Product Analytics & A/B Testing 2026 — Mixpanel / Amplitude / PostHog / Heap / Optimizely / Plausible / Segment Deep Dive

"Data beats opinions, but opinions beat no data." — Jim Barksdale, ex-Netscape CEO (the line Mixpanel and Amplitude have quoted on their blogs more than any other)

As of May 2026, the product analytics market has hardened into a four-camp structure that no single tool spans: managed SaaS (Mixpanel, Amplitude), open-source self-hosted (PostHog, Matomo), privacy-first (Plausible, Fathom, Umami), and data infrastructure (CDPs: Segment, RudderStack, Snowplow). A company that takes data seriously usually runs two or three of these at once — for example, Segment for collection → Mixpanel for analysis → Statsig for experiments.

This post covers more than 25 products in one pass: positioning, pricing, API surface, limitations, and market events including "the tool that died in 2024 (June)" and "the tool that got acquired in 2024 (Hotjar → Contentsquare)." We explain why a solo developer picks Plausible, why a Series B startup begins with PostHog, and why an enterprise still cannot leave Adobe Analytics.

1. The 2026 product analytics map — managed / open-source / privacy-first / CDP

Product analytics is no longer "drop a GA snippet on the page." In 2026, the surface area a company actually touches looks like this.

The 2026 landscape collapses into four camps.

• Managed SaaS: Mixpanel and Amplitude still dominate. Both shipped AI features (Spark, Audience GenAI) in 2025, and as their pricing kept climbing, more Series B/C startups defected to PostHog.
• Open-source all-in-one: PostHog grew explosively in 2024 and 2025. One platform now covers analytics, session replay, feature flags, experiments, and LLM observability. Matomo is the older self-hosted incumbent, the de facto standard for European governments and public agencies.
• Privacy-first: As GDPR and ePrivacy enforcement tightened and cookie banner fatigue peaked, "no cookies, no PII, just enough traffic data" became the new default for SaaS marketing sites. Plausible, Fathom, and Umami own this slice. (June shut down in 2024 — covered below.)
• CDP (Customer Data Platform): Segment (under Twilio, with the perennial 2024 spinoff rumor) still leads. Open-source RudderStack and warehouse-native Snowplow challenge from below. The 2026 trend is "warehouse-native CDP" — making the warehouse the single source of truth.

The first decision branch is always "how much self-hosting effort can you take on?" PostHog is free if you self-host, but it eats an SRE's time. Mixpanel is expensive but one click and you're done.

2. Mixpanel — the long-time leader, still the standard in 2026

Mixpanel was founded in 2009 and effectively created the "event-based analytics" category. In 2026 it remains the standard benchmark for managed analytics — the phrase "a Mixpanel-style analytics tool" is used as a common noun.

Core product lineup (May 2026):

• Events: Event-plus-property analytics, unlimited event definitions
• Funnels: Multi-step conversion funnels, A/B comparison
• Cohorts: User segment definitions, exportable to external tools
• Retention: N-day retention, per-cohort retention curves
• Impact Report: Causal estimation of how a feature launch moved a metric
• Spark (2025): GenAI-driven natural-language querying — "what share of users who signed up last week converted to paid?"
• Session Replay (2024): Late entrant, lighter than Fullstory
• Warehouse Connectors: Query Snowflake, BigQuery, or Redshift directly (no reverse ETL)

Pricing (May 2026):

• Free: 1M events per month, unlimited users, 1-year retention
• Growth: Usage-based on events beyond 1M (from about $20/month)
• Enterprise: Quoted, with SSO, SOC2, HIPAA, dedicated instances

Strengths:

• The most polished analytics UX. Funnel and retention visualizations are a notch above the competition.
• Cohort Sync pushes segments out to Braze, Iterable, HubSpot, and similar marketing tools.
• The 2025 Spark feature really does work in natural language (subjectively more accurate than Amplitude AI).

Limitations:

• Usage-based pricing means a traffic spike blows up the bill.
• Mobile SDKs lag behind Amplitude (especially Flutter and React Native).
• Data governance features are weaker than Amplitude Govern.

