Mobile & Product Analytics 2026 Deep Dive - Firebase, Amplitude, Mixpanel, PostHog, Branch, AppsFlyer, RevenueCat, Airbridge

"What gets measured gets managed. But what gets mismeasured gets mismanaged." — Peter Drucker (reinterpreted)

In May 2026, mobile and product analytics is no longer "drop in one Google Analytics snippet and call it a day." Product Analytics, MMP (Mobile Measurement Partner), CDP (Customer Data Platform), Session Replay, Feature Flags, and Subscription Analytics have each settled into independent categories, and most growth-stage companies run four to six of these in parallel.

This guide compares more than thirty tools — Firebase Analytics, GA4, Amplitude, Mixpanel, PostHog, Heap, Adobe Analytics, AppsFlyer, Adjust, Branch, Singular, Kochava, Airbridge, Adbrix, RevenueCat, Adapty, Segment, RudderStack, Hightouch, LaunchDarkly, Statsig, and more — across category, pricing, and real-world fit. We also cover the post-ATT privacy landscape, server-side tagging, and how Korean and Japanese companies actually wire all of this together in production.

1. The 2026 Analytics Map - Product / MMP / CDP / Replay / Flags / Subscription

Start by getting the categories right. The single most common confusion among new PMs and engineers is treating "Mixpanel vs AppsFlyer" as a vendor showdown — it is not. They answer different questions.

Product analytics answers "what does the user do inside our app?" while MMP answers "how did the user find our app?" These are fundamentally different problems, and neither tool replaces the other. CDP is the plumbing that delivers consistent data to all of the above, and session replay adds qualitative texture to the quantitative numbers.

The headline rule: pick the category first, then compare products inside it. If you mix categories you will never decide.

2. Firebase Analytics + GA4 - The Google Baseline

Firebase Analytics is the analytics module of Firebase, the platform Google acquired in 2016. It is free and event-based. With Google Analytics 4 (GA4) launching in 2020 and Universal Analytics officially retired on July 1, 2023, GA4 became the de facto web-plus-app standard, sharing its backend with Firebase Analytics.

Key characteristics:

• Free — 500 event types, 25 user properties, 500 conversion events per month. The only realistic option for early-stage startups that need to start tracking without a budget conversation.
• Auto-collected events — about 25 events including app_open, session_start, first_open, screen_view, and in_app_purchase are emitted by the SDK automatically.
• Free BigQuery export — link Firebase to BigQuery and stream raw events at no cost. This is the single most powerful escape hatch from GA4 UI limitations.
• Crashlytics integration — crashes, ANRs, and user behavior on one screen.
• Remote Config — free A/B testing and progressive rollout.

The limitations are equally clear. UI to ingest latency is 24-48 hours. Cohort and funnel builders are far simpler than Amplitude or Mixpanel. Cross-platform user stitching is weak. Without going directly into BigQuery, deep insights are hard.

The practical 2026 pattern is "Firebase Analytics + BigQuery + Looker Studio": collect raw data for free, query it directly in BigQuery, and build dashboards in Looker Studio (formerly Data Studio). This stack carries most seed-to-Series-A companies.

3. Amplitude - The Product Analytics Standard

Amplitude, founded in 2012, is the de facto standard for product analytics. 2024 revenue was around 300 million USD, with over 60,000 customers including major Korean companies like Toss, Karrot, Coupang, and Baemin.

Core features:

• Events plus user properties — every analysis lives on the three axes of event_name, user_properties, and event_properties.
• Funnels — visualize conversion flows like app_open to signup_start to signup_complete to first_purchase.
• Retention — break down Day 1, Day 7, Day 30 retention in one view.
• Behavioral cohorts — dynamic segments like "users who did add_to_cart in the last 7 days but never converted."
• Pathfinder — automatically surface what users did before and after a given event.
• Amplitude Experiment — A/B testing module built on the ExperimentX acquisition (2021).
• Audiences — export cohorts directly to Braze, Iterable, or Facebook Custom Audiences.

Pricing tiers: Free (10 million events per month), Plus (from 49 USD per month), Growth (custom, typically 50K-100K USD per year), Enterprise (200K USD per year and up). The free tier is generous, but once MAU exceeds about a million you get forced into Growth and pricing jumps fast.

