A Knowledge Drawback in Disguise.

Not too long ago, Starkware initiated their much-awaited airdrop. Like, most airdrops it resulted in a ton of controversy. Which in a tragic method, would not actually shock anybody anymore.

So why is it the case that this retains on occurring over and over? One may hear just a few of those views:

• Insiders simply need to dump and transfer on, cashing out billions

• The workforce did not know any higher and did not have the suitable counsel

• Whales ought to have been given extra precedence since they create TVL

• Airdrops are about democratising what it means to be in crypto

• With out farmers, there isn’t a utilization or stress testing of the protocol

• Misaligned airdrop incentives proceed to supply unusual unwanted side effects

None of those view is incorrect, however none of those views are utterly true by themselves. Let’s unpack just a few takes to ensure now we have a complete understanding of the issue at hand.

There exists a basic stress when doing an airdrop, you are selecting between three components:

• Capital Effectivity

• Decentralisation

• Retention

You usually find yourself in a state of affairs the place airdrops strike effectively in 1 dimension, however hardly ever strike steadiness between even 2 or all 3. Retention particularly is the toughest dimension with something north of 15% usually unprecedented.

• Capital effectivity is outlined as the standards used to what number of tokens you give to a participant. The extra effectively you distribute your airdrop, the extra it should turn into liquidity mining (one token per greenback deposited)—benefiting whales.

• Decentralisation is outlined as who will get your tokens and below what standards. Current airdrops have adopted the strategy of going for arbitrary standards as a way to maximise the protection of who will get stated tokens. That is usually factor because it saves you from authorized troubles and buys you extra clout for making individuals wealthy (or paying for his or her parking fines).

• Retention is outlined as how a lot do customers stick round after the airdrop. In some sense it is a approach to gauge how aligned had been your customers together with your intent. The decrease the retention, the much less aligned your customers had been. 10% retention charges as an business benchmark imply just one in 10 addresses are literally right here for the suitable causes!

Placing retention apart, lets study the primary 2 in additional element: capital effectivity and decentralisation.

To know the primary level round capital effectivity, let’s introduce a brand new time period referred to as the “sybil co-efficient”. It principally calculates how a lot you profit from splitting one greenback of capital throughout a sure variety of accounts.

The place you lie on this spectrum will finally be how wasteful your airdrop will turn into. If in case you have a sybil coefficient of 1, it technically means you are working a liquidity mining scheme and can anger a lot of customers.

Nonetheless whenever you get to one thing like Celestia the place the sybil coefficient balloons out to 143, you are going to get extraordinarily wasteful behaviour and rampant farming.

This leads us to our second level round decentralisation: you need to finally assist “the small man” who’s an actual person and taking the prospect to make use of your product early — regardless of them not being wealthy. In case your sybil coefficient reaches too near 1 then you are going to be giving near nothing to the “small man” and most if it to the “whales”.

Now that is the place the airdrop debate turns into heated. You may have three lessons of customers that exist right here:

1. “small guys” who’re right here to make a fast buck and transfer on (perhaps utilizing just a few wallets within the course of)

2. “small guys” who’re right here to remain and just like the product you have made

3. “industrial-farmers-who-act-like-lots-of-small-guys” right here to utterly take most of your incentives and promote them earlier than transferring to the subsequent factor

3 is the worst, 1 continues to be form of acceptable and a couple of is perfect. How we differentiate between the three is the grand problem of the airdrop drawback.

So how do you resolve for this drawback? Whereas I haven’t got a concrete resolution, I’ve a philosophy round learn how to resolve this that I’ve spent the previous few years fascinated by and in addition observing first-hand: project-relative segmentation.

I will clarify what I imply. Zoom out and take into consideration the meta-problem: you’ve all of your customers and also you want to have the ability to divide them up into teams primarily based on some form of worth judgement. Worth right here is context-specific to the observer so will range from undertaking to undertaking. Making an attempt to ascribe some “magical airdrop filter” is rarely going to be enough. By exploring the info you can begin to grasp what your customers actually appear to be and begin to make data-science primarily based selections on what the suitable approach to execute your airdrop is thru segmentation.

