Charles Cushing

Castles vs. Colliders: Observations on community dynamics

I’ve helped build and/or run 3 communities:

Over time, I’ve developed some observations on how communities form, grow, and decay. [1]

I’ve tried to summarize these observations below. This is not meant to be comprehensive or applicable to all communities (n=3), so take it for what it is :)

Community formation

A new community’s formational epoch is a special, and pre-ordains many of the later outcomes. Here are a few observations about this period.

As long as there is change, there will be new communities

Who is relevant to whom changes constantly. The market shifts — then suddenly, software engineers want to talk to hospital administrators, AI researchers want to talk to truckers, climate scientists want to talk to entrepreneurs, retail investors want to talk to each other.

Most new professional communities start due to a trigger. StartOps formed while hundreds of new D2C companies were being created, each figuring out ops best practices for the first time. On Deck started when abundant early-stage capital brought thousands of new founders into the market.

There will always be evergreen categories of community (e.g. for personas like “Product managers”). But changing market conditions also lead to changing relevance, which leads to new community formation opportunities.

Convening community is a form of network arbitrage

At any given moment, there are lots of strangers who would benefit from knowing each other.

If one person or organization (a “convener”) knows a critical mass of relevant-but-unconnected people, they have the opportunity to bring them together as a community. Doing this does three things:

  1. It unlocks value for the community members, transforming potential social energy into kinetic social energy.
  2. It creates a beacon that draws in other, similar people.
  3. It confers a measure of gratitude and status to the convener.

By creating a common space for relevant people to connect — removing the need for themselves to make 1:1 introductions — the convener unlocks exponentially more ties than would be otherwise practical. The convener also retains partial credit socially for the connections made via the community.

By starting a community, a convener is performing network arbitrage — in that they reap a risk-free profit (paid in social capital) from collapsing the distances across their own social graph.

The atomic unit of community is favors

I’ve read a number of authors who assert that the defining aspect of a community is a “shared purpose” or “common goal”. This is often true, but there are counter-examples: look at people living in the same small town, or whose kids go to the same elementary school. These people have something in common, but it’s rarely their purpose or their personal goals. Yet some of the strongest communities form in these groups.

Instead, I believe the lowest common denominator for community is this:

A willingness to help others solely on the basis of mutual community affiliation.

If we share a community, then I am (tautologically) pre-disposed to help you, even if I don’t know you that well personally. And a community where members don’t help each other simply isn’t a community.

For a new community-joiner, the “magic moment” often comes when they’re helped by a stranger, with no expectation of direct reward. Once a community member has received help, they are more likely to help someone else, creating a positive flywheel of reciprocity.

There’s a formula for community helpfulness

When you convene a group of people, how do you know if they’ll help each other? I believe the equation looks something like this:

Helpfulness = Recognition x Capability x E(Reciprocity)

Here’s what I mean by each term:

  1. Recognition — To what extent do members recognize each other as peers?

    The norms around peers, respect and status vary wildly in different environments (think: a public school PTA vs. a country club). But as a general rule in professional communities, people want to be around others of their own status or higher. This also works best when the status ladder is fuzzy, and everyone has roughly similar “potential energy”.

    If someone joins a community and feels they are unambiguously the most accomplished person in the room, they’ll leave. If they are inspired by the people around them, they’ll stay and invite their friends — potentially starting a powerful flywheel of referrals.

  2. Capability — Can members help each other?

    The ability for members to offer relevant value to each other — in the form of knowledge, work, friendship, investment, or anything else — is predicated on how many vectors of similarity they share:

    • Early-stage founders are similar (1 vector)
    • Early-stage fintech founders even more so (2 vectors)
    • Early-stage fintech founders selling to Chief Compliance Officers at mid-market banks are even more so (3 vectors)

    If your community members aren’t sufficiently alike in the problems they face, they won’t be able to help each other. If they can’t help each other, they won’t engage or stick around. [2]

  3. E(Reciprocity) — Reciprocity in a community is weird. It isn’t about whether the person you helped will eventually return the favor. It’s about whether anyone else in the community will eventually return the favor. More on this below.

    Even if the status recognition and capability conditions are met, it’s possible that people still aren’t that helpful to each other. This is largely a function of culture, which is often set in the earliest days of the community. If you walk into a community and everyone is giving generously of their time, you’ll do the same. If they’re not, you won’t (and the community likely won’t last long — there’s a lot of survivorship bias here).

There are obviously many nuances not captured by this equation, but I think it hits the most important ingredients.