Amplitude's strength is letting analysts go deep without SQL. Its weakness is that raw data export is gated behind paid plans, which is one of the reasons PostHog and Heap make raw data freedom a core marketing pitch.

4. Mixpanel - Amplitude's Rival

Mixpanel, founded in 2009, predates Amplitude by three years and was the first-generation product analytics tool. It led the market for years, then ceded the top spot to Amplitude between 2017 and 2020. 2024 revenue is estimated around 150 million USD.

Functionally, Mixpanel is nearly at parity with Amplitude. Funnels, Retention, Cohorts, Insights, Flows, and Impact modules all exist, and pricing is similar. The differentiators:

• Boards — bundle several charts into a static dashboard view. Lighter than Amplitude's Notebook.
• Formulas — write inline expressions like event_a / event_b * 100 directly on charts. Amplitude requires defining a Custom Event for the same calculation.
• Free tier — 100 million events per month — ten times Amplitude's, but with a one-year data retention cap.
• Group Analytics — analyze accounts or organizations rather than individual users. A natural fit for B2B SaaS.

The selection rule is crisp. For individual user analysis (B2C, gaming, media), Amplitude has more integrations and case studies. For account-level analysis (B2B SaaS), Mixpanel's Group Analytics is more natural. That said, Amplitude has been investing heavily in Account Analytics since 2024 and the gap has narrowed.

5. PostHog - The Open Source All-in-One

PostHog, founded in 2020, is an open-source product analytics platform and the fastest-growing analytics tool of 2026. Its Series D in 2024 raised 70 million USD at an 800 million USD valuation.

PostHog's differentiator is "all-in-one" — under one SDK and one dashboard you get:

• Product Analytics — funnels, retention, cohorts at Amplitude or Mixpanel quality
• Session Replay — recording quality on par with FullStory
• Feature Flags — progressive rollout like LaunchDarkly
• A/B Experiments — statistical experimentation like Statsig
• Surveys — in-app polls and NPS
• Heatmaps — click and scroll heatmaps
• LLM Observability — added in 2024 for tracking LLM call latency, cost, and quality
• Data Warehouse — a ClickHouse-backed warehouse built in

Pricing: Free (1 million events per month, unlimited if self-hosted). Cloud pricing is usage-based, roughly 0.00005 USD per event after the first million free. Self-hosting is completely free but operationally heavier.

PostHog is powerful because it is genuinely open source — raw data export is not gated, EU and sensitive data can be fully isolated by self-hosting, and the all-in-one bundle removes vendor sprawl. The trade-off is that UI polish trails Amplitude or Mixpanel slightly, though the gap has closed dramatically across 2025.

6. Heap - Auto-Capture and Retroactive Analytics

Heap, founded in 2013, was acquired by ContentSquare in 2024. Its core differentiator is "auto-capture."

Amplitude, Mixpanel, and PostHog all rely on manual instrumentation — engineers add analytics.track('Add to Cart', { product_id }) calls at every interaction. Forget to add the track call on a new button and the data is silently gone forever.

Heap's auto-capture records every click, page view, and form submit automatically. Analysts later define which CSS selector maps to "the checkout button," and a brand new metric appears retroactively across historical data. This is called Retroactive Analytics, and "you do not need to predefine your data" is its strongest pitch.

Two downsides:

• Event noise — capturing every click explodes data volume and mixes meaningful events with accidental mis-clicks.
• Selector fragility in SPAs and React — dynamic class names like div[class="abc-123"] change every build and break selectors. Adding explicit data attributes such as data-heap-id has effectively become the best practice.

Heap shines for marketing and growth teams that want to start fast without engineering involvement. It is stronger on web than in mobile apps, where auto-capture reliability in React Native and Flutter is still uneven.

7. Adobe Analytics + Customer Journey Analytics

Adobe Analytics descended from the 2009 Omniture acquisition and was once the market share leader for enterprise web analytics. Customer Journey Analytics (CJA), launched in 2020 on Adobe Experience Platform, is the next-generation product, designed to plug directly into cloud data warehouses like BigQuery and Snowflake.

