Choose Mixpanel or Amplitude when product teams need to explore behavior through a broad analytics environment. Build a focused reporting layer when a few critical questions depend on definitions, account relationships, or operational joins unique to your product. Begin with an event contract: a clicked button, a started import, and a completed import are not interchangeable outcomes.
Both products now span more than charts. That makes older comparisons built around “analytics versus experimentation” unreliable. The decision should be based on the workflow and entitlements your team will actually use, with the same event stream and identity rules in each trial.
Compare the measurement contract
Official sources were reviewed September 26, 2026. Vendor capabilities are cited; the custom column describes Looski's proposed limited implementation. These are not hands-on accuracy or performance benchmarks.
| Criterion | Mixpanel | Amplitude | Build your own |
|---|---|---|---|
| Current free allowance | Public Free plan lists up to one million events per month. Pricing | Public Free plan lists two million events per month. Pricing | Collection, storage, and query costs still grow with event volume. |
| Funnel analysis | Funnels are a documented analysis surface. Funnels | Funnel Analysis supports defined conversion steps. Funnel setup | Implement a named journey with explicit ordering and conversion window. |
| Identity | Simplified ID Merge documents linking device and user identities. Identity | MTU guidance explains anonymous/known identities and merging. Identity and billing | Own identity rules, deduplication, corrections, and auditability. |
| Seats | Free plan lists unlimited seats. Pricing | Current pricing describes unlimited seats across plans. Pricing | Access control and onboarding become application responsibilities. |
| Session replay | Current Free offer lists a replay allowance. Plans | Replay joins observed sessions to analytics context. Replay | Retain a replay service if visual diagnosis is important. |
| Experimentation | Current Free/Growth offerings include flags and experimentation with limits. Plans | Current plans include experimentation with package-specific limits. Plans | Keep assignment and statistical analysis separate from ordinary funnel reporting. |
| Billing interpretation | Compare event, replay, and experimentation usage separately. Pricing | Event-priced offers coexist with documentation for MTU-priced contracts. Billing | Budget ingestion, retention, compute, monitoring, and analyst support. |
| Account-level question | Test group/account definitions with the intended plan. | Test account definitions with the same dataset and entitlement. | Join users to accounts using effective dates and explicit business rules. |
| Late events | Verify the configured report's treatment of event time and later arrivals. | Verify the same conversion-window behavior. | Define when a result is provisional and when it is finalized. |
| Best reason to extend | Preserve exploratory analysis while adding authoritative business joins. | Preserve exploratory analysis while adding authoritative business joins. | A trusted answer to a specific operational question, not every possible chart. |
Do not mix events, users, and experiments in a price comparison
The current public pages use event allowances prominently. Amplitude's MTU documentation remains relevant to MTU-priced plans: an MTU is a unique user with at least one event in a calendar month, with identity rules affecting the count. It would be misleading to label every current Amplitude offer “priced by users” or apply an old Starter allowance to the present Free plan. Amplitude MTU guide, current pricing.
Obtain an exact quote for your event volume, replay volume, experiments, retention requirements, and governance needs. Record currency, term, included usage, and overage behavior. The September 26, 2026 source review establishes current pricing structures; it does not establish the cost of your tracking plan. Mixpanel calculator, Amplitude billing rules.
| Cost or migration item | Evidence required | Why it matters |
|---|---|---|
| Instrumentation | Event inventory, frequency, properties, and ownership | A noisy tracking plan increases cost without necessarily answering more questions. |
| Identity | Anonymous-to-known examples and account membership changes | Different identity rules can change both reporting and billing. |
| Historical migration | Export format, timestamps, IDs, and supported import semantics | Copying events does not automatically recreate old cohorts and charts. |
| Parallel operation | The same input stream and reconciliation queries | Differences must be explained before the old source is retired. |
| Custom analytics | Query optimization, backfills, tests, and support | A correct first query is not a maintained analysis platform. |
| Pilot | Known journeys with expected results | Chart agreement is meaningful only when definitions also agree. |
Buy exploration when questions change every week
Our analysis favors a commercial platform when product managers routinely ask new segmentation, conversion, retention, and behavioral questions. The ability to investigate without requesting a new report is part of what you buy. A fixed custom dashboard might answer today's question quickly while slowing every question that follows.
Run the trial with the people who make decisions. Ask them to explain a drop-off, narrow it to a meaningful cohort, inspect supporting evidence, and share the conclusion. Record where they need engineering help. Do not select solely on how attractive the default dashboard looks.
Make the funnel definition explicit before comparing totals. Specify who enters, whether order matters, the allowed time between steps, and what counts as success. Use a server-confirmed import completion when the business question concerns a completed import. Browser clicks can be useful diagnostic evidence without becoming the authoritative completion event.
Build the business definition that deserves to be owned
A bounded custom layer could answer whether a newly activated account completes its first successful import within an agreed period. Store the account's activation event, import attempts, final outcomes, and relevant plan state. Join membership changes carefully so a user moving between accounts does not rewrite history accidentally.
Keep raw events and derived measures separate. Give every measure a definition version, owner, and test cases. If a definition changes, show whether historical values were recomputed or only future values use the new rule. Otherwise a dashboard can appear to improve because its denominator changed.
Use stable event IDs to deduplicate retries. Preserve event time and receipt time to investigate delayed arrival. Mark recent cohorts as incomplete when they have not had the full opportunity to convert. A conversion rate that ignores that exposure difference can mislead even when every query runs correctly.
Retain mature collection, replay, experimentation, or exploration tools where they continue to save work. A shared event contract can feed both the vendor platform and a warehouse-backed custom report; ownership does not require immediately retiring every analytics subscription.
A pilot with answers known in advance
Create test journeys for an anonymous visitor who signs in, a user on two devices, a failed import followed by success, a duplicated completion event, an out-of-order event, and an account that never finishes. Include a user whose account membership changes during the observation window.
Calculate expected results by inspecting the underlying records. Reconcile each platform and the custom query against those expectations. Investigate disagreements instead of averaging the numbers or assuming the larger count is more accurate.
Measure time to a defensible answer, effort required to correct instrumentation, and how often an analyst can explain a result without developer intervention. Buy the platform that supports exploration; build the few business definitions and joins that must remain stable across tools.
Frequently asked questions
Can a custom dashboard replace product analytics?
It can replace a defined set of recurring analyses. It does not automatically replace exploratory analysis, experimentation, or every team's questions. List the supported decisions and expand only when the value is clear.
What would a first custom product analytics workflow include?
Define a small event vocabulary, stable account identifiers, and the exact success condition. Store events with timestamps and schema versions, then produce a reproducible journey report linked to operational outcomes. Expose a narrow interface for the product decisions the team makes repeatedly.
How should we compare the cost of Mixpanel, Amplitude, and a custom build?
Include instrumentation and data-quality work in all options. For custom analytics, estimate storage, query compute, and the continuing analyst time required to maintain definitions. Compare the same scope and planning horizon, including implementation, retained services, maintenance, and support. A lower subscription bill alone does not establish a lower total cost.