Why does nobody do that? That is one other article that I will be writing sooner or later however the very lengthy TLDR is that it is a laborious knowledge drawback that requires knowledge experience, money and time. Not many groups are prepared or in a position to try this.

The final dimension that I need to speak about is retention. Earlier than we speak about it, it is most likely finest to outline what retention means within the first place. I might sum it as much as the next:
“

quantity of people that an airdrop is given to
---------------------------------------------
quantity of people that preserve the airdrop

What most airdrops make the basic mistake of is making this a one-time equation.

With a purpose to reveal this, I assumed some knowledge may assist right here! Fortunately, Optimism has really executed on multi-round airdrops! I hoped I might discover some simple Dune dashboards that gave me the retention numbers I used to be after however I used to be sadly incorrect. So, I made a decision to roll up my sleeves and get the info myself.

With out overcomplicating it, I wished to grasp one easy factor: how does the share of customers with a non-zero OP steadiness change over successive airdrops.

I went to: https://github.com/ethereum-optimism/op-analytics/tree/predominant/reference_data/address_lists to get the checklist of all addresses that had participated within the Optimism airdrop. Then I constructed a bit of scraper that might manually get the OP steadiness of every handle within the checklist (burned a few of our inside RPC credit for this) and did a bit of information wrangling.

Earlier than we dive in, one caveat is that every OP airdrop is impartial of the prior airdrop. There isn’t any bonus or hyperlink for retaining tokens from the earlier airdrop. I do know the explanation why however anyhow let’s stick with it.

Given to 248,699 recipients with the standards accessible right here: https://group.optimism.io/docs/governance/airdrop-1/#background. The TLDR is that customers got tokens for the next actions:

• OP Mainnet customers (92k addresses)

• Repeat OP Mainnet customers (19k addresses)

• DAO Voters (84k addresses)

• Multisig Signers (19.5k addresses)

• Gitcoin Donors on L1 (24k addresses)

• Customers Priced Out of Ethereum (74k addresses(),

After working the evaluation of all these customers and their OP steadiness, I obtained the next distribution. 0 balances are indicative of customers who dumped since unclaimed OP tokens had been despatched on to eligible addresses on the finish of the airdrop (as per https://dune.com/optimismfnd/optimism-airdrop-1).

Regardless, this primary airdrop is surprisingly good relative to earlier airdrops executed that I’ve noticed! Most have a 90%+ dump fee. For less than 40% to have a 0% steadiness is surprisingly good.

I then wished to grasp how every standards performed a job in figuring out whether or not customers had been more likely to retain tokens or not. The one problem with this system is that addresses may be in a number of classes which skews the info. I would not take this at face worth however reasonably a tough indicator:

One time OP customers had the best proportion of customers with a 0 steadiness, following customers who had been priced out of Etheruem. Clearly these weren’t one of the best segments to distribute customers to. Multisig signers had been the bottom which I believe is a superb indicator since it is not apparent to airdrop farmers to setup a multi-sig the place you signal transactions to farm an airdrop!

This airdrop was distributed to 307,000 addresses however was quite a bit much less considerate imo. The standards was set to the next (supply: https://group.optimism.io/docs/governance/airdrop-2/#background):

• Governance delegation rewards primarily based on the quantity of OP delegated and the size of time it was delegated.

• Partial gasoline rebates for energetic Optimism customers who’ve spent over a certain quantity on gasoline charges.

• Multiplier bonuses decided by extra attributes associated to governance and utilization.

To me this intuitively felt like a nasty standards as a result of governance voting is a straightforward factor to bot and pretty predictable. As we’ll discover out beneath, my instinct wasn’t too off. I used to be stunned simply how low the retention really was!

Near 90% of addresses held a 0 OP steadiness! That is your ordinary airdrop retention stats that individuals are used to seeing. I might love to enter this deeper however I am eager to maneuver to the remaining airdrops.