Key characteristics:

• Calculated Metrics — Excel-grade formulas for complex KPI definitions.
• Cross-channel — unify Web, App, Offline POS, and Call Center data under one user ID.
• Anomaly Detection — automatic time-series outlier detection.
• Cohort Tables — cohort analysis.
• Workspace — analyst-grade free-form canvas.

Pricing is not public, but typically 100K-500K USD per year. Implementations come with 4-6 months of consulting.

Adobe Analytics retains overwhelming preference among finance, telecom, airlines, and large retail. In Korea, LG U+, Shinhan Card, and Lotte Shopping are notable users. Where Amplitude and Mixpanel are for "PM and growth teams," Adobe Analytics is for "analyst and BI teams."

8. Pendo / WalkMe / Userlane / Appcues - Onboarding Overlays

This category is often labeled "Digital Adoption Platform (DAP)." The core value is launching in-app guides, tooltips, checklists, and NPS surveys with no code, with a side benefit of analytics on the usage of those guides themselves.

• Pendo (founded 2013, around 200 million USD revenue) — the B2B SaaS standard. Combines guides, analytics, and feedback. Used in Korea by Toss, Karrot, and others.
• WalkMe (founded 2011, acquired by SAP in 2024) — the enterprise standard. Particularly strong as an employee guidance overlay inside large SaaS like SAP, Salesforce, and Workday.
• Userlane — Europe-based, GDPR-friendly.
• Appcues — startup-friendly, lowest pricing.
• Chameleon — design-friendly.
• Userflow — emphasizes no-code.

Pricing is typically MAU-based at 10K-100K USD per year. These tools complement product analytics rather than replace Amplitude or Mixpanel. For the narrow problem of "drop-off during onboarding," however, they often deliver more intuitive insights than a generic funnel.

9. MMP - AppsFlyer / Adjust / Branch / Singular / Kochava

MMP (Mobile Measurement Partner) is the tool that tracks where a mobile install came from — Facebook, Google, TikTok, Apple Search Ads, influencer campaigns, and so on. It is a fundamentally different category from product analytics.

AppsFlyer (founded 2011, Israel) — global market share leader, around 250 million USD revenue in 2024. Integrated with 12,000-plus ad networks. Fastest to support SKAdNetwork 4.0 and 5.0. Number one in Korea as well.

Adjust (founded 2012, Berlin) — number two globally. Acquired by AppLovin in 2021 for one billion USD, still run as an independent brand. Strong in Japan and Europe, heavy gaming-advertiser concentration.

Branch (founded 2014, US) — leader in deep linking. Attribution is a secondary feature, but the deferred deep link experience ("if the app is not installed when a user taps an ad, route them to the store, then drop them on the original screen after first launch") is the de facto standard.

Singular (founded 2014, US) — marketing analytics plus attribution combined. Strong ROAS (Return on Ad Spend) analysis with ad spend integration.

Kochava (founded 2011, US) — privacy-first positioning, with an Indirect Measurement method that does not depend on IDFA or GAID.

Tenjin — exclusively for mobile game advertisers.

The single most important MMP selection criterion is "does it integrate with all the ad channels I actually buy from?" If you run on Korean or Japanese channels like Kakao Moment, Naver GFA, or Line Ads, the global MMPs alone are insufficient and a local MMP like Airbridge or Adbrix becomes mandatory.

10. Airbridge / Adbrix - The Korean MMPs

Airbridge is built by Ab180, founded in Korea in 2014, and shares the Korean market with AppsFlyer. Its Series C in 2022 raised 58 billion KRW, and it has been expanding to Japan, Southeast Asia, and the US.

Differentiators:

• Integration with Asian channels — Kakao Moment, Naver GFA, Line, Line Ads — channels that global MMPs typically support 1-3 months later are first-class in Airbridge.
• MMM (Marketing Mix Modeling) integration — Airbridge MMM, launched in 2024, estimates channel contribution statistically without relying on IDFA.
• Native Korean-language support — the standard advantage of Korean SaaS.

Adbrix, built by IGAWorks in 2012, predates Airbridge. It is more common on the publisher (ad network) side than on the advertiser side. After IGAWorks' 2023 acquisitions, its data integration capabilities have strengthened.

