How to Measure Discord Event Success: Count Who Comes Back, and Give Each Format Two Runs

The scorecard, before the reasoning

The usual way to judge a community event is to count who attended. Replace that count with the scorecard below. It keeps one line for every run of every event format, and it records what attendees did after the event as well as how many of them there were.

EVENT SCORECARD (sample figures, to show the layout only)

Format            Run  Interested  Attended  Rating     Came back  Posted
Workshop           1       15         10     4.2 (6)       n/a     8 of 10
Workshop           2       14         11     4.4 (7)     8 of 10   9 of 11
   Decision: KEEP
Watch-along        1       70         48     3.4 (12)      n/a     6 of 48
Watch-along        2       66         21     3.1 (5)     5 of 48   3 of 21
   Decision: DROP
Question session   1       30         19     4.0 (8)       n/a     9 of 19
Question session   2       31         20     4.1 (9)     4 of 19   8 of 20
   Decision: CHANGE ONE THING, then two more runs

Rating     = average on a 1 to 5 scale (number of raters in brackets)
Came back  = of the people at the previous run, how many attended this one
Posted     = attendees who posted anywhere in the server in the next 7 days

Two words need defining. A format is a type of event, such as a workshop, a watch-along or a question session. A run is one occurrence of that format.

The sample lines make the case. The watch-along had the fullest room on the card by a wide margin, and it is the format being dropped. The workshop had the smallest room, and it stays. Attendance alone would have produced the opposite decisions.

I served as Head Discord Moderator for Google Developers, and a card like this is how I keep track of events today.

How each column gets filled

Two of these columns rest on things Discord provides. A Scheduled Event is Discord's built-in event listing, which members open to see what is planned and when. According to Discord's help center (2023), members can press "Interested" on an event and are notified when it starts.

For the Posted column, Discord's help center (2024) states that the Members page shows a message count for each member. That page works on desktop or in a browser. Write down each attendee's count on the day and again seven days later, and anyone whose count rose has posted. With a long list, check every third name by the same rule each time, or give the counting to an automation bot.

The attendance list deserves care, because two columns depend on it. One staff member, named before the event starts, writes down every name present at the midpoint and posts the list in a staff channel, which is a channel only the team can see. The list can be kept by hand or by an automation bot. I use the midpoint because the opening minutes include members who look in and leave, and the last minutes hold only those with time to spare. Label each list with the format, the run number and the date.

Why the Attended column cannot give the verdict

For anyone running a creator, partner or ambassador programme, attendance is the figure leadership asks for and the easiest one to supply. It misleads in three ways.

It grades the invitation. How many people arrive depends mostly on how many saw the announcement, how many reminders they received and whether the hour suited them. A middling event announced five times draws more than a strong one announced once.

It pays out for novelty. Anything new draws the curious on its first outing, and a guest draws the guest's followers. A crowd like that says nothing about whether the format should return every month.

It treats a regular and a tourist as equals. The member who attends every session and the member who looked in once are each counted as one. Your programme depends on the first kind.

The effect on the calendar is predictable. One large crowd keeps a weak format scheduled for months, since nobody wants to cancel the best-attended event the server has had. A small session with the same faithful attendees gets removed for low numbers.

Reading Interested against Attended

This pair of columns isolates the invitation from the event. It produces three patterns.

1. Interested is high and Attended is low. Members liked the idea, and the hour did not suit them. Change the slot and leave the format alone.
2. Interested is low and Attended is high. People came because of a reminder or a friend on the day. Point the next announcement at the event listing so that interest is registered earlier.
3. Both are low. The topic missed the members it was written for. Go back to the goal sentence described further down.

Treat this pair as a check on the announcement and the slot. A format is never kept or dropped on it.

The rating, and how far to trust it

In the final minute, while people are still present, a staff member posts one message in the event channel, the text channel where the chat for the event happens.

Before you go: one tap to rate this session.
1 = not worth the hour        5 = well worth the hour

And if you have ten seconds, one question:
What would have made this more useful to you?