// Mixpanel JavaScript SDK 2026
import mixpanel from 'mixpanel-browser'

mixpanel.init('YOUR_PROJECT_TOKEN', {
  debug: false,
  track_pageview: 'url-with-path',
  persistence: 'localStorage',
  api_host: 'https://api-eu.mixpanel.com', // EU data residency
})

// Identify user
mixpanel.identify('user_42')
mixpanel.people.set({
  $email: 'alice@example.com',
  plan: 'pro',
  signup_date: new Date().toISOString(),
})

// Track event
mixpanel.track('Checkout Completed', {
  amount: 99.0,
  currency: 'USD',
  items: 3,
})

Server-side tracking follows the same design.

# Mixpanel Python SDK (backend)
from mixpanel import Mixpanel
import os

mp = Mixpanel(os.environ['MIXPANEL_TOKEN'])

mp.track('user_42', 'Subscription Renewed', {
    'plan': 'pro',
    'mrr': 49.0,
    'tenure_months': 14,
})

3. Amplitude — expanding into experimentation, the governance leader

Amplitude was founded in 2012 and went public in 2021 (NASDAQ: AMPL). It is Mixpanel's direct competitor and from 2024 through 2025 it differentiated along two axes: experimentation (Experiment) and data governance (Govern).

Core product lineup:

• Analytics: Events plus cohorts plus funnels, roughly at parity with Mixpanel
• Experiment: A/B testing plus feature flags, built on ClearBrain (acquired 2022)
• CDP: Built from Iteratively (acquired 2022) as a Segment alternative
• Govern (2024): Event taxonomy with approval workflow, enterprise data catalog
• Audience GenAI (2025): Build cohorts in natural language — "users who signed up last month but didn't return this week"
• Session Replay (2024): Late entrant
• Data Assistant: Natural-language querying (Mixpanel Spark equivalent)

Pricing (May 2026):

• Starter (Free): 50k events per month, limited cohort analysis
• Plus: From $61 per month (about$750 annually), 10k MTU (Monthly Tracked Users)
• Growth / Enterprise: Quoted, with SSO, HIPAA, SOC2, EU data residency

Strengths:

• Best-in-market mobile SDKs — React Native, Flutter, iOS, and Android are all first-class.
• Experiment integration means "how did this experiment affect retention" is a one-click analysis.
• Govern is just about the only solution that stops event-definition chaos at the enterprise scale.
• The free tier is enough to start serious analytics work.

Limitations:

• The UI is heavier than Mixpanel and the learning curve is steeper.
• The Experiment stats engine is simpler than Statsig or Optimizely.
• MTU-based pricing escalates fast for high-traffic B2C.

// Amplitude Browser SDK 2 (v2, 2024)
import { init, track, identify, Identify } from '@amplitude/analytics-browser'

init('YOUR_API_KEY', {
  serverZone: 'EU',
  defaultTracking: {
    sessions: true,
    pageViews: true,
    formInteractions: true,
    fileDownloads: true,
  },
})

// Update user properties
const identifyEvent = new Identify()
identifyEvent.set('plan', 'pro')
identifyEvent.add('login_count', 1)
identify(identifyEvent)

// Track event
track('Checkout Completed', {
  amount: 99.0,
  currency: 'USD',
})

The Experiment SDK is imported separately.

// Amplitude Experiment
import { Experiment } from '@amplitude/experiment-js-client'

const exp = Experiment.initializeWithAmplitudeAnalytics('DEPLOYMENT_KEY')

await exp.start({ user_id: 'user_42' })

const variant = exp.variant('checkout-button-color')
if (variant.value === 'green') {
  // Render the new button
}

4. PostHog — open-source all-in-one (analytics + replay + flags + experiments + LLM)

PostHog is a British startup out of Y Combinator (2020) that closed its Series D ($430M valuation) through 2024 and 2025 and grew explosively. As of May 2026 it is the most successful execution of the "everything on one platform" strategy.

What PostHog ships in a single product:

• Product Analytics: Events, funnels, retention, cohorts (a Mixpanel and Amplitude alternative)
• Session Replay: User screen recording (a Fullstory and Hotjar alternative)
• Feature Flags: Feature flags, gradual rollouts (a LaunchDarkly alternative)
• Experiments: A/B tests with a built-in stats engine (an Optimizely and Statsig alternative)
• Surveys: In-app surveys (a Hotjar Surveys alternative)
• Web Analytics (2024): GA4 alternative, cookieless mode supported
• LLM Observability (2025): Track OpenAI and Anthropic API calls, evaluate prompts
• Data Warehouse (2024): Sync BigQuery, Snowflake, Stripe, and HubSpot; run SQL inside PostHog
• CDP (2025, beta): Forward events to external tools

Pricing:

• Cloud Free: 1M events, 5,000 session replays, and 1M flag calls per month, free
• Cloud Paid: Usage-based on overage (about $0.00045 per 1M events)
• Self-hosted: Free, with Hobby (single-node) and Production (K8s) variants
• Enterprise: Quoted, with SSO, SAML, SOC2, and dedicated instances

The open-source license is MIT for most modules, with a separate license for cloud-only features (some SSO, audit logging, and the like).

