How to Measure Protocol Retention Right

TL;DR:

• A returning wallet is not automatically a retained user.
• Anchor cohorts on the first completed value action, not the first transaction.
• Read retained actors, retained economic value and retained core behaviour together.

Protocol retention starts with the product job

The usual retention chart is easy to build and easy to misread. It counts a wallet that returns in a later period, then calls the result stickiness.

That can be directionally useful. It is not enough for an operating decision.

For a protocol, the retained action has to match the product's actual job. A repeated swap, a maintained lending position, a second liquidity action, or a renewed stake can all be meaningful. A dust transaction or an automated router interaction is not necessarily evidence of a returning customer.

I start with one question: what action would make the team genuinely believe this user received value and chose to come back?

The cohort anchor changes the answer

Do not anchor a cohort on the first approval, page touch, or submitted transaction. Anchor it on the first completed value-bearing action.

That definition will differ by protocol. A DEX may use a completed swap above a reasonable threshold. A lending protocol may use a confirmed borrow or deposit. A liquidity product may use an LP position that remains active long enough to be meaningful.

This is the same distinction that mattered in Morpho's migration analysis. The useful success metric was not a button click or a submitted transaction. It was a validated migrated position. Once we combined the product funnel with onchain settlement, we could see the product issue behind the metric and fix it. Ethereum migrated liquidity went from roughly $6M to roughly $32M, a 433% increase.

Retention needs the same discipline. Count the action that represents actual value, then study who repeats it.

Keep wallet and entity views separate

Wallets are observable and useful. They are not people.

One actor can use several wallets. A router, vault, multisig, market maker, or bot can create activity that is technically real but irrelevant to a user-retention decision. I would rather show a wallet-level view plus a filtered or entity-adjusted view, with the assumptions visible, than produce a single clean curve nobody can defend.

The aim is not perfect identity resolution. It is decision-grade confidence. Segment known contracts and obvious automated behaviour. Document exclusions. Keep an entity model conservative.

Retain behaviour and value, not just addresses

The useful retention read has three layers:

1. Retained actors: wallets or identified entities that return.
2. Retained core behaviour: repeat swaps, deposits, borrows, positions, votes, or stakes.
3. Retained economic value: recurring fees, volume, balances, TVL, or borrow value.

Those curves will often disagree. That disagreement is usually the insight.

In a private analysis of concentrated-liquidity pools, fewer than 1% of LP wallets controlled roughly 82% of cohort TVL. A dashboard that celebrated returning addresses without showing concentration would have missed the operating risk. The right leadership question was not whether liquidity appeared to return. It was whether the protocol could retain resilient liquidity if a small number of entities changed behaviour.

Segment before you call a result healthy

Break retention down by acquisition source, chain, contract path, asset, and user type. Incentivised wallets, aggregators, direct users and sophisticated LPs should not be blended into one cohort.

Then compare the week or month after incentives decline. If wallet recurrence disappears with rewards, that is not a failure to hide. It is a decision input for incentive design, product work, and growth spend.

The dashboard should make the next question easier to answer: which source, segment, or product step deserves attention?

A retention model people can act on

For every headline retention metric, record the cohort anchor, reporting entity, filters, return window, source tables, owner, and limitation. Review it on the cadence that matches the product. Weekly retention may make sense for trading. Monthly or rolling retention may be better for lending, staking, or LP activity.

The goal is not a prettier curve. It is a stable definition that product, growth, finance, and leadership can use to decide what to change next.

If your protocol needs a retention model that connects wallet behaviour to product and economic value, see Unchain Data's Product Analytics service.