In this article

An account keeps completing its weekly work, but the people doing that work have changed. A user-retention chart calls the original users lost; an account-retention chart calls the customer retained. Both can be correct. Amplitude’s account-level reporting makes the distinction explicit: groups, such as companies or teams, can be the unit of analysis instead of individual people. For a collaborative product, the useful question is which kind of continuity you need to understand.

Keep the customer and the person in view

A person can stop using a product while their employer continues to rely on it. A colleague may take over the work, a role may change, or a team may consolidate activity around fewer people. Conversely, several individuals may stay active while the account no longer completes the job for which it bought the product.

An account-return measure should therefore name the behavior that qualifies the account. “Someone logged in” answers a weaker question than “the team completed its recurring task.” Start with the customer job, then decide which events establish that it happened and which account they belong to.

Mixpanel’s Group Analytics allows metrics to aggregate by a group identifier rather than a user identifier. That changes the entity being counted. It does not automatically give the chosen event a better business meaning.

The same activity produces two different retention stories

Consider a hypothetical collaborative reporting product. Ten users across four customer accounts complete a report in the starting month. In the following two months, count a person or an account as returning only if they complete at least one report in that month. All identifiers below are invented.

AccountStarting-month report completersMonth 1Month 2
AA1, A2A1, A2A3, A4
BB1, B2, B3B1B1
CC1, C2C1, C2None
DD1, D2, D3D1, D2, D3D1, D2, D3, D4

In this scenario, A1 and A2 leave account A before month 2 and A3 and A4 join it. D4 joins D. Users do not belong to multiple accounts.

Measure from the same hypothetical ledgerMonth 1Month 2
Starting users who return8 of 104 of 10
User return rate80%40%
Starting accounts that return4 of 43 of 4
Account return rate100%75%
All users active in that month87

User return rate = original users completing the task in the later month ÷ original users. Account return rate = original accounts with at least one task completion in the later month ÷ original accounts.

Month 2 has seven active users, but only four belong to the original ten-user cohort. Counting all seven as retained would turn new participation into returning participation. The account result is different because new members can continue an existing customer’s work.

Read the account patterns before prescribing a fix

Account A shows complete turnover of its original active people while the account continues the task. That calls for an investigation into continuity: was the handover intentional, and can the new team operate without the former members? A win-back message to A1 and A2 would not answer those questions.

When new people take over an existing customer’s work, onboard invited teammates from the account’s prepared state while measuring each person’s first useful action separately.

Account B has a different pattern. The account still qualifies as retained, but participation has narrowed from three people to one. That might be efficient specialization or unwanted dependence on a single person. The activity record alone does not decide which. Talk to the account and examine whether the remaining user can complete the intended workflow reliably.

Account C has no qualifying activity. Investigate whether the job disappeared, moved elsewhere, or stopped being completed. Account D has continued use from all original members and one new contributor. That is broader participation, but it is not proof of expansion revenue or an intention to renew.

The account-level total usefully compresses these patterns. It also hides them. Keep the underlying participation view available when the aggregate changes or when the next action depends on which people are involved.

Attribute the event to the account where the work happened

Identity becomes especially important when a consultant, administrator, or other user belongs to several customer accounts. Assigning all of that person’s activity to every account would manufacture continuity. The event needs the account context in which the work took place.

Amplitude distinguishes event-level groups from group membership associated with a user, and notes that changes to group assignments and properties apply to new data. Its account-reporting documentation is a reminder to define the rule before interpreting the chart. Decide how transfers, mergers, deleted accounts, and reused identifiers will be handled, and preserve enough history to explain a change.

For work shared across people, an account may also complete a sequence that no single person completes alone. Amplitude permits that in a group-level funnel. Decide whether collective completion fits the actual customer job; do not use it merely because it produces a larger number.

Pair continuity with participation

Use account return to ask whether the customer organization continues doing the relevant work. Use original-user return to ask whether the people in the starting cohort continue doing it. Then inspect participation within the account when the two diverge.

Those views support different product decisions. A handover problem may need transferable ownership and clearer permissions. Concentrated use may need either stronger collaboration or better support for a specialist role. A lost recurring task may need a different product intervention entirely. The choice becomes easier once the chart identifies whose continuity it describes.

Sources

Something we should clarify?

Suggest a correction