Strengths:

• Truly all-in-one. You can skip or delay buying other tools (a Series B can ship on PostHog alone).
• Self-hosting actually works (Docker Compose in one line).
• Pricing comes out to roughly one-third to one-fifth of competitors.
• LLM Observability shipped in June 2025 and exploded in popularity among AI startups.

Limitations:

• Mobile SDKs are a tier behind Amplitude in maturity.
• The UI is sometimes heavy (the cost of cramming everything into one screen).
• Data governance (anything like Govern) is weak — event-definition chaos arrives quickly.

// PostHog JavaScript SDK
import posthog from 'posthog-js'

posthog.init('phc_YOUR_PROJECT_KEY', {
  api_host: 'https://eu.posthog.com', // or https://app.posthog.com
  person_profiles: 'identified_only', // skip profile creation for anonymous users (cheaper)
  capture_pageview: true,
  capture_pageleave: true,
  session_recording: {
    maskAllInputs: true,
    blockClass: 'ph-no-capture',
  },
})

// Event
posthog.capture('Checkout Completed', {
  amount: 99.0,
  currency: 'USD',
})

// Feature flag
if (posthog.isFeatureEnabled('new-checkout-flow')) {
  // Render the new flow
}

// Experiment (get the variant)
const variant = posthog.getFeatureFlag('button-color-test')
// variant === 'control' | 'green' | 'blue'

A 2025 server-side (Python) pattern for tracking LLM calls became popular.

# PostHog LLM Observability (2025)
from posthog import Posthog
from openai import OpenAI

posthog = Posthog(
    project_api_key='phc_YOUR_KEY',
    host='https://eu.posthog.com',
)

# Auto OpenAI integration (tracks request, response, tokens, cost)
from posthog.ai.openai import OpenAI as PHOpenAI
client = PHOpenAI(posthog_client=posthog)

resp = client.chat.completions.create(
    model='gpt-4o-mini',
    messages=[{'role': 'user', 'content': 'hello'}],
    posthog_distinct_id='user_42',
    posthog_trace_id='conversation_abc',
    posthog_properties={'feature': 'support-chat'},
)

5. Heap — the autocapture pioneer, acquired by Contentsquare in 2024

Heap was founded in 2013 and acquired by Contentsquare in 2024 (exact figure undisclosed, estimated around $500M). It pioneered the "capture every event with zero code" category and in 2026 remains the standard for it.

How it works:

• Drop the Heap SDK in one line and every click, form input, and pageview is captured automatically.
• Later, in the UI, you say "name this button click 'Signup Clicked'" — retroactive definition.
• As a result, the three-week "add tracking code → ship → wait for data" loop disappears.

Core features:

• Autocapture: All interactions captured automatically
• Snapshot Recording: Session replay (added 2022)
• Journey: Visualization of the actual user path (Sankey diagram)
• Effort: Measures how often a user repeats the same action (a frustration signal)

Pricing: Undisclosed, quoted. Roughly from $12,000 per year for small accounts, with enterprise commonly at$50k or more.

Strengths:

• The fastest analytics setup — drop it on the marketing site and you have data the next day.
• Retroactive definition lets you "answer questions you didn't ask in advance."
• After the Contentsquare acquisition, integration with heatmaps and UX analytics is underway.

Limitations:

• Pricing is expensive and opaque.
• Autocapture produces noisy data (spam clicks, bots — filtering is a chore).
• Mobile autocapture lags the web.
• From a "data governance" perspective it is an anti-pattern — events are never defined up front.