That is by far one of the best executed airdrop by the OP workforce. The standards is extra subtle than earlier than and has a component of “linearisation” that was talked about in earlier articles. This was distributed to about 31k addresses, so smaller however simpler. The main points are outlined beneath (supply: https://group.optimism.io/docs/governance/airdrop-3/#airdrop-3-allocations):

• OP Delegated x Days = Cumulative Sum of OP Delegated per Day (i.e. 20 OP delegated for 100 days: 20 * 100 = 2,000 OP Delegated x Days).

• Delegate should have voted onchain in OP Governance throughout the snapshot interval (01-20-2023 at 0:00 UTC and 07-20-2023 0:00 UTC )

One vital element to notice right here is that the standards for voting on-chain is AFTER the interval from the final airdrop. So the farmers that got here within the first spherical thought “okay, I am completed farming, time to maneuver on to the subsequent factor”. This was sensible and helps with this evaluation as a result of take a look at these retention stats!

Woah! Solely 22% of those airdrop recipients have a token steadiness of 0! To me this alerts the waste on this airdrop was far lower than any of the earlier ones. This performs into my thesis of retention being vital and extra knowledge that having multi-round airdrops has extra utility than individuals given credit score for.

This airdrop was given to a complete of 23k addresses and had a extra attention-grabbing standards. I personally thought the retention of this may be excessive however after fascinated by it I’ve a thesis for why it was most likely decrease than anticipated:

• You created participating NFTs on the Superchain. Complete gasoline on OP Chains (OP Mainnet, Base, Zora) in transactions involving transfers of NFTs created by your handle. Measured throughout the trailing one year earlier than the airdrop cutoff (Jan 10, 2023 – Jan 10, 2024).

• You created participating NFTs on Ethereum Mainnet. Complete gasoline on Ethereum L1 in transactions involving transfers of NFTs created by your handle. Measured throughout the trailing one year earlier than the airdrop cutoff (Jan 10, 2023 – Jan 10, 2024).

Absolutely you’d suppose that individuals creating NFT contracts can be indicator? Sadly not. The info suggests in any other case.

Whereas it is not as dangerous as Airdrop #2, we have taken a reasonably large step again by way of retention relative to Airdrop #3.

My speculation is that if they did extra filtering on NFT contracts that had been marked as spam or had some type of “legitimacy”, these numbers would have improved considerably. This standards was too broad. As well as, since tokens had been airdropped to those addresses instantly (reasonably than having to be claimed) you find yourself in a scenario the place rip-off NFT creators went “wow, free cash. time to dump”.

As I wrote this text and sourced the info myself, I managed to show/disprove sure assumptions I had that turned out to be very invaluable. Particularly, that the standard of your airdrop is instantly associated to how good your filtering standards is. People who attempt to create a common “airdrop rating” or use superior machine studying fashions will fail susceptible to inaccurate knowledge or a lot of false positives. Machine studying is nice till you attempt to perceive the way it derived the reply it did.

Whereas writing the scripts and code for this text I obtained the numbers for the Starkware airdrop which can be an attention-grabbing mental train. I will write about that for subsequent time’s submit. The important thing takeaways that groups ought to be studying from right here:

1. Cease doing one-off airdrops! You are capturing your self within the foot. You need to deploy incentives form of like a/b testing. A number of iteration and utilizing the previous’s learnings to information your future goal.

2. Have standards that builds off previous airdrops, you are going to improve your effectiveness. Really give extra tokens to those who maintain tokens on the identical pockets. Make it clear to your customers that they need to stick to at least one pockets and solely change wallets if completely obligatory.

3. Get higher knowledge to make sure smarter and better high quality segmentation. Poor knowledge = poor outcomes. As we noticed within the article above, the much less “predictable” the standards, the higher the outcomes for retention.

In case you’re actively pondering of doing an airdrop or need to jam about these items, attain out. I spend all my waking hours fascinated by this drawback and have been for the previous 3 years. The stuff we’re constructing instantly pertains to all the above, even when it would not appear so on the floor.

Facet notice: I have been a bit out of the loop with posting resulting from poor well being and many work. Which means content material creation usually finally ends up sliding off my plate. I am slowly feeling higher and rising the workforce to make sure I can get again to having a daily cadence right here.