A community’s founding members and rituals carry high inertia

After a satellite has been shot into space, the ground team can nudge the trajectory with maneuvering thrusters — but a lot of the initial momentum came from the launch. [3]

In a community, the organizers can make corrections with changes to policy and membership criteria; but likewise, the momentum is heavily influenced by initial conditions.

Most critically, the legitness of the seed group effectively sets the upper bound of legitness that you can expect in future members. This is because if a more-legit person enters your community, and doesn’t see other people like them, they will churn.

Every community also has rituals. This could be as simple as Friday drinks at a certain bar, or a custom of celebrating small wins in a Slack channel. These rituals usually play a functional role — e.g. to foster trust, or to exchange certain types of information.

Rituals may be good at their function, or bad at it. Regardless, when a community starts, the creation of rituals feel organic — everybody’s making things up as they go. After a certain point, experimenting with new rituals (or retiring old ones) can start to feel contrived or upsetting to older members. So you get a certain amount of lock-in, whether or not your set of rituals is functional (or scalable).

Platform matters

Having great people seed your community is necessary but not sufficient. The context in which they meet and communicate is equally important. Platform matters.

Different communities will need different platforms. A community of senior executives may like to gather and bond in the private room at a restaurant, but balk at the idea of a WhatsApp group. A community of gamers may feel perfectly at home on Discord, but get awkward when meeting IRL.

When I started StartOps, my first instinct was to use a private LinkedIn group, since LinkedIn is where our members were most active. We launched, and engagement was awful. Then, I switched to Slack, and suddenly, communication lit up. Just the UX of LinkedIn groups had turned them off.

Platform may not be static. During the pandemic, all cohorts in On Deck moved to Zoom. We got great at creating a comfortable and exciting atmosphere online. As the pandemic ended, it became harder and harder to keep the community online. Eventually, we switched back to IRL-first to adapt to this new reality. Many communities that start online end up spilling over into the real world, and vice versa.

When you onboard someone into a community, you force them to adopt your platform. So platform is not necessarily an impediment to joining. But once a member is in, the choice of platform has a big impact on how long they’ll stick around before they churn.

If you’re going to collect information, do it early

To get into a community, people will share a tremendous amount of valuable personal information — things they’d never share publicly, like their vendors, their side projects and hobbies, their personal interests, and their availability for work.

This data, if shared with other members, forms a rich way for people to discover each other and make connections. Most importantly, this data can be used to demonstrate how the person can be useful, and how they need help. This gives members the binding sites to connect with each other, or keep each other in mind for other opportunities.

After joining a community, it becomes order-of-magnitude more difficult to collect information from members. So it’s important to nail data collection in application and onboarding.

Individual behavior in communities

The macro behavior of a community is really the sum of hundreds or thousands of individual interactions. So I’ll start here with a few patterns I’ve noticed about individual behavior and incentives.

People seek to join communities in order to get leverage

Communities give a new joiner two powerful forms of leverage:

New community joiners will seek those most relevant

To get value out of a community, members will usually seek out the people who are most relevant to them. This process often organically happens in 3 phases:

  1. Prospecting. Most communities don’t share their membership publicly. So when someone joins a community, their first step is to see who else is in the room with them. This is why things like “welcome mixers” and “new member meet-and-greets” are a thing in pretty much every community.

    At On Deck, we experimented with a bunch of rituals to assist with prospecting. The first was the #intros channel in Slack — we prompted all members to post a short, fun intro about their personal lives and what they were working on. We also did a 1-2 hour session called Rapid Fire Intros, where 80+ people would pitch themselves for 90 seconds each. It may sound like overkill, but actually each format would reveal very different things about a person. We did both of these activities within the first few days of onboarding.

  2. Sampling. Based on their prospecting, a typical community-joiner might earmark anywhere from 3 to 30 people to follow up with 1:1 and go deeper. Different people approach this stage with different levels of proactivity and intentionality.

    These initial meetups help both parties determine whether it’s worth building a long-term relationship, and what kind — is the relationship one of friendship, advisory, investment, hiring, or something else?

  3. Mining. The sampling convos determine whether later ones will follow. Most people met through sampling will end up in one of three categories:

    1. Friends and trusted colleagues (“Gems”). These become true long-term relationships who enter your social circle in a meaningful way, and transcend the initial bond of mutual community affiliation.
    2. Supporters and weak ties (“Ore”). These are people you may only interact with occasionally, but there is a mutual feeling of appreciation and desire to help.
    3. One-and-done interactions (“Tailings”). Some relationships just don’t click. When there is a lack of mutual relevance, or vibes are off, both parties may leave the meeting with no intention of reconnecting.