// Heap autocapture (one line)
window.heap = window.heap || []
heap.load('YOUR_HEAP_APP_ID')

// Identify user
heap.identify('user_42')
heap.addUserProperties({ plan: 'pro' })

// Explicit events are optional (you can mix them with autocapture)
heap.track('Subscription Renewed', { mrr: 49.0 })

6. Pendo — the standard for product adoption and in-app guides

Pendo was founded in 2013 and acquired by Thoma Bravo in 2021 ($2.6B). As of May 2026 it is the de facto standard in the B2B SaaS in-app guide and onboarding category. "You've seen the Pendo, right?" is a running joke among SaaS PMs.

Core products:

• Product Analytics: Similar to Mixpanel and Amplitude, though shallower
• Guides: In-app guides, tooltips, walkthroughs, onboarding checklists — no code required
• Feedback: In-app NPS, feature request voting
• Roadmaps: Public roadmaps (a Productboard alternative)
• Listen (2024): Voice-of-customer analysis, GenAI summaries
• Adopt: For internal employees (measures adoption of SaaS rolled out by HR/IT)

Pricing: Undisclosed, quoted. Roughly from $20,000 per year, with enterprise commonly at$100k or more.

Strengths:

• You can change in-app UX without shipping code — the PM operates it directly.
• Feature adoption and active-user metrics connect across analytics and guides.
• Deeply entrenched in enterprise SaaS like Workday and Salesforce.

Limitations:

• Very expensive.
• Analytics depth lags Mixpanel and Amplitude.
• No "drop a single line on the site" lightness — setup is heavy.

Competitors: Appcues (small to mid), Userpilot (best value), WalkMe (enterprise, most expensive), Chameleon (developer-friendly), Userflow (rising in 2024).

7. Contentsquare + Hotjar — UX analytics plus heatmaps, consolidated

Contentsquare was founded in France in 2012, raised Series E ($500M) in 2021, **acquired Hotjar in 2023** (estimated$500M), and acquired Heap in 2024. As of May 2026 it is unambiguously the leading group in UX analytics.

Positioning of the three products:

• Contentsquare: Enterprise UX analytics, zone-based click maps, frustration scoring
• Hotjar: Mid-market, casual entry point with heatmaps, session replay, and surveys
• Heap: Retroactive autocapture events

Through 2024 and 2025 the three products are being integrated, but as of May 2026 they still operate under separate UIs. The Contentsquare core differentiates with composite metrics like "Frustration Score" and "Engagement Score."

Hotjar remains the default choice for SaaS marketing sites.

• Heatmaps: Click, move, scroll
• Recordings: Session replay
• Surveys + Feedback: In-app NPS and surveys
• Funnels (2025): Simple funnels

Hotjar pricing:

• Basic (Free): 35 sessions captured per day
• Plus: From $32/month (billed annually)
• Business: From $80/month
• Scale: Quoted

<!-- Hotjar tracking code (one line for the whole site) -->
<script>
  (function(h,o,t,j,a,r){
    h.hj=h.hj||function(){(h.hj.q=h.hj.q||[]).push(arguments)};
    h._hjSettings={hjid:YOUR_SITE_ID,hjsv:6};
    a=o.getElementsByTagName('head')[0];
    r=o.createElement('script');r.async=1;
    r.src=t+h._hjSettings.hjid+j+h._hjSettings.hjsv;
    a.appendChild(r);
  })(window,document,'https://static.hotjar.com/c/hotjar-','.js?sv=');
</script>

8. Fullstory / Optimizely / VWO / Statsig / GrowthBook — the other heavy hitters

Fullstory (2014, US) is the session-replay incumbent. Series E in 2024 ($25M), GenAI-based "Frustration Signals" shipped in 2025. Pricing is undisclosed and skews enterprise.

• Strengths: Smoothest replay UX, can be combined with backend API traces.
• Weaknesses: Opaque pricing, analytics features are secondary.

Optimizely (2010, US) is the A/B-testing incumbent. Merged with Episerver in 2020 and re-positioned as a DXP (Digital Experience Platform). In 2026 it is still the enterprise A/B benchmark.

• Products: Web Experimentation, Feature Experimentation (for developers), Content Marketing, CMS
• Pricing: Quoted, typically over $50k per year
• Strengths: Sophisticated stats engine (Stats Engine, Sequential Testing)
• Weaknesses: Expensive, steep learning curve

VWO (Visual Website Optimizer, 2010, India) is the value alternative to Optimizely.

