PostHog

PostHog vs Amplitude is the analytics-layer decision for almost every PLG SaaS we talk to. The honest split is not "open-source vs enterprise" — it's stage, team shape, and whether your event taxonomy is governed yet.

Typical fit: who each tool is built for

Typical PostHog customer - Indie or Series A–B PLG SaaS, under ~1M MAU, founder or eng-led GTM. - Wants one tool covering product analytics, session replay, feature flags, LLM observability. - Engineering owns the SDK; no full-time analyst yet. - Budget sensitive; pay-as-you-go preferred over annual commit. - Operator pattern, not vendor claim: AI-native startups consolidating LangSmith- or Helicone-style LLM cost tracking into the same tool that runs funnels.

Typical Amplitude customer - Series B+ PLG or multi-product SaaS with a named analytics or RevOps owner. - Event taxonomy is written down; PQL criteria exist; CRM writeback is owned. - Experimentation is a discipline (program, not one-off A/Bs). - Procurement requires SCIM, audit logs, SOC 2, DPAs, and master agreements. - Operator pattern, not vendor claim: teams ready to use Amplitude AI Global Agent + MCP into Claude/Cursor on clean data — not as a substitute for writing definitions down.

Neither if you're… - An in-app-guidance buyer — your gap is not analytics. See Userpilot or Pendo. - A sales-led outbound shop without a product surface — these are not pipeline tools.

When PostHog wins

PostHog wins when the question is "what are users doing, and can we ship and measure a fix this sprint?" — and one bill is worth more than enterprise governance.

• Input: PostHog SDK direct or Segment/RudderStack ingress; server-side capture for backend events.
• AI step: Max AI for natural-language analytics, replay auto-summaries, LLM-obs span tagging on token cost and latency. Narrower surface than Amplitude's agentic positioning — and intentionally so.
• Human review: Engineer or PM validates events and flag rollouts before prod; RevOps reviews cohort definitions before any CRM writeback.
• Writeback: Cohort exports to Customer.io or HubSpot lists; Slack/Linear webhooks on funnel anomalies; flag toggles in-app; replay deep links in tickets.
• Metric: Activation by cohort, flag-variant conversion lift, $/feature on LLM spans, replay-watch time per CSM.

Concrete wins: AI-native startup wanting one tool for analytics + LLM cost; PLG team consolidating replay + flags + analytics; regulated industry that needs self-host.

When Amplitude wins

Amplitude wins when the question is "is our analysis trustworthy enough to drive CRM routing and forecasting?" — and governance, experimentation depth, and AI on clean data justify the bill.

• Input: Instrumented events from web/mobile, Segment/RudderStack ingress, account/group properties, CRM traits via native syncs.
• AI step: Global Agent answers questions and builds charts; specialized agents monitor dashboards, replay, feedback, web conversion; MCP exposes data to Claude/Cursor/ChatGPT and can write to Jira/Linear/Notion.
• Human review: Analytics/RevOps validates taxonomy and approves cohort definitions before CRM sync; CSM interprets health signals before outreach.
• Writeback: Audiences to Braze, Customer.io, HubSpot, Salesforce; Slack/email agent alerts; MCP-driven tickets.
• Metric: Cohort conversion lift, PQL→Opp rate, health-score precision vs. churn, time-to-insight on ad-hoc questions.

Concrete wins: Series C+ multi-product org with named analysts; procurement requiring SCIM + audit; experimentation program shipping dozens of variants per quarter.

When you need both

Most teams don't run both — the analytics layer is a foundation, not a shared workload. Exceptions:

• Migration windows. Run PostHog and Amplitude in parallel 30–60 days during a migration to validate cohort parity before cutover (detailed below).
• Embedded telemetry. Some teams keep PostHog for in-product LLM observability and dev telemetry while standardizing GTM analytics on Amplitude. Operationally honest, but doubles event-instrumentation discipline.

For most readers, pick one analytics layer. Compose guidance on top via Userpilot or Pendo; compose CRM writeback via Hightouch. See the CSM health score playbook and the RevOps lead scoring playbook.

Pricing and per-account math

Sources: PostHog pricing and Amplitude pricing (both checked 2026-06-14). Amplitude Startup Scholarship gives one year of Growth free if you have less than $10M funding and under 20 employees.

Crossover math (verify against your own event volume):

• At 200K MTUs / ~20M events/mo with light replay, PostHog stays well under Amplitude Plus.
• Around 1M MTUs / 100M events/mo with heavier replay sampling and LLM obs, PostHog's metered bill begins to approach Amplitude Plus or low-end Growth.
• Past 10M MTUs, Amplitude's Growth/Enterprise contracts often beat metered PostHog if you also need governance, experimentation, and AI agent surface — otherwise the math stays in PostHog's favor for analytics-only.

