day, told players to leave negative reviews, and said the team "didn't talk about it enough." Sony reversed within days and the reviews came back with it.

The decisionAn angry thread with no reply from you is the cheapest problem you will ever fix. But a reply on its own is not the fix. Say what you are going to do about it and when, even if the honest answer is that you are not going to do anything. Then put a crisis process in place, so the next thing that breaks does not catch you the same way.

Signal 04

Do you enable your contributors?

Having a few regulars who help out is a good start. Recognising them in front of everyone else, backing them when they get something wrong, and giving them the information they need is what turns willing players into people you can rely on.

Check it this week

Take the last 20 questions a player answered, or the last player-led thing that happened in your server. Name the people behind them. Then ask what they have ever had from you for doing it.

▼ The Day Before

Volunteers ran the community and answered for the studio. When information they had been given turned out to be wrong, Fntastic blamed them in public. The volunteer replied that it had come from the team. Then the studio deleted the Discord those people had built, and everything they had contributed went with it.

▲ Warframe

Digital Extremes picks volunteer chat moderators out of the player base and gives them an accolade that shows on their in-game profile. Forum moderators are invited directly. The wiki administrators have a title of their own, Tenno Chroniclers, chosen alongside DE. The work is unpaid, but it is named and the studio stands behind it.

The decisionName and equip the three people already trying to help, before you need them. You almost certainly know who they are.

Signal 05

What have you promised that you have not shipped?

Every unshipped public promise is an open position against your credibility. Players keep the ledger whether or not you do, and it gets settled on your storefront.

Check it this week

List everything you have said publicly that you would do. Mark each shipped, in progress, or quietly dropped. The quietly dropped column is your risk.

▼ Overwatch 2

A mode promised in 2019 sat unanswered in every channel players gathered in until Blizzard cancelled it four years later. The ledger was never closed where the community could see it, so it got closed somewhere worse: roughly 90% negative across more than 100,000 Steam reviews. About 63,000 of those tie to the NetEase split, so the number is confounded, though the order of events is not.

▲ Baldur's Gate 3

Through three years of Early Access, Larian posted community updates in the places players already were, saying what had changed, what was coming next, and what they had decided against. Saying no in public, on a schedule, is what stops a promise turning into a debt. They left nothing outstanding at 1.0.

The decisionClose the ledger publicly, including the things you dropped. Saying "we are not doing this after all" costs far less than letting people find out at launch. Get in the habit of answering suggestions on a schedule too. If an idea got traction in your server, it has earned a yes or a no.

Let's be honest about this

What community management will not fix

  • It will not fix a game people do not want to play. Concord had design problems that sat well outside anyone's Discord.
  • It will not undo a monetisation model players reject. That is a business decision, and no amount of talking fixes it.
  • It will not compensate for a broken build. Battlefield 2042 needed patches as much as it needed presence.

What it gives you is warning. Every failure above was survivable at some point, and in each one somebody in the community knew before the studio did.

What to do next

Run the five checks yourself. It takes an afternoon and it will tell you more than your dashboard does. Start there, I mean that.

The harder part is reading them. A weak first-day ratio and a room where every answer comes from you look almost identical from the inside, and they point at completely different problems with completely different fixes. Get that read wrong and you spend a quarter fixing the wrong thing.

Let's talk

So, what did you find? Bring your five answers and I will tell you which one to act on first, what to do about it, and what you should expect to see.

Book a session →

Or email me at contact@a-gamelabs.com