• Products: Testing, Personalization, Insights (heatmaps), Engage (in-app messages)
• Pricing: Starting Tier from $279/month, Growth from$649/month
• Strengths: Visual editor that lets marketers build tests directly
• Weaknesses: Stats engine simpler than Optimizely or Statsig

Statsig (2021, US, founded by ex-Meta) was the 2024-2025 dark horse of experimentation. Offers a combination of LaunchDarkly and Optimizely at a generous free tier.

• Products: Feature Flags, Experiments, Product Analytics, Session Replay (2024), Warehouse Native (2025)
• Pricing: Very generous free tier (1M events per month, unlimited users, unlimited flags). Enterprise is quoted.
• Strengths: Stats engine at Meta-internal-tool levels of rigor (Continuous Exposure, CUPED)
• Weaknesses: Developer-leaning UX, hard for marketers

# Statsig Python SDK
from statsig import statsig, StatsigUser

statsig.initialize('secret-server-key')

user = StatsigUser(user_id='user_42', email='alice@example.com')

# Feature flag
if statsig.check_gate(user, 'new_checkout_flow'):
    # New flow
    pass

# Experiment (get a variant)
exp = statsig.get_experiment(user, 'button_color_test')
color = exp.get('color', 'blue')  # default blue

GrowthBook (2022, US) is the open-source experimentation platform. Where PostHog is all-in-one, GrowthBook is "experimentation-only, open-source."

• License: MIT
• Products: Feature Flags, A/B Tests, Bayesian plus frequentist stats engines
• Data: Analyzes directly inside your data warehouse (Snowflake, BigQuery, ClickHouse) — no data movement
• Pricing: Self-hosted free, Cloud Free (unlimited flags, 5 experiments), Cloud Pro $20/seat/month
• Strengths: Warehouse-native experimentation, statistical transparency (every calculation is documented)

9. Plausible / Fathom / Umami / Matomo / Simple Analytics — privacy-first

Across 2024 and 2025, as EU GDPR plus ePrivacy enforcement and cookie-banner fatigue peaked, the market exploded for "no cookies, no personal identifiers, just traffic data is enough." The five leaders in this category.

Plausible Analytics (2018, Estonia)

• Open source (AGPL v3) plus a managed SaaS
• 100% cookieless, GDPR/PECR/CCPA compliant, EU data residency
• 1KB script (about 1/45 the size of GA4)
• Pricing: $9/month for 10k pageviews,$19 for 100k, $49 for 1M (annual billing saves 33%)
• Self-hosting is free, 30 minutes with Docker Compose

<!-- Plausible tracking (one line) -->
<script defer data-domain="example.com" src="https://plausible.io/js/script.js"></script>

Fathom Analytics (2018, Canada)

• Closed source, managed SaaS only
• ISO 27001 certified, EU or CA data residency by choice
• Pricing: From $15/month for 100k pageviews
• Strengths: UI is slightly more polished than Plausible, excellent email reports

Umami (2020, US, full-stack developer Mike Cao)

• Open source (MIT), 100% self-host friendly
• One-command deploy on Vercel plus PostgreSQL
• Pricing: Self-hosted free, Umami Cloud from $9/month (10k events)
• Strengths: Lightest setup, Next.js friendly

# Umami self-host (Docker Compose)
git clone https://github.com/umami-software/umami.git
cd umami
docker-compose up -d
# Visit http://localhost:3000

Simple Analytics (2018, Netherlands)

• Closed source, managed SaaS
• 100% cookieless, EU data residency
• Pricing: From $19/month for 100k events
• Strengths: Tracks carbon footprint, B-Corp certified

Matomo (2007, New Zealand, formerly Piwik)

• Open source (GPL v3), the oldest self-hosted analytics
• Managed SaaS option available (EU, UK, or US data residency)
• Pricing: Self-hosted free, Cloud from EUR 29/month (50k pageviews)
• Strengths: De facto standard at European governments and public agencies, nearly 100% feature parity with GA
• Weaknesses: Self-hosting is heavy (MySQL recommended, Redis recommended)

Pirsch (2021, Germany)

• Closed source, managed SaaS
• 100% cookieless, German data residency, no cookie consent needed
• Pricing: From EUR 5/month for 10k pageviews
• Strengths: Cheapest in class, vetted by German GDPR lawyers

Sherlock (2022, US)

• Closed source, specializes in SaaS activation scoring
• Surfaces "how sticky is this user on our product" as a 0-100 score
• Pricing: Quoted

10. CDP — Segment (Twilio) / RudderStack / Snowplow

Segment (2011, US, acquired by Twilio for $3.2B in 2020) is the definer of the CDP category. Twilio reshuffled around it in 2024 and a spinoff rumor surfaced for a while, but as of 2025 it stays under Twilio.