Adison is the attribution-plus-rewarded-ads integration solution within the IGAWorks group.

The standard configuration for Korean mobile advertisers is dual MMP — "AppsFlyer for global channels plus Airbridge for domestic channels" — or, to save cost, Airbridge solo. The dual-MMP gotcha is that install counts can double-count, so teams must adopt an explicit Single Source of Truth (SSOT) policy and designate one MMP as the master of record.

11. The Japanese MMP Market - Adjust Tokyo, F.O.X, Singular

The Japanese mobile market is similar to Korea's but distinct. Adjust Tokyo is effectively the number one MMP, and gaming advertisers like Cygames, Mixi, KLab, and gumi overwhelmingly prefer it.

• Adjust Tokyo — opened the Tokyo office in 2014. Holds many top-tier Japanese gaming clients. Specialized for Japanese-language support and gaming ad workflows.
• F.O.X (Fox Insight) — a subsidiary of Cyberowl (now Septeni Holdings). A Japan-native MMP with deep integration into Japanese channels like Line Ads, Yahoo! Ads, and AmebaAd.
• Singular Japan — growing among e-commerce and D2C advertisers (non-gaming).
• AppsFlyer Japan — used by global-friendly advertisers like Mercari and Rakuten.

Unlike Korea, Japan's mobile ad spend is dominated by gaming, and ROAS/LTV measurement accuracy can make or break the business. Gaming advertisers treat Adjust's SKAN tooling and Singular's ROAS Dashboard as nearly mandatory infrastructure.

Japanese advertisers also rely more heavily on server-to-server (s2s) measurement and probabilistic attribution than Korean advertisers, in part because the 2023 amendment to Japan's APPI (Act on the Protection of Personal Information) further tightened IDFA and GAID usage.

12. SKAdNetwork - Apple's Privacy Attribution

App Tracking Transparency (ATT), introduced in iOS 14.5 in 2021, rewired the paradigm of mobile ad measurement. Unless a user explicitly taps "Allow Tracking," ad SDKs cannot read the IDFA (Identifier for Advertisers). From 2024 through 2026, the industry-average IDFA opt-in rate has hovered around 20-30 percent.

Apple's replacement for IDFA is SKAdNetwork (SKAN). The core mechanics:

• The OS itself handles all information between ad click and install.
• Advertisers receive only the install event plus a conversion value (0-63).
• Postbacks arrive with a 24-48 hour delay.
• Precise per-campaign, per-creative, per-publisher attribution is effectively impossible.

SKAdNetwork 4.0 (2023) split conversion value into a two-tier model (coarse: low/medium/high, plus fine: 0-63), and 5.0 (2025) added re-engagement attribution and multi-postback.

MMPs like AppsFlyer, Adjust, Branch, and Airbridge translate SKAN data into a familiar attribution surface. The most important decision an advertiser makes is the conversion value mapping schema: "which user behaviors (first purchase, reaching level 5, ad click, and so on) map to which conversion value bits?" That mapping defines LTV and ROAS measurement fidelity.

13. Google Play Privacy Sandbox - Android's Answer

On Android, Google has been rolling out Privacy Sandbox for Android in phases since 2022. The core components:

• Topics API — interests are stored on the device and only category labels (for example "travel" or "sports") are exposed to advertisers, replacing ad identifiers.
• Protected Audience API (FLEDGE) — retargeting executes on-device.
• Attribution Reporting API — click-to-install-to-conversion attribution without IDFA or GAID.

This rollout is staged beta across 2024-2026 with full launch projected for 2027. Unlike Apple's SKAN, Google promises richer, more advertiser-friendly data, and the GAID (Google Advertising ID) is not disappearing immediately.

From the advertiser's perspective, the key point is that both SKAN and Privacy Sandbox are handled by the MMP. You almost never integrate directly with the OS APIs — you just keep AppsFlyer, Adjust, or Airbridge SDKs up to date.

14. CDP - Segment / RudderStack / Hightouch / mParticle

CDP (Customer Data Platform) promises "instrument once, send anywhere." Instrument app and web events one time, then fan them out simultaneously to dozens of destinations including Amplitude, Mixpanel, AppsFlyer, Braze, Iterable, and Snowflake.