Do not buy on price alone past 1M MTUs. Buy on whether you have the analytics-team maturity to use Amplitude's depth, or whether you'd rather invest that headcount elsewhere and stay on PostHog.

Feature overlap and gaps

Reading this matrix: PostHog leads on bundled LLM obs and self-host; Amplitude leads on governance, experimentation, and AI-agent surface area. Neither is missing the other's headline feature — the depth and operating cost are what differ.

The buying mistakes we see most

1. Buying Amplitude before taxonomy discipline. Procurement signs Enterprise while three squads disagree on "Signed up." Global Agent inherits the ambiguity and ships confident wrong cohorts to Salesforce. Fix taxonomy first; upgrade when definitions are written down.
2. Outgrowing PostHog without a migration plan. Event volume doubles, autocapture stays on globally, the bill spikes. Set ingestion filters and an event allow-list before traffic scales; model the curve quarterly.
3. Treating PostHog replay or LLM obs as enterprise-grade. Competitive at sub-1M MAU; not feature-equivalent at Series C scale. If replay is a governance object (legal, support, security), audit depth before sign-off.
4. Buying Amplitude Enterprise before Plus discipline. SCIM and mutual-exclusion cohorts while still on spreadsheet lifecycle definitions. Plus is enough until taxonomy is stable.
5. Self-hosting PostHog without an infra owner. Postgres + ClickHouse + Kafka ops cost real platform-eng time. Cloud wins on total cost for most teams; self-host is a regulatory call, not a budget one.

What to test in week 1

PostHog test (engineering-led, ≤5 days):

1. Pick one activation metric tied to revenue ("completed onboarding step 3 within 24h"). Document the event definition.
2. Instrument (or confirm autocapture caught it); build the funnel; sample 5 dropped users from replay.
3. Ship one flag-controlled tweak to the broken step.
4. Measure: activation lift, replay-watch time per CSM, total elapsed time vs. prior stack.

Amplitude test (RevOps + analytics-led, ≤5 days):

1. Pick one PQL or expansion cohort tied to revenue ("used feature X twice in 14 days").
2. Document event definitions in a shared doc; fix the top duplicate or orphan instrumentation issue.
3. Build the cohort; manually review 10 accounts against CRM records.
4. Sync a test audience to CRM or post a weekly Slack summary — human approved.
5. Measure: % of cohort accounts where CRM activity matched product truth; time spent vs. prior manual pull.

If step 3 of the Amplitude test fails (taxonomy is too messy to trust), do not enable AI agents at scale — fix data first. The same principle applies to PostHog at scale.

Migration and coexistence

Bidirectional migration is real in this category. Run the new tool in parallel for 90 days before cutover:

• Event parity: Re-instrument the top 20 events in the new tool's SDK. Debug at the SDK level, not just at Segment.
• Cohort parity: Define the same 5 production cohorts in both tools and reconcile counts weekly. 5% delta is normal; 30% means an instrumentation gap.
• CRM writeback: Keep production writeback on the incumbent until parity holds two consecutive weeks; cut over one cohort first.
• Replay + flags: Treat as separate migrations. Many teams keep PostHog flags + Amplitude analytics during transition (or vice versa) — operationally honest, double the maintenance.
• Contract risk: Amplitude Enterprise has co-term and renewal cliffs; PostHog pay-as-you-go has none, but skipping spend caps is its own risk.

If you outgrew the smaller tool, migrate. If an AI demo wowed a VP, run the week-1 test first.

FAQ

Does PostHog's LLM observability replace LangSmith or Helicone? For cost and latency tracking, yes. For deep prompt versioning, evals, and trace-level debugging, LangSmith and Helicone still go further.

Is Amplitude's Global Agent an AI SDR? No. It informs GTM with product usage; it does not send cold email or replace Salesforce-native sequencers. See the AI account research use case for where analytics agents actually fit in the GTM stack.

Can RevOps use PostHog as a CDP? For lightweight cohort syncs to Customer.io or HubSpot, yes. For enterprise CRM writeback at SLA, pair PostHog with a dedicated reverse-ETL layer like Hightouch.

Does Amplitude include session replay at the free tier? Yes — included on Starter as of the 2026 platform packaging. Verify on the pricing page before committing replay-heavy workflows.

Does gtmpod earn commission on either tool? Yes on PostHog; no on Amplitude (employer disclosure instead). Both stated above.