The core value prop: define the event once, send it simultaneously to 250+ downstream tools (Mixpanel + Amplitude + HubSpot + Salesforce, and so on).

Core products:

• Connections: Sources (web, mobile, server) to destinations (analytics, marketing, ads)
• Personas (Engage): Unified user profiles, cohort sync
• Protocols: Event schema validation and governance
• Functions: Serverless data transformations (Lambda-style)
• Reverse ETL (2022): Data warehouse to SaaS tools

Pricing (May 2026):

• Free: 1,000 MTU per month, 2 sources
• Team: From $120/month, unlimited sources
• Business: Quoted, typically over $50k per year

// Segment Analytics.js 2 (2024)
import { AnalyticsBrowser } from '@segment/analytics-next'

const analytics = AnalyticsBrowser.load({ writeKey: 'YOUR_WRITE_KEY' })

analytics.identify('user_42', {
  email: 'alice@example.com',
  plan: 'pro',
})

analytics.track('Checkout Completed', {
  amount: 99.0,
  currency: 'USD',
})

// This one call fans out to Mixpanel, Amplitude, GA4, HubSpot, Slack, and 100 other tools.

RudderStack (2019, US) is an open-source Segment alternative. Post-Series C (2025), it took a "warehouse-first CDP" stance.

• License: AGPL v3 (open source) plus Cloud (SaaS)
• API is nearly Segment-compatible (migration is a half-day exercise)
• Data lands in your warehouse first (Snowflake, BigQuery), then routes to SaaS — a governance edge over Segment
• Pricing: Self-hosted free, Cloud Free (1M events/month), Pro quoted

Snowplow (2012, UK) is the original open-source CDP. AGPL license, Series B ($40M) in 2022. In 2026 it rebranded as a "behavioral data platform."

• Very rich data model (event schema validation, JSON Schema based)
• Loads data straight into your warehouse (Redshift, BigQuery, Snowflake, Databricks); routing to external tools is incidental
• Pricing: BDP Cloud quoted (roughly over $50k/year), open source is free but operationally heavy
• Strengths: Data-engineer friendly, the most rigorous schema governance
• Weaknesses: Heavy to operate (needs Kinesis plus Lambda plus Iceberg-class infrastructure)

Hightouch / Census (reverse ETL space)

• Tools that treat the data warehouse as the single source of truth and sync to external SaaS
• The core part of a "warehouse-native CDP"
• Hightouch: Series C, quoted pricing
• Census: Series B, similar pricing band

11. AI in analytics — Mixpanel Spark / Amplitude Audience GenAI / PostHog AI

Through 2024 and 2025, every analytics product shipped a GenAI feature in lockstep. The reality, as of May 2026:

Mixpanel Spark (2025):

• Natural-language chart generation — "share of users who signed up last week and reached checkout"
• Automatic event and cohort recommendations
• Automatic insight discovery ("D7 retention dropped by 12% this week. Cause: mobile app v4.2 launch.")
• Subjectively more accurate than competitors (thanks to Mixpanel's refined event model)

Amplitude Audience GenAI (2025):

• Natural-language cohort definition — "users who signed up last month but didn't return this week"
• Recommended experiments (which hypothesis to test for users who saw a given feature)
• Data Assistant (Mixpanel Spark equivalent)
• Integrated with Govern to warn on misdefined events

PostHog AI (2025):

• "MaxAI" chatbot — present across all PostHog UI
• Automatic SQL generation (PostHog is ClickHouse-based and exposes a SQL interface)
• Automatic session-replay summaries — "this user got frustrated on the checkout page and refreshed three times"
• Integrated with LLM Observability (it can track its own AI calls)

Limitations (common across all products):

• "Natural-language query" is about 70% accurate. The other 30% shows the wrong chart.
• Automatic insight discovery is noisy (mostly exogenous variables like weekend or viral effects).
• Costs creep up (LLM calls accumulate and show up on the bill).