• Segment (founded 2011, acquired by Twilio in 2020 for 3.2 billion USD) — market share leader, 400-plus destinations.
• RudderStack (founded 2019) — open-source alternative to Segment with near-identical workflow and self-hosting available.
• mParticle (founded 2013) — enterprise-only, strong audience integrations.
• Tealium (founded 2008) — the long-standing incumbent, with deep security certifications (SOC 2, HIPAA, FedRAMP).
• Treasure Data (founded 2011, an Arm subsidiary in Japan) — strong in Japan and the broader Asia market.
• Hightouch (founded 2021) — pioneer of the Reverse ETL category. Flips the direction: Snowflake or BigQuery to Salesforce, Braze, and so on.
• Census — Hightouch's main competitor.
• Snowplow (founded 2012, open source) — self-hosted event collection pipe, data-team-friendly.

CDP pricing is MTU-based, typically 10K-300K USD per year. Segment is free up to 10K MTU; pricing climbs sharply beyond that.

RudderStack vs Segment comes down to data sovereignty. Segment routes through Twilio data centers, which EU and financial-services companies often cannot accept for governance reasons, pushing them toward RudderStack's self-hosted option.

Hightouch's Reverse ETL is a new pattern. Traditional CDP flow is "app to CDP to destination," but Hightouch is "Snowflake to destination." Once a company has already centralized data in Snowflake or BigQuery, it becomes cleaner to treat the warehouse as the single source of truth than to route everything through a separate CDP. This pattern has been eating share fast across 2024-2026.

15. RevenueCat - The Subscription Analytics Standard

iOS and Android in-app purchase (IAP) is only possible through OS APIs — StoreKit (iOS) and Google Play Billing Library (Android) — and the two diverge completely on receipt format, subscription state model, refund policy, and Family Sharing. Building a subscription stack in-house means validating receipts on the server for both platforms, ingesting renewal, cancel, and refund webhooks, and tracking expirations and resubscribes — typically a 3-6 engineer 6-12 month project.

RevenueCat (founded 2017) collapses that into a one-line SDK call. 2024 revenue was around 50 million USD with over 30,000 apps in production, making it the market leader.

Features:

• SDK — iOS, Android, React Native, Flutter, Unity, and Web Billing. Purchases.shared.purchase(product) is the full path to a completed payment.
• Subscription Status — receipt validation, renewal, expiration, and refund tracking all server-side and automatic.
• Webhooks — renewal, cancel, and refund events pushed to your backend.
• Charts — MRR, churn rate, LTV, ARPU subscription dashboards out of the box.
• Experiments — price and paywall A/B tests.
• Targeting — different prices and paywalls per user segment.
• Paywalls — no-code paywall builder (launched 2024).

Pricing is MTR (Monthly Tracked Revenue) based: free up to the first 2,500 USD of revenue, then 1 percent of revenue beyond that (separate from Apple/Google's fees). At a million dollars of monthly revenue, RevenueCat costs around 800 USD per month, dramatically cheaper than building and operating the equivalent in-house.

16. Adapty / Qonversion / Glassfy - RevenueCat's Challengers

Adapty (founded 2020) is RevenueCat's most formidable challenger. Differentiators are paywall A/B testing and deeper CRM integration. Pricing is comparable.

Qonversion (founded 2020) leans open-source-friendly and was early to address usage-based pricing. Its free tier is more generous than RevenueCat's (free up to the first 10K USD of revenue).

Glassfy (founded 2021) is the smallest of the four but is EU/GDPR-friendly, with infrastructure hosted in the EU.

Selection rules are crisp.

• Fastest start, most case studies, most stable SDK -> RevenueCat
• Paywall experiments, CRM integrations -> Adapty
• Cost minimization, open-source-friendly -> Qonversion
• EU data residency, GDPR -> Glassfy

In 2026 nearly every new subscription app adopts one of these four. Building it in-house has practically disappeared.

17. Session Replay - FullStory / LogRocket / Smartlook / PostHog / Sentry

Session Replay plays back a real user session like a video. Where event funnels cannot tell you "why did this user abandon the checkout page?", session replay shows you visually.