At ODF, we’d consider it a great outcome if a new joiner explored conversations with 30-40 people, and left with 5-10 supporters and 1-3 friends.

If this process seems like a lot of effort — it is. Getting great outcomes from a community is non-trivial, and requires high conviction that it will be worth it. Those who aren’t willing to make this effort may leave most of their potential value on the table.

First impressions have an outsized impact on retention

When a new member joins a community, they will be highly sensitive to the status, capabilities, and E(reciprocity) of those they are meeting.

Many communities are too big for a new joiner to meet everyone. So instead, they’ll make an inference about the group based on their initial interactions. In my experience, this can be as few as 3 conversations. After these first encounters, the new member will decide how much to continue investing in the community.

This point was brought home for me in a conversation with an ODF founder. He took a three initial “sampling” conversations and was disappointed — the founders he talked to were earlier-stage than he was, and in different sectors. He was about to churn when he took an inbound fourth conversation. His bond with that person blew him away, and encouraged him to go find more relevant people. He ended up being a big net promoter. But we were one conversation away from losing him.

Maintaining community-built relationships requires high individual effort

When it comes to maintaining and using your social network, “community” by itself is not a panacea. Some communities offer ongoing access to digital campuses (Slack, Discord, etc.). Some have ongoing events where you might bump into other people you know and serendipitously catch up. But there are many other actions needed to keep your network healthy and useful:

Most communities do not have great infrastructure in place to support these activities. Memories fade, profile data goes stale, Slack channels become noisy and saturated with people you don’t know.

Joining a community can level the playing field between all members when it comes to meeting new people. But they do little to impart the skills, aptitude, and “ground game” needed to keep your new network healthy. Long-term community benefits accrue disproportionately to the minority of members who have these social skills, and are willing to put in the work.

Community dynamics over time

Community dynamics are complex, and hard to predict with certainty. Still, from my 10,000-foot-view as a community administrator, I’ve seen some interesting patterns emerge.

Communities exchange value via the “Electron Cloud of Reciprocity”

In order for a community to work, there is some game theory at play — each member must believe that each other member is there altruistically, willing to give at least as much as they take. Communities fall apart if they become too openly selfish or transactional.

It’s hard to fake this altruism, so people actually follow through! The result is a crazy sort of value exchange whereby community members offer help upfront and with no concrete claim to reciprocation.

I like to think of it as the Electron Cloud of Reciprocity. Members throw helpful favors up into the cloud, unsure of where they’ll land, and unsure of whose favors they’ll pick up. Value exchange does happen, but it’s highly unpredictable.

For scarcity-minded or transactional people, this dynamic is also what keeps them from joining communities in the first place. It’s an article of faith to think that you’ll get $X worth of value from a community, without knowing exactly what form that value will take or when it will arrive.

The ECoR is what enables that magic moment I mentioned above: someone helping you merely on the basis of your mutual community affiliation.

Ultimately, communities are very positive-sum. Indeed, they embody the economic concept of comparative advantage — if Member A is great at design and bad at fundraising, and Member B is bad at design and great at fundraising, they can exchange help to everyone’s benefit.

In this way, people joining communities can be self-serving without being transactional.

In the long run, every community becomes an information marketplace

After an initial burst of connectivity and openness, new community joiners’ social circles become saturated and they generally stop seeking to meet new people.

At this point, the main utility of the community is that it’s a better-than-average way to get a question answered — especially when the answer to that question is a matter of taste, opinion, or tribal knowledge (the things you can’t search on Google).

Members seek answers either by posting in a public channel (we call this the #asks channel in On Deck and StartOps), and/or by pinging individual people they met through the community.

Communities naturally degrade with scale and time

Neither scale nor time are kind to the core dynamics of communities. Here are the forces that break communities down:

  1. Greater scale = Less intimacy. Posting a request to 150 people you recognize is low-risk. Responding to a request made by a friend you know well is its own reward. The same is not true with 1,500 people, or 15,000.
  2. Greater scale = More noise. Finding the relevant person to talk to (whether receiving or giving help to the community) gets progressively harder the more people there are.
  3. Greater scale = Lower recognition. As a community grows, there is usually some natural mean reversion in terms of joiners’ status (as well as more variance). This feeds back further into the issues of noise and intimacy.
  4. Longer time = Lower need. When someone first joins a community, they get a huge burst of utility as they make new connections and fulfill their goals. After this initial period, the ongoing value of the community tends to dip and plateau at a lower level.
  5. Longer time = Colder relationships. It’s not uncomfortable to text someone you met once, a week ago. It can be very uncomfortable to text someone you met once a year ago. Colder relationships are less useful than warm ones.
  6. Longer time = Stale information. When you first join a community, you might make a mental note that Avery is a great pitch deck designer. Months later, you will likely have forgotten this. The mental mapping of who is in the community and how they can help goes stale.

