12th March, 2024

Onchain Retention for Smart Accounts

How sticky are smart accounts? Safe joined up with BundleBear for a deep analytics dive into the retention rates of ERC4337 accounts across the ecosystem.

Clément Bihorel



In its fairly recent history, the web3 community and its protagonists seemingly acknowledged the need for solving a key barrier to adoption: the user experience that Externally Owned Accounts (EOAs) convey.

To improve the overall user experience and accessibility of web3, the idea of Account Abstraction (AA) was introduced to web3 enthusiasts. There is plenty of existing literature and explainers on the topic, so we will not dive into the details of what AA is in this article.

In a nutshell, AA aims to decouple EOAs and user interactions from blockchain systems, and instead use personal smart contract accounts as the main means of interacting with blockchains. AA opens up new security, flexibility and security paradigms. In this domain, Safe and the 4337 standard have been leading the way to adoption.

If AA improves the usability and accessibility of web3, great! What do we build?

The late 2010s’ fuelled the dreams of many to decentralise everything. From web2 consumer applications to supply chain tracking and energy trading, an immense array of use cases were envisioned as being prone to disruption through the prism of decentralisation. Initial Coin Offerings (ICOs) promised new ownership structures for blockchain based projects and their participants. Social media, transport, messaging and other commodities were meant to undergo a technological revolution, the revolution of decentralisation.

Some years down the line, the adoption of crypto-based products in the day-to-day life of a regular internet user has not reached the traction it had initially hoped for. The mass market has not yet migrated away from the web2 behemoths for the benefit of a more decentralised world or improved self-agency. The reasons for this are multifold and couldn’t possibly be summarised in this article, but one key reason is that the web3 ecosystem does not seem to have broken through as shipping applications that users genuinely want to use and engage with, or build the tools and infrastructure to attract teams experienced in doing so.

Drawing Parallels between Web2 and Web3

When looking at successful web2 consumer apps with product lenses, we can see that many have one thing in common: they build sticky products that users love to go back to and engage with. If you’re working with a team building a mobile app for instance, you’re first likely going to look at:

  • Number of app downloads

  • Number of active users

  • Retention

  • Customer lifetime value

While web3 comes with default transparency, it’s hard to compare apple to apple any of these key metrics, because blockchains are by definition permissionless and cannot prevent things like sybil attacks. This is particularly important as many apps, protocols and networks bootstrap themselves by creating incentive structures that reward what they assume are users. This has led the web3 industry to gradually become more thoughtful in how these incentives are designed, but there is still much work to do, as the the recent zkSync farming activity suggests.

In the context of smart accounts (and how the space is currently fragmented), we could look at account deployments as a proxy for app downloads (note that this does not hold true as soon as accounts become truly portable). While only tracks 4337 native account deployments, it gives a good overview of account deployments from different providers. Safe deployments can also be tracked here on Dune.

The notion of active users and retention can partially remain the same, looking at the frequency at which a user engages with an application. However while onchain events are public by default, frontend analytics aren’t, so the scope of our analysis is limited to data available on the blockchain.

Customer lifetime value is hardly comparable, as web3 introduces a new paradigm of turning customers into shareholders (or prosumers) of systems they interact with, challenging the definition of customer and the standard calculation methods behind this metric.

With account abstraction unlocked and user experience increasingly expected to match the one of web2, it becomes interesting to look at onchain data to understand current user behaviour, and to gauge the traction of web3 products.

Defining Retention

For this article, we assume that retention is the best onchain proxy to evaluate the stickiness and health of a web3 application. As explained in the introduction, it is currently hard to map other key concepts to the onchain world due to the permissionless nature of these systems. We plan to re-visit the concept of active users, customer lifetime value and deployments in future articles.

According to the most plain and generic definition of retention provided to us by Wikipedia, retention is defined as “…the ability of a company or product to retain its customers over some specified period. High customer retention means customers of the product or business tend to return to, continue to buy or in some other way not defect to another product or business, or to non-use entirely.”

Beyond the abstracted definition (pun intended), retention is often derived to create key performance indicators such as the Customer Retention Rate (more can be found here).

There are actually many documented reasons why a product team would want to optimise their user retention rates. Take for example this HBR article, which suggests that acquiring new customers could be five to 25 times more expensive than retaining existing ones. Furthermore, this research published by Fred Reichheld, the inventor of the Net Promoter Score, shows that increasing customer retention rates by 5% increases profits by 25% to 95%. Other research conducted by Bain even suggests that returning customers spend 67% more over time than first-time customers.