• FullStory (founded 2014) — enterprise leader. Powerful page indexing and search. Expensive but precise.
• LogRocket (founded 2016) — developer-friendly. Captures Redux store and console.log alongside the session for debugging power.
• Smartlook (founded 2016, Czech Republic) — best price-to-feature ratio. Auto event capture is strong.
• Hotjar (founded 2014, acquired by ContentSquare in 2023) — web heatmaps plus replay.
• Mouseflow — Hotjar alternative.
• ContentSquare — enterprise UX analytics, currently integrating Hotjar and Heap after acquiring both.
• PostHog Session Replay — bundled inside PostHog. Best price-to-value.
• Sentry Session Replay — bundled inside Sentry. Recording the 30 seconds before an error is its killer pattern.
• Datadog RUM Session Replay — bundled inside Datadog's RUM.
• Microsoft Clarity — fully free. Less polished but a great starting point.

Pricing is typically MTU-based at 100-5,000 USD per month. The biggest differentiators are retention period (3 months vs 1 year) and masking policy (automatic PII redaction quality).

Mobile session replay is harder than web. Reconstructing iOS and Android UIKit, SwiftUI, and Jetpack Compose screens requires native instrumentation, and text input masking is significantly more involved than on the web. PostHog, Sentry, and Smartlook are investing heavily in this space.

18. Feature Flags - LaunchDarkly / Statsig / Optimizely / GrowthBook

Feature Flags let you toggle features on and off without deploying code. They are the core infrastructure behind A/B testing, progressive rollouts, kill switches, and beta access.

• LaunchDarkly (founded 2014) — enterprise standard, around 200 million USD revenue in 2024. The most expensive but the most stable.
• Statsig (founded 2021) — Meta alumni founders. Unlimited free tier plus integrated A/B testing. Series C in 2024 valued the company around 700 million USD, with the fastest growth in the category.
• Optimizely Flags — Optimizely's flag product, anchored on the Rollouts acquisition.
• PostHog Feature Flags — bundled inside PostHog for free.
• Flagsmith — open source, self-hostable.
• GrowthBook — open source, with strong statistics.
• Unleash — open source, the cleanest self-hosted option.
• Eppo — experimentation platform specialized in statistical analysis.
• Vercel Flags — bundled inside Vercel for free.

Pricing varies dramatically. LaunchDarkly is MAU-based at 50K-500K USD per year. Statsig is free up to 100 million MAU. PostHog Flags is free up to 1 million MTU.

Statsig's unlimited-free strategy shook the market. Companies frustrated by LaunchDarkly's price have been migrating in droves, and 2026 share is shifting rapidly. LaunchDarkly's stability and advanced features (SDK breadth, governance) still command a premium for enterprises.

19. A/B Testing - The Statistical Pitfalls

A/B testing tools overlap heavily with feature flags but diverge on the sophistication of the statistical engine.

• Optimizely Web/X — first-generation A/B testing. Expensive but the most validated.
• Statsig — feature flags and A/B testing under one SDK. Variance-reduction techniques like CUPED (Controlled-experiment Using Pre-Experiment Data) are first-class.
• GrowthBook — open source. Statistics engine built by PhDs and respected for rigor.
• Eppo — Airbnb alumni founders. Statistically the most advanced in the category.
• PostHog Experiments — free, with basic statistics only.
• Amplitude Experiment — A/B under the same SDK as Amplitude analytics.
• VWO — India-based, strong value pricing.
• AB Tasty — France-based, strong in Europe.
• Vercel Toolbar A/B — bundled inside Vercel.

The classic pitfall in A/B testing is treating statistical significance (p-value below 0.05) as the whole picture. In practice you must also consider:

• Sample size — typically at least 5,000 per variant.
• Run duration — minimum one week, usually two weeks to capture weekend versus weekday patterns.
• Peeking problem — checking results daily and stopping early once "significance" appears is a well-known false positive trap.
• Novelty effect — users react to anything new initially and revert later.
• Multiple comparison — tracking 10 metrics simultaneously almost guarantees one of them looks significant by chance.

Eppo, Statsig, and GrowthBook provide statistical engines that correct for these issues automatically.