Add the five number emoji beneath the message so that a rating takes one tap. Write down the average and how many people rated. The wider subject of what to ask as an event closes, and how quickly, is one I cover in a separate piece.

Use the rating for one job only: comparing the second run of a format with its first. The format, the slot and the question are the same, so a clear move in the average points to something inside the session. The answers to the open question tell you what it was. They also tell you which single thing to change, if the decision turns out to be a change.

Return, the two columns that decide

Came back asks how many of the people at run one attended run two. Match names, never totals. Twenty people at each run may be the same twenty or forty different members, and those two results mean opposite things.

Posted asks how many attendees posted anywhere in the server during the seven days after the run. A high share means the event sent members back into the community. A low share means the hour left no trace once it ended.

I give these two columns more weight than the rating. Tapping a number takes a second. Returning costs a member another hour, with no novelty to pull them in, so it is the stronger statement about what the first hour was worth.

Two runs, held equal

A new format runs exactly twice before it is judged. Before run two, confirm each of these:

How to promote an event is its own subject, and the only requirement here is that both runs get equal treatment.

Two runs is the smallest test that works. A single run cannot separate the format from the novelty of a first outing, and the Came back column stays empty until a second run exists. Going past two without a decision is how a weak format becomes a fixture. Decide a few days after the seven-day check on run two.

Turning the numbers into a decision

Fix the meaning of good and weak before any numbers arrive.

• Keep the format when rating and return are both good. It receives a regular slot, and you keep recording its numbers.
• Change one thing and run it twice more when the rating is good and return is weak. Members enjoyed the hour and found no reason to come again. A typical change is making each session continue from the last.
• Change one thing and run it twice more when return is good and the rating is weak. Members keep coming for the group or the habit while the hour disappoints. A typical change is the length or the structure.
• Drop the format when both are weak after two runs.

A change means one change. With two, you cannot tell which one moved the numbers. I allow each format one round of changes, so after four runs it is either kept or dropped.

Then say the decision in the server, in the channel where events are announced. Members who watch a format vanish without a word conclude that events here are abandoned. Members who read the reason conclude that events are tested and that their ratings counted.

We tested the watch-along twice this month. Thank you to everyone who
joined and rated it. Ratings were low on both runs and few people came
back for the second, so we are retiring it. The weekly session stays
exactly where it is, and the next format we test will be posted here.

One sentence before anything is scheduled

Every event starts with a single sentence naming the company goal it serves.

This event exists so that (which members) (do what),
which supports (company goal).

A creator programme might write: this workshop exists so that new ambassadors publish their first piece, which supports the goal of more active creators each quarter. When the sentence cannot be written, the event does not go on the calendar. Without a goal, attendance is the only thing left to judge an event by, and attendance is the number this method replaces.

The sentence also focuses the Posted column. If the goal is published work, check whether attendees posted in the channel where work is shared, and note that figure beside the general one.

How much to test at once

1. One new format under test at a time. Two tests draw on the same members and the same announcement space, so neither produces a clean result.
2. One fixed recurring event that is never the format under test. It gives members one event that stays put while others are tried.
3. A monthly review of the scorecard. I allow thirty minutes, with the programme owner and the person who kept the attendance lists.

The review answers three questions:

• Which formats have a decision due this month?
• Has any kept format seen return fall on two runs in a row? If so, it goes back under test.
• Which format is tested next, and what is its goal sentence?

The schedule this card serves

Measuring events and deciding which formats stay belongs to engagement rhythm: the recurring schedule of events and prompts that brings members back week after week. Two other layers sit beside it. Community reporting carries results to leadership, and onboarding is where a new member first learns that events happen.

The measuring is covered here in full, and it needs nothing more than a staff channel, one document and a person to keep the list. What the card cannot do is arrange the kept formats across a week, turn its lines into a report leadership will read, or bring new members to their first event.

A scorecard this small takes a few minutes per event, and it replaces a debate about attendance with a record of what members did next. More at danieljeong.org.