Retention, as an observable concept, is used as a pillar for user behaviour analysis and is used in tandem with measurement and analysis of metrics such as: Customer Retention Rate, Customer Churn Rate, Repeat Purchase Rate etc. (hotjar, user pilot).

Among these, the Customer Retention Rate (CRR) is arguably the most popular and commonly used metric as the basis for further analysis. Most companies strive to get that rate as high as possible, because it captures the company’s ability to retain existing customers and build a loyal customer base. Note here that the notion of a customer is introduced, and that customer in this context are typically defined as individuals or entities which engage in transactions with a particular business.

Some average retention rates for software products, provided by Adjust.Some average retention rates for software products, provided by Adjust.

Some average retention rates for software products, provided by Adjust

Retention rates are also industry dependent, and vary depending on the type of business model your company operates. For instance, social media apps relying on the attention economy theory will sell good and products indirectly to their end users via advertisements, while neo-banks might not value user engagement as much.

Onchain Usage and Retention of Smart Accounts

The onchain world allows us to draw parallels with metrics like Customer Lifetime Value, Repeat Purchase Rate etc. as transactions are recorded publicly. For this article, we decided to focus primarily on usage and retention as these two are closely related and observable events on blockchains.

When users swap, bridge, transfer, mint, deploy etc, these events are immediately available and become public knowledge for anyone using a block explorer and analytics solutions like Dune.

As hinted in the previous sections, user retention is not solely defined through the prism of transactional events. The notion of “active users” is also a pillar of retention and user analysis. In web2, these numbers are typically available through frontend-based analytics and provided by platforms like Google Analytics.

Account Abstraction’s thesis is that all web3 accounts should become smart contract accounts. In 2023, this thesis has led many projects to launch their own smart account offering to try and capture this market. However, while there is undoubtedly growing adoption of the Safe account and the ERC4337 standard, there is little evidence that smart accounts are enabling a new wave of sticky web3 apps and products. And while retention still gives us a sense of user engagement and the overall health of web3 applications and protocols, we should bear in mind:

  • Onchain retention, defined as accounts executing transactions, overlooks that many web3 products might not optimise for onchain events; e.g. web3 social media apps might have high engagement and retention rates while having little onchain conversion events. There is also of course a business case for web3 companies to build and monetise offchain services around and outside of onchain events, for instance through intents, pseudo transactions like UserOps and other infrastructure services.

  • Sponsored and meta transactions, as paymasters and other sponsoring logic now introduce the ability for users to perform onchain actions with virtually no-cost for the user, something previously impossible using Externally Owned Accounts (EOA) only. Account providers, networks, applications and infrastructure teams now run programs for builders where paymasters sponsor user interactions, making it easier and cheaper (free) than ever to interact with an application or protocol for airdrops etc.

These considerations being made, we will now look at ERC4337 active users and retention following a shared convention between Bundlebear and Safe.

How does the current onchain app landscape look and compare in terms of smart account engagement and retention?


In this analysis, Bundlebear and Safe define the retention rate by looking at the onchain activity of newly created accounts week over week, from their first transaction. From a cohort of 100 accounts which first executed a transaction within the same week, we analyse how many of these users made a transaction in the following week, the week after, and so on. This type of retention analysis is sometimes referred to as Cohort Retention Analysis.

ERC4337 Accounts Usage Analysis

ERC4337 is a standard for implementing smart accounts. Since its official launch in March 2023, adoption of the standard has grown rapidly and more than 3 million ERC4337 accounts have been created.

Safe recently soft-launched our official ERC4337 Module, which allows the Safe account to plug and play into the ERC4337 standard, bundlers and paymasters. Our partner Pimlico makes it easy to integrate via their permissionless.js libary, and Safe has plans to roll out more support for ERC4337 via our APIs, SDKs and other services.

In this section, we will:

  • Investigate the size of the existing ERC4337 user base

  • Analyse the onchain activity of ERC4337 accounts

  • Measure the economic activity of ERC4337 accounts

How many ERC4337 accounts are active?

Unlike EOA wallets, ERC4337 accounts don't directly make onchain transactions. They first make UserOperations (UserOps) - meta-transactions that get batched in onchain transactions by offchain transaction builders, also known as Bundlers.

Below, we look at the number of accounts that made at least one UserOp (including deployment) within a month.

Monthly active accounts making at least one UserOp (including deployment) by network.Monthly active accounts making at least one UserOp (including deployment) by network.

Since November of last year, at least 400,000 accounts have sent one pseudo-transaction every month, and February recorded a record high with more than 900,000 accounts doing at least one transaction. Polygon’s share of accounts is 95%, making it the most popular chain in the ERC4337 ecosystem. As per a previous analysis, this could be due to Polygon being among the cheapest chains on which to use an ERC4337 account.