20. Server-Side Tagging - GTM Server, Snowplow

Traditional analytics has the client (browser or app) SDK send events directly to the analytics server. Server-side tagging is becoming the standard for the following reasons:

• Ad blockers block client SDK requests. uBlock Origin blocks about 90 percent of requests to GA, Mixpanel, and similar.
• iOS Safari ITP caps third-party cookies at one day.
• GDPR/CCPA forbid third-party data transfer without explicit consent.
• Data accuracy — fingerprint-based blockers are nearly impossible to evade when sending directly from the client.

Server-side tagging has clients send events to a first-party domain like analytics.mycompany.com, then your server forwards them to Google, Facebook, Amplitude, and so on. Ad blockers cannot block first-party traffic.

Tooling:

• Google Tag Manager Server Container — the standard. Runs in a Docker container on GCP App Engine, typically 100-500 USD per month.
• Segment / RudderStack — CDPs that also play the server-side role.
• Snowplow — top choice for self-hosting.
• Tealium iQ — enterprise.

In 2026 server-side tagging is not optional for any serious marketing or analytics operation. Letting ad blockers eat 30-50 percent of marketing data silently corrupts the data foundation of every decision.

21. Privacy Regulation - GDPR / CCPA / LGPD / PIPA

Regulations to review before deploying any analytics:

• GDPR (EU, 2018) — the strictest. Explicit opt-in, right to be forgotten, data portability.
• CCPA/CPRA (California, 2020/2023) — opt-out based, US standard.
• LGPD (Brazil, 2020) — closely modeled on GDPR.
• PIPEDA (Canada) — GDPR-aligned.
• PIPA (South Korea) (2011, amended 2023) — administered by KISA. Explicit consent required for third-party ad SDKs.
• APPI (Japan) (2003, amended 2023) — mixed opt-in/opt-out.
• PDPA (Singapore) — opt-in.

Core obligations are nearly identical:

• Explicit consent banner before any analytics SDK loads.
• Data deletion requests answered within typically 30 days.
• Data residency — EU user data on EU servers.
• Data Processing Agreements (DPA) signed separately with each analytics vendor.

Tooling is standardized on CMP (Consent Management Platform) vendors: OneTrust, Cookiebot, Iubenda, Termly, and Didomi (France). Pricing is 1K-10K USD per year per domain.

The 2026 reality is a tradeoff: privacy-friendly equals higher cost plus measurable data loss. But GDPR fines run as high as 4 percent of revenue or 20 million euros — whichever is greater — so consent-free analytics is no longer a real option.

22. Cost - Events × Users × Retention

Analytics tool pricing is almost always a product of the following factors:

• MTU (Monthly Tracked Users).
• Events per user.
• Data retention (1 year vs 5 years).
• Destination count, for CDPs.

A rough 2026 cost map at 1 million MAU:

In aggregate a 1 million MAU company typically spends 300K-1M USD per year on analytics. The most effective levers for reducing this:

1. Event diet — review every event monthly and drop the ones nobody queries.
2. Sampling — track one out of every ten users rather than all of them.
3. Self-hosting (PostHog, RudderStack, GrowthBook) — infra cost grows, but per-event volume becomes unlimited.
4. BigQuery/Snowflake integration — analyze raw data directly instead of paying a SaaS premium.

23. Korea - Toss, Karrot, Coupang's Analytics Stacks

Toss — Toss has used Amplitude as its main product analytics tool since 2020. According to a 2024 Toss tech blog post, Toss processes about 30 million MAU and 5 billion daily events. For cost efficiency, Toss ingests raw events first into its own BigQuery-based data warehouse and only forwards critical events (sampled) to Amplitude. MMP setup is dual-rail with Airbridge plus AppsFlyer, and payment analytics is largely homegrown.

Karrot — Karrot's standard stack is Amplitude + AppsFlyer + Firebase. According to a 2023 Karrot tech blog post, raw events land in Snowflake, BI runs on Looker, cohort analysis happens in Amplitude, and experiments use an in-house A/B platform (Karrot Experiments). As Karrot expands globally, it has been consolidating from Airbridge toward AppsFlyer.

Coupang — Coupang has the highest in-house analytics ratio. It is reported to combine Adobe Analytics, in-house tooling, and partial use of Mixpanel, supported by a very large data engineering organization