an experiment on maintaining high growth DAOs’ community stickiness
“A decentralized autonomous organization (DAO) is an organization represented by rules encoded as a
computer program that is transparent, controlled by the organization members and not influenced by a
central government”, Wikipedia.
I view each DAO as a living organism/city-state that grows with its members in the web3 world/metaverse.
As the concept of DAO begins to seep through more fields and professions outside of the web3 context,
bigger DAOs begin to grow and expand at an unprecedented pace. However, there are little tools in the
current ecosystem that can support the highest growth DAOs and there is still a lot of friction when users
interact in these emergent entities.
After joining now one of the biggest DAOs, Friends with Benefits (LINK), in August, I realized that it is
extremely hard to make friends and feel at home in this new found community. Although I’ve asked
questions, participated in discussions, joined calls, and followed some members on twitter, it has been
extremely hard to find people I can genuinely talk to and bond with in this community. It felt like moving to a new city all by my self with no jobs, no friends, and no outlet to meet anyone in person (I currently live
in a city with no FWB presence). Moreover, it is very intimidating to speak out in a large group chat/
channelswhen you are new to the community, making everything that much harder. I’d imagine that this
has happened and will happen to many organizations’ newcomers. The sad truth is that no matter how
great the on-boarding process or how active the community is, it is the tool/current way of building that
ultimately limits the size of active members. Therefore, I ran a three day experiment, similar to Lauren’s
SOMEONE (LINK), acting as a content curation bot for FWB on three of my friends to see if they would
react positively to my messages.
DAO comes from the two Chinese characters 道/岛, which stands for “way” and “island” respectively.
This represents the purpose of this project - to find a way to help model DAOs’ social model after the equivalent of a physical location.
The very core of this project is to view/design the social aspect of a community model as the equivalence
of its physical counterpart. We note that the community model in cities, villages, tribes, etc. are all self
sustaining, clearing out the old and make way for the new. However, in the case of high growth online
communities, we face the unique challenge of morphing the community model from tribes to cities in
very little time. This inherently leads to the issue of the social system lagging behind in relation to the
size of the growing community. Therefore, a positive loop that catalyzes members of the community to
engage is extremely important. This leads to our bot that will try to push the right content to the right
members in an attempt to increase the likeliness of community engagement.
key questions / answers:
( how do we measure timeliness?
see WHEN users engage
-> NEED read reciepts
-> NEED feedback from user at the end of the experiment
( how do we measure relevance?
determines the wizard of oz match making
-> NEED user’s input
-> NEED to instruct users to engage (like, no effect, dislike)
-> NEED to explain IN the moment
( how do we measure influence? WHEN users engage with content
-> NEED to measure baseline score
-> NEED qualitative measurement
( how do we obtain personal goals? assuming they would join a community with a purpose or some interest in mind
-> NEED user’s input through a survey like questionarie
main goal and key questions:
( main goal: to understand the effects of engagement with timing and frequency
( key question : How might we create a self-sustaining online community?
( key question : How might we build a symbiotic relationship between the community and the user?
( key question : How might we maintain community stickiness during high growth periods?
( key question : How might we emulate the behavior of the CUI while still sound human?
( Users would engage with the community if they can see the relevant content
( A content filtration algorithm would block out the noise for individual user
Note* no conversation can be displayed due to the FWB policy & other privacy concerns
The testing session was ran for three days for three different testers, a previous survey was sent out to
gain insights on when to send messages and which channel each participant preferred.
tester1 / metaverse, de-fi, social tokens, promote your stuff
tester2 / trading crypto, learning crypto, setup
tester3 / nft-alpha-leaks
After determining the topics for each tester, I tested on the timing, frequency, and density of each
day1 / 1-time summary of previous day’s activity
day2 / divided evenly into 2/3 threads per notifications
day3 / heavily curated information based on personal knowledge
During each of the interactions, testers had to react postively, negatively, or had no effect. From the extrapolation of patterns in the data, we could generate our feedback and eventually find the impact
these factors had on testers.
Timing did not matter too much, since testers replied fairly quickly. In general, replies tend to happen
when people are more available. However, the prompt replies may be due to the fact that testers
were close friends of mine and didn’t take me as a bot as much as they would with a real bot.
time slot1 / around lunch time (fastest average response time)
time slot2 / morning/breakfast (slowest average response time)
time slot3 / evening/night time (middle average response time)
Frequency mattered greatly, and testers reacted best when frequency is higher than just once per
day, but still low enough to be magnitudes lower than the amount of notifications testers recieve in
the channels they have chosen.
once per day / 2 no effect and 1 positive
3-5 times per day / 2 positive and 1 no effect
10 times per day / 2 no effect and 1 negative
Density correlated heavily with frequency, which meant that it also had a similar result. People
enjoyed the material when it had some density, but not too thick that makes it feels like a bot like
daily newsletter. On the other hand, they didn’t want things that did not have depth at all.
high density / 2 no effect and 1 positive
mid density / 2 positive and 1 no effect
low density / 2 no effect and 1 negative
Testers’ reaction to how and which content is chosen and presented followed our basic intuition.
The more personalized catering messages were better received, while the ones that are bot like
and felt like a summary had little to no interactions
summary / no interaction
summary filtered / some interaction
personally chosen / some to high interaction
( Content mattered the most out of everything we have tested
( Timing does not play a vital role, but sending it when testers are available has increased effect
( Lower frequency (less spammy) of interaction resulted in higher engagement overall
( Frequency played less of a role when the content was relevant
( Acting like a human was more important than expected
( Relevancy can be created by the context the bot presents information in
( Density of each interaction impacted engagement
The overall testing was fairly dissatisfying as no generalizable patterns were discovered during the testing
process. The fact the bot had to be extremely different and vary from person to person means that it
requires a great deal of testing and engineering for us to achieve a community catalyst that helps the
social model to grow at the pace of the community. There are some further testings I want to do, but I also
believe that many DAOs have not really considered about the issue they will face in times of scaling.
Therefore, I stand by my point of view of building online communities like cities and will want to keep
investigating on how to construct a catalyst that can speed up the morphing time of an online
community’s social system.
Questions to consider (Final Project) :
A more extended and elaborate period of testing to gain more insights
Generalized patterns across different groups needs to be identified
Interactions using the bot between multiple members within a community need to be tested