To refine the data above, we start classifying accounts based on the amount of UserOps they sent over a month. This helps us dissociate active accounts from one-time action accounts.

As we previously explained, smart accounts now introduce the ability for projects to sponsor transactions, lowering down the barrier for “activity farming” and not requiring native tokens for users to interact with web3 applications. This can result in accounts deploying and sending only one UserOp to mint an NFT for instance.

Active accounts sending UserOps transactions on a monthly basis.Active accounts sending UserOps transactions on a monthly basis.

In dark grey, we highlight accounts sending more than 5 UserOps and in light grey accounts only sending one UserOp.

In February, there were roughly 127,000 accounts making 5+ UserOps a month. This is 8% of all active accounts over the periods analysed. For comparison, 16% of monthly active Ethereum EOA wallets make five or more transactions a month. Since September 2023, the number of accounts making over 5 UserOps in a month has grown by 11x (from 11,000 to 127,000).

Weekly retention rate of ERC4337 accounts from December 2023 to date.Weekly retention rate of ERC4337 accounts from December 2023 to date.

Since the beginning of the year, the average 4-week retention rate of ERC4337 accounts is 3.16%. This means that 4 weeks after their creation, only 3.16% accounts continue to be sending at least one UserOp per week.

What are ERC4337 accounts doing onchain?

We can categorise UserOps based on the type of application the smart account was interacting with. This categorisation is provided to us by Bundlebear.

Monthly UserOps activity categorised by type of activity.Monthly UserOps activity categorised by type of activity.

NFT drops and social interactions are currently the two most popular type of interactions made with ERC4337 accounts. We can break this down further and segment UserOps by the specific application they are interacting with, as shown below.

Monthly UserOps activity segmented by contract interaction.Monthly UserOps activity segmented by contract interaction.

Since the beginning of the year, the onchain chess game Anichess has been responsible for the large majority of UserOps, followed by other applications including Cyberconnect, a web3 social network responsible for a large number of UserOps in 2023.

How much volume goes through ERC4337 accounts?

While we’ve figured that ERC4337 accounts are popular accounts for minting, claiming NFTs and interacting with games and social applications, we will now look at how much volume they are responsible for.

Monthly ERC20 transfer volume by network in USD.Monthly ERC20 transfer volume by network in USD.

In January, ERC4337 accounts transferred close to $11M worth of ERC20s, which has been so far the all time high. In February, this trend went downward with less than $6M in volume. Ethereum still largely dominates ERC20 transfers by volume, while Polygon is third behind Arbitrum and Ethereum.

Average 4-week retention of ERC20 usage by ERC4337 accounts is 0.62%.Average 4-week retention of ERC20 usage by ERC4337 accounts is 0.62%.

As shown above, the average 4-week retention of ERC20 usage by ERC4337 accounts is 0.62%. On average, only 0.62% of accounts that make their first ERC20 transfer/approval will interact with an ERC20 during the 4 following weeks.

65% of the token transfers made by ERC4337 accounts are worth $10 or less.65% of the token transfers made by ERC4337 accounts are worth $10 or less.

In the graph above, we segment the ERC20 transfers by amount transferred. 65% of the token transfers made by ERC4337 accounts are worth $10 or less. The 5.6% of transfers that are worth more than $100 accounted for 97% of the total transfer volume. The data was analysed from April 1, 2023 to January 31, 2024.

Monthly native transfer volume is dominated by Ethereum, followed by Arbitrum and Polygon.Monthly native transfer volume is dominated by Ethereum, followed by Arbitrum and Polygon.

If we compare ERC20 volume to native token transfer volume (i.e. ETH transfers on Ethereum, Optimism, Arbitrum, Base and MATIC transfers on Polygon), ERC20 volume is much greater. In January and February, native transfer volume ranged roughly between $500 and $600K.

Once again, Ethereum dominates the volume despite only having a small percentage of active accounts and UserOps. Arbitrum is the second biggest chain for transfer volume with 16% share followed by Polygon with 8%.

The average weekly retention of native transfer activity by ERC4337 accounts.The average weekly retention of native transfer activity by ERC4337 accounts.

The average 4-week retention of native transfer activity by ERC4337 accounts is only 1.73%. There is a lot of room for improvement here particularly because native transfer retention is weaker than the overall average (2.17%) and slightly better the ERC20 average (0.62%).

Power law remains: 94% of ERC4337 native transfers are worth less than $10.Power law remains: 94% of ERC4337 native transfers are worth less than $10.