At On Deck, we were always fighting against these forces. The #asks channel is a case in point: in ODF1, founders were willing to share detailed, private information in this public channel, and people who barely knew them would compassionately help. By ODF18, the #asks channel was still incredibly active, but requests tended to be more transactional, and more people would respond privately.

You can combat community degradation with cohorts, slow growth, and uncompromising admissions standards

There are clearly exceptions to this rule — communities that seem to only get better with age and scale. But for any community, there are 3 tools that can be used to combat degradation:

  1. Cohorts. A cohort (or “batch”, or “class”) is a community-with-a-community. This lets new joiners form their own intimate social bonds at a manageable scale, instead of trying to graft them onto a monolith.
  2. Uncompromising admissions standards. If your community starts with the top people in a field (the proverbial “best of the best”) then there is much lower risk of community degradation. This is why the networks of Harvard and Stanford, YC, Rhoades Scholars, etc. have such longevity. If there is nowhere else for the best people to go, they will keep coming to you, as long as you don’t drop your quality bar.
  3. Slow/linear growth. The best way to break Rule #2 is to try to grow faster than your supply of legit people. If your demand for legit people outstrips supply, you’ll get mean reversion and flight. (YC is currently testing the limits of this; it will be interesting to see how it plays out).

There is also a 4th category here: software. Many of the destructive forces that act on a community — forgetting who is relevant, relationships going cold, noise — could be fought with the right platform. At this point, you begin to transform from a community into a social network.

I have yet to see a tool that perfectly handles the community-to-social-network transition. But there is no theoretical barrier to it — several companies have actually done it, like Facebook and LinkedIn (before messing things up with their ads-driven business models). With some of the titans of the last decade finally falling, I expect we will see new, large entrants into this arena in the coming years.

Conclusion: Castles vs. Colliders

I made you wait a long time to see what I mean by “castles vs. colliders” in the title of the post. Thanks for making it this far.

In the last two years, my mental model of community has shifted in a pretty substantial way.

Coming into ODF, I thought of the community as a castle. When people enter your castle, they wear your livery, they fly your flag, and they stay within your walls. People in ODF would proudly brandish their membership on their LinkedIn and Twitter, and we optimized for trying to keep as many people engaged as possible as the community grew (effectively keeping them “in the castle”).

A dense medieval city enclosed by high stone walls — the "castle" model of community.

Today, my mental model is of a particle collider. Community is a place where people are temporarily smashed together and new relationships can form.

Particle-collision tracks spraying out from a central impact — the "collider" model of community.

After the initial collision, it’s up to community members to take those relationships where they want. People participate in lots of different communities, smashing into each other again and again in different settings. This is a good thing.

I’ll conclude with a bold statement: In the long term, there are no communities, just individual relationships. Joining a community is a great way to immediately intersect with helpful people. If you have open slots in your social circle, you may take some of those people and — with care and attention — turn them into your friends, colleagues, or supporters.

In the long term, a shared community affiliation can lower the friction of meeting someone new, and calibrate your level of trust.

But just being part of a community does not guarantee friendship or help. That comes down to you as an individual, and how much thought and care you put into tending your own relationships.

I hope you enjoyed reading this. Do you have examples, counter-examples, or observations that should have made the list? Are you building a social network or community platform that makes use of these dynamics? If so, feel free to reach out!

[1] What kind of community are we talking about? “Community” means many things to many people. The word has been used to describe your local neighbors, a Saturday morning running club, your church or temple congregation, the people you regularly DM with on Twitter, or the 500k person ecommerce Discord you joined.

To clarify, though: the type of community that I’m talking about is a closed-access professional community, i.e. the kinds that I’ve built and run before. These typically cater to a specific industry or title, have a selective bar for admission, are sometimes (but not always) pay-to-join, and are directly or indirectly focused on generating commercially useful relationships.

[2] Even people who seem like they might be mutually relevant on the surface may have less in common than you think. We saw this at On Deck: even amongst the narrow-seeming category of “early stage founders”, relevance was also a factor of sector, business model, and funding raised.

[3] Disclaimer: I am not a rocket scientist 🚀 and this may be completely wrong.

Get future essays

One email when a new piece goes up.