PostHog vs Mixpanel: should an onboarding experiment count exposed users or flag requests?

Both now support native flags and experiments. For onboarding, check exposure timing and identity first, then distinguish PostHog request allowances from Mixpanel monthly experiment users. Teams with existing Mixpanel data need not migrate because of outdated impressions.

Source: posthog.com
November 2, 2026
6 min

When replacing a long onboarding form with a step-by-step flow, the question is whether the new flow improves first-time completion, not which analytics platform has the longer feature list. PostHog and Mixpanel now both offer native feature flags and experiments. Teams already using Mixpanel should not start a migration based on its past product reputation. A more useful comparison is which platform can reuse existing events, correctly identify users who actually see onboarding, and keep costs manageable until the experiment ends. This article draws on official documentation; it does not include experiments run in practice or performance tests.

Separate assignment from seeing onboarding

Imagine users land on the home page after logging in and enter onboarding only when they click “Create your first project.” The application may request a flag and obtain an assignment on the home page, but that does not mean the user has seen the new form. If everyone receiving an assignment is included in the onboarding experiment, the team may actually be answering a different question. Before launch, specify eligibility, the page or action at which exposure is recorded, and the point from which completion events are counted. This example is experimental-design advice, not a result already produced by either platform.

PostHog’s exposure documentation states that the standard SDK path records $feature_flag_called when getFeatureFlag is called or a flag is read from server-side evaluateFlags results. Experiments identify participants through exposure events, and metric events before exposure are excluded from analysis. The call location is therefore part of the experiment definition. If the application must fetch a value in advance, but participation should begin only upon entering onboarding, consider a custom exposure event with the correct variant properties. A successful flag API request alone does not establish that participant counts are accurate.

Mixpanel’s documentation also distinguishes assignment from exposure: the former determines the version assigned, while the latter occurs when the application actually uses that version to render. It offers native flags and can track exposure through $experiment_started. Teams with an existing CDP event pipeline can also implement assignment and exposure reporting separately. This option helps retain existing collection paths, but responsibility for sending events must be explicit to avoid one path reporting automatically and another logging the same event again. Compare actual SDKs and instrumentation rather than treating “supports experiments” as completed integration.

Next, verify identity. Mixpanel allows variant assignment by user distinct_id, device device_id, or a group key; this assignment key cannot be changed after the flag is enabled. If onboarding spans anonymous visits and registration or login, decide first whether the study concerns the same person, the same device, or the entire company account. PostHog also notes that multiple devices, cleared storage, or evaluation before and after identifying a user can expose the same user to multiple versions. For this scenario, use test accounts to follow anonymous, login, refresh, and device-switching paths and check that exposure and completion events still correspond correctly. This matters more than deciding whose reports look better.

One million requests is not one million experiment users

PostHog’s current pricing page lists monthly free allowances of one million product analytics events and one million feature flag requests, with experiments billed through flags. There are at least two usage ledgers here: events and flag requests. This is not one million free experiment users, nor should every in-app flag read be treated directly as a billable network request. Estimate the budget using the SDK, caching, and actual calling pattern, then examine overage pricing. A small number of frequent visitors and a large number of one-time visitors can produce different usage profiles.

Mixpanel Free currently includes one million events per month, 1,000 monthly experiment users (MEU), and up to ten active flags. Growth includes 5,000 MEU and up to 50 active flags. The page’s statement that it can scale to 100,000 MEU does not mean 100,000 are included by default. MEU counts distinct users exposed within a calendar month. Official rules also state that one user participating in up to 30 experiments counts as one MEU, with separate rules beyond that range. When evaluating an onboarding experiment, first estimate the number of actual participants rather than substituting site-wide event volume.

Allowances affect more than the bill. Mixpanel’s experimentation documentation states that when Free reaches its MEU limit, experimentation pauses until the next monthly reset or an upgrade, while flags continue serving traffic. Growth overages involve billing. If 2,000 users are expected to enter onboarding in a month, do not assume Free can cover the entire experiment even if analytics events remain far below one million. This headcount is only a budgeting example, not evidence of enough users to detect a particular improvement. Sample requirements still depend on the baseline, the change to be detected, and experimental design.

With reliable events already in place, check the extra work first

If a team already maintains trustworthy first-completion events, funnels, and user identity in Mixpanel, checking whether its current native experimentation meets requirements is often more sensible than rebuilding the entire data setup for one onboarding experiment. Its experiment reports can select existing metrics and analyze exposure events generated by external flags. If migration is needed, list the work involved in historical metric definitions, identity continuity, and parallel data validation rather than comparing only the numbers in two free plans.

If the team already uses flags and product analytics in PostHog, it can reuse the same event and experiment pipeline. Its experiment documentation also links variants to session replay, making it easier to inspect user experiences after metrics change. For teams starting from scratch, this connected workflow is worth piloting, but it does not automatically eliminate exposure definitions or data-quality work. Do not enable collection that the current experiment does not need just to use the platform’s full feature set.

Finally, compare both using the same acceptance criteria: who is eligible, when exposure occurs, how completion events map to identity, whether expected usage covers the observation period, and who deploys and rolls back changes. Teams already on Mixpanel should prioritize validating incremental integration; teams already on PostHog should prioritize reusing their existing pipeline. Only a clear capability or cost gap warrants migration. What makes an onboarding experiment trustworthy is whether participants and metrics align, not treating flag requests as the number of people who saw the new flow.

Content note:This article is based on publicly available information and is intended as a reference for evaluating tools. Features, pricing, and related policies may change at any time; please refer to the latest official information.

Original source:posthog.com