94% of native transfers are worth less than $10. Similar to ERC20 transfers, there is a power law where the 2.4% of transfers account that are worth $100+ create 96% of the dollar-denominated volume (measurements made between April 2023 and February 2024).

Overall, we can observe that when looking into ERC20 and native transfers, the retention numbers are very low. As mentioned in the introduction, this could mean that “activity farming” drives a lot of activity, with users minting and interacting with applications without having to bear the costs themselves and favouring low value activity.

What types of accounts are people using?

Most developers building wallets and applications on top of ERC4337 use smart account providers instead of building their own. We can determine which of these providers are most popular by looking at account deployment data.

Monthly smart account deployment by factory.Monthly smart account deployment by factory.

In February, close to 500,000 accounts were deployed through Zerodev factories, followed by Biconomy with over 300,000 deployments. These two providers currently cover over 90% of the deployments.

Monthly UserOps sent grouped by Factory.Monthly UserOps sent grouped by Factory.

However, when looking at UserOps, one can observe that Biconomy is actually leading over Zerodev. In February, close to 2 million transactions were coming from Biconomy accounts while Zerodev covered less than 600,000 transactions.

ERC4337 Account Usage Insights Summary

To summarise, we’ve observed that:

  • Since the beginning of the year, the average 4-week retention rate is only 3.16% for ERC4337 accounts

  • Including deployment data, February recorded an all time high in terms of active accounts with more than 900,000 accounts deployed

  • In February, 127,000 accounts made over 5 UserOps. This type of active account has been growing steadily since December

  • ERC20 transfers recorded an all time high in volume in January with roughly $11M in volume, and native ETH transfer remain very low comparatively

  • Ethereum still accounts for the biggest volume of both ERC20 and native transfers

Now that we’ve looked at ERC4337 accounts and how they interact with the onchain world, we will now look at Safe’s relative activity. As a reminder, Safe only recently released its ERC4337 module, as a means for the accounts to interact with the ERC4337 ecosystem. For a long time however, Safe has supported a wide array of other relaying systems, including Gelato, which allows our partners to offer relayed and sponsored transactions.

Safe Account Usage Analysis

In this section, we will dive into Safe account usage and retention. The analysis includes the onchain activity of both Safe{Core} and Safe{Wallet}.

We will run a similar analysis than the one applied to ERC4337 accounts, though on Safe accounts. Safe accounts also leverage pseudo transactions (similar to UserOps), as the account is designed around a multi-signature scheme. This scheme allows developers and users to guarantee higher levels of security than most ERC4337 accounts.

While this article looks at account usage and retention, overall the Safe smart account ecosystem dominates the market by all measures; over 7 million Safe accounts were created, close to 40 million Safe transactions were executed and the Safe accounts are currently securing over $110 billion in assets. That’s comparable to the amount stored in the ETH Staking contract.

Similar to the previous section, we will aim to answer the following:

  • How many Safe accounts are active?

  • What are Safe accounts doing onchain?

  • How much volume goes through Safe accounts?

How many Safe accounts are active?

Safe accounts that made at least X txs per month to calculate active/inactive accountsSafe accounts that made at least X txs per month to calculate active/inactive accounts

On the first of February, over 224,000 Safe accounts fell under the definition of “active”, meaning sending at least one transaction per month. Note that even when carrying a Month over Month analysis of active accounts, both the cumulative and new active account count is growing. The new active account count is growing at an average 40% Month over Month, while the total number of active accounts has also been growing, from 9% in October 2023 to 30% in February 2024.


Conducting a weekly cohort analysis on Safe accounts since the beginning of the year, we notice that while the retention rate at week 4 greatly varies between cohorts, with a December cohort going as high as 72.94%, retention always stays above 10%. This is 3 times higher than the average ERC4337 account.

What are Safe accounts doing onchain?

When querying the most used onchain functions of Safe accounts; we can identify, by order of usage, ETH transfer, transfer, swap and verifyproof as the most used transactional functions (other called functions are rather internal functions to the contract). The following cohort analyses show the retention of accounts calling the same function, week over week, and the retention of selected cohorts for these functions throughout time.

On week 4, the average retention rate of accounts calling ETH transfer is 11.66%. On week 4, the average retention rate of accounts calling ETH transfer is 11.66%.
The average retention rate on week 4 for the transfer function is 34.77%.The average retention rate on week 4 for the transfer function is 34.77%.

The transfer function is the second most used function by Safe accounts. As the table above suggests, the retention rate for accounts calling this function on week remains above 11% at its lowest, and 73.36% at it highest. The average retention rate on week 4 for this function is 34.77%, meaning that 1 out of 3 account execute a transfer on the first week will still execute a transfer 4 week after the account was created.

The average retention rate on week 4 for the swaps function is 28.22%.The average retention rate on week 4 for the swaps function is 28.22%.

When looking at the average retention rate in week 4, we also observe an average 28.22% retention rate for swaps.

The average retention rate on week 4 for the verifyProof function is 21.53%.The average retention rate on week 4 for the verifyProof function is 21.53%.

verifyProof is a call executed by providers to verify the account owner against an identity, it is commonly found in identity solutions like Worldcoin and Polygon ID (source). The average retention rate in week 4 is also above average at 21.53%.

From this analysis, we can observe that ERC4337 accounts behave differently than Safe accounts onchain. While Safe accounts are typically used to perform higher value, financial or identity related transactions, ERC4337 accounts appeared to be primarily used for minting NFTs and engaging in low value transactions, with much weaker retention rates, almost like “disposable” accounts.

How much volume goes through Safe accounts?

Source: AlliumSource: Allium

ETH transfers represent the biggest volume of transfers coming from Safe over the past months. Over the past 6 months, the total USD value of transfers done in ETH has been consistently growing, to reach a staggering $3 billion of transfers in February.

ERC20 volume from Safe accounts over months.ERC20 volume from Safe accounts over months.

The chart above depicts the ERC20 volume from Safe accounts over months. In February, ERC20 transfers coming from Safes accounted for close to $14M in volume, a down from over $22M in January. Comparatively to ERC4337 native accounts, Safe ERC20 transfers are roughly two times more voluminous (Source: Dune).


The chart above depicts the relationship between the transfer size and the number of accounts performing the transfer. 400 thousand accounts transferred between 1 and 10 USD while close to 2 million accounts transferred over 100 USD from November 26, 2018 to February 29, 2024.

Safe Account Usage Insights Summary

To summarise, we’ve observed that:

  • The average 4th week retention rate for Safe accounts always stays above 10%, hence 3 times higher than the average ERC4337 account

  • The most used functions currently are; ETH Transfer, transfer, swap and verifyproof

  • In February, ETH native transfers currently represented over 3 billion USD in volume

  • In January and February, Safe ERC20 transfers are roughly two times more voluminous than ERC4337 native accounts

Beyond Retention

While onchain analytics can yield some interesting insights and we could dig further into the details explaining spikes in usage and retention by looking at individual applications (hint: we will), the scope of onchain analysis remains limited compared to offchain analytics. As the space grows in maturity, we can hope to witness the rise of more advanced analytics, similar to how this Farcaster dune dashboard merges both offchain and onchain events under one unified view.

However, with web3’s ethos being to bring the world onchain and upgrades like Dencun making it cheaper than ever to interact onchain, we have the unique opportunity to craft and shape the meaning of existing and new concepts that relate to smart account retention. Provided the importance of account portability and the permissionless nature of web3, we even have the opportunity to question and put the concept of retention under scrutiny. Some examples:

  • Defining metrics that analyse the relationship between the transactional behaviour of accounts and the value they hold

  • Formalising with data the notion of prosumer (rethinking customer value)

  • Identifying correlations between the type of events emitted by an account and its stickiness

However while these explorations represent exciting opportunities, they’re at a similar risk than accounts facing fragmentation. Without cooperation on defining the standards and finding a common nomenclature for defining onchain retention, we will make it increasingly difficult to communicate clearly about smart account usage.

This first article kicked off a series of explorations that we hope will bring forward more analyses of onchain smart account products. For more extensive data deep dives, look out for upcoming quarterly reports with Messari, with the first one published here.

Talk to us

Together with our partners and the ecosystem, Safe aims to explore and refine the nomenclature of onchain retention and account usage, in a mission to make smart accounts the default way of living onchain. Part of this work recently translated into our analytics team building standard templates to help our partners measure the health of their engagement and the ecosystem they build. If you’re a team interested in the topic, give us a shout, we’d love to talk with you.

Let’s build

If you’re a project looking to build a product using smart accounts and ERC4337, or any other standard, come talk to us! World-class projects like Worldcoin, Gnosis Pay, Rabby, Setter, (and many more we look forward to announce over the coming weeks), are already building with us.

Special thanks to: Daniel Partida, Peter, and Kofi for their contributions.

Read more




Safe Modular Smart Account Architecture – Explained




ERC-4337 integrations now supported in Safe{Core}’s Account Abstraction stack




Pimlico & Safe: Advancing Account Abstraction with Launch of ERC-4337 Module


Get the Alpha

Sign up to hear the latest from Safe in your inbox