// PostHog AI (MaxAI) programmatic call - 2025
import { PostHogAI } from 'posthog-ai'

const max = new PostHogAI({ apiKey: 'phc_YOUR_KEY' })

const result = await max.query({
  prompt: 'Compute the share of users who signed up last week and reached checkout',
  context: { project_id: 12345 },
})

console.log(result.sql)      // Generated SQL
console.log(result.data)     // Execution result
console.log(result.chart)    // Recommended chart type

12. Korea / Japan — Toss Data, Kakao Data, Mercari, AbemaTV

Korea

Toss (Viva Republica) is famous for its in-house DBR (Data-Backed Recommendation) platform. It barely uses external analytics tools and runs an internal stack (Iceberg + Trino + Superset + a homegrown experimentation platform). At the 2025 Toss Data Conference, it unveiled "Tossfeed," a real-time event processing platform on Kafka + Flink + ClickHouse.

Toss Payments also started shipping an analytics SDK (Toss Events) for external use in 2025. An attempt to standardize Korean fintech data.

Kakao varies by business unit:

• KakaoTalk: in-house analytics (originally Hadoop + Hive, migrated to Trino + Iceberg in 2024)
• KakaoPay: Mixpanel plus in-house tools
• Kakao Games: game analytics tools (Adjust, Singular) plus in-house

Coupang is AWS-native, with Redshift + S3 + an in-house BI (KuPro). Does not use Heap or Mixpanel.

Naver runs in-house analytics on its own NCloud. Hardly uses external SaaS.

Korean startups (Series A to C):

• Most common combo: Amplitude (strong on mobile + generous free tier)
• Or: Mixpanel (strong on web)
• 2024 to 2025: accelerated migration to PostHog (open source + cost)
• Marketing site: Google Analytics 4 + Hotjar (default)
• A/B testing: Optimizely transitioning to Statsig or GrowthBook

Japan

Mercari runs Looker + BigQuery + an in-house experimentation platform. Looker is the analytics standard, while experiments run on a homegrown tool called Eclipse. Added LLM observability (LangSmith) in 2024.

CyberAgent (AbemaTV) runs its own data infrastructure (BigQuery + Looker + an in-house experimentation tool). Because it powers the ads business, it barely uses external SaaS.

Rakuten uses Adobe Analytics + Tealium iQ (CDP). A classic enterprise stack.

LINE (a Naver subsidiary) runs in-house analytics (originally Hadoop + Trino + Iceberg in 2024). Does not use external SaaS.

Japanese startups:

• SmartHR, freee: Mixpanel + Hotjar (most common)
• Japanese D2C: KARTE (a Japan-made CDP + chat + analytics all-in-one, founded by Plaid in 2014)
• B2B SaaS: Pendo (the in-app guide standard)

Japan-specific tools

• KARTE (Plaid Inc, 2014, listed on the Tokyo Stock Exchange): Japan's answer to Segment + Heap + Hotjar combined. Dominant in Japan's EC market.
• Beusable (Korea, 2014): A strong heatmap and session-replay vendor in both Korea and Japan.
• Ptengine (2010, Japan and China): Heatmap plus A/B testing, popular with Japanese SMBs.

13. Who should pick what — solo / startup / growth / enterprise / privacy

Solo developer / side project

• Analytics: Plausible or Umami (self-hosted free) or PostHog Free
• Session replay: Microsoft Clarity (entirely free, operated by Microsoft)
• A/B testing: GrowthBook Cloud Free
• CDP: Not needed. Send events directly.
• Budget: $0/month

Seed to Series A startup

• Analytics: PostHog Cloud (start on Free, 1M events covered)
• Session replay: PostHog built-in
• A/B testing: PostHog built-in or Statsig Free
• CDP: Not needed (PostHog covers some destinations)
• Budget: $0 to$200/month

Series B to C startup (growth stage)

• Analytics: Amplitude Plus or Mixpanel Growth (event taxonomy governance becomes essential)
• Session replay: Fullstory or PostHog (run in parallel)
• A/B testing: Statsig or GrowthBook (stats engine rigor starts to matter)
• CDP: Segment Team or RudderStack (if five or more marketing tools are connected)
• In-app guides: Userpilot or Chameleon
• Budget: $1,000 to$10,000/month

Enterprise (Series D+, revenue over $50M)

• Analytics: Amplitude Enterprise plus an in-house data warehouse (Snowflake)
• Session replay: Fullstory Enterprise
• A/B testing: Optimizely or Statsig Enterprise (some shops run both)
• CDP: Segment Business or Snowplow (BDP Cloud)
• In-app guides: Pendo or WalkMe
• UX analytics: Contentsquare
• BI: ThoughtSpot or Looker
• Governance: Amplitude Govern plus Segment Protocols, side by side
• Budget: over $250,000/year

Privacy-first (medical, government, EU public sector)

• Analytics: Matomo self-hosted (EU government standard)
• Session replay: Forbidden (risk of capturing PII)
• A/B testing: GrowthBook self-hosted
• CDP: Snowplow self-hosted (no external data transfer)
• Marketing site: Plausible or Fathom (EU data residency)
• Budget: Infrastructure only

Tools that died or were acquired in 2024 (for the record)

• June (Series A then shut down in 2024): launched grandly as "analytics for B2B SaaS" but got squeezed between PostHog and Mixpanel and wound down. Provided migration guides to the end.
• Hotjar → Contentsquare acquisition (2023)
• Heap → Contentsquare acquisition (2024)
• Iteratively → Amplitude acquisition (2022, CDP integration)
• Snowplow Analytics Open Source v2 → license changed to AGPL (2021, separating commercial cloud from open source)

14. References

Managed analytics

• Mixpanel Documentation — https://docs.mixpanel.com/
• Mixpanel Spark Announcement — https://mixpanel.com/blog/spark/
• Amplitude Documentation — https://amplitude.com/docs
• Amplitude Audience GenAI — https://amplitude.com/blog/audience-genai
• Heap Documentation — https://developers.heap.io/
• Pendo Documentation — https://support.pendo.io/

Open-source analytics

• PostHog Documentation — https://posthog.com/docs
• PostHog GitHub — https://github.com/PostHog/posthog
• PostHog LLM Observability — https://posthog.com/docs/llm-observability
• Matomo Documentation — https://matomo.org/docs/
• Matomo GitHub — https://github.com/matomo-org/matomo

UX analytics / session replay

• Contentsquare — https://contentsquare.com/
• Hotjar Documentation — https://help.hotjar.com/hc/en-us
• Fullstory Developer — https://developer.fullstory.com/
• Microsoft Clarity — https://clarity.microsoft.com/
• LogRocket — https://docs.logrocket.com/

Experimentation / feature flags

• Optimizely Documentation — https://docs.developers.optimizely.com/
• VWO — https://help.vwo.com/
• Statsig Documentation — https://docs.statsig.com/
• GrowthBook Documentation — https://docs.growthbook.io/
• GrowthBook GitHub — https://github.com/growthbook/growthbook
• LaunchDarkly — https://docs.launchdarkly.com/

Privacy-first analytics

• Plausible Analytics — https://plausible.io/docs
• Plausible GitHub — https://github.com/plausible/analytics
• Fathom Analytics — https://usefathom.com/docs
• Umami GitHub — https://github.com/umami-software/umami
• Simple Analytics — https://docs.simpleanalytics.com/
• Pirsch Analytics — https://docs.pirsch.io/

CDP

• Segment Documentation — https://segment.com/docs/
• RudderStack Documentation — https://www.rudderstack.com/docs/
• RudderStack GitHub — https://github.com/rudderlabs/rudder-server
• Snowplow Documentation — https://docs.snowplow.io/
• Snowplow GitHub — https://github.com/snowplow/snowplow
• Hightouch — https://hightouch.com/docs
• Census — https://docs.getcensus.com/

Korea / Japan

• Toss Tech Blog — https://toss.tech/
• Kakao Tech Blog — https://tech.kakao.com/
• Coupang Engineering — https://medium.com/coupang-engineering
• Mercari Engineering Blog — https://engineering.mercari.com/
• CyberAgent Developers Blog — https://developers.cyberagent.co.jp/blog/
• KARTE (Plaid) — https://karte.io/

Standards / reference reading

• Customer Data Platform Institute — https://www.cdpinstitute.org/
• W3C Web Analytics Working Group — https://www.w3.org/community/web-analytics/
• GDPR Official Text — https://gdpr-info.eu/
• ePrivacy Regulation — https://www.europarl.europa.eu/RegData/etudes/BRIE/2018/623544/EPRS_BRI(2018)623544_EN.pdf
• "Trustworthy Online Controlled Experiments" — Ron Kohavi et al., Cambridge 2020 (the canonical text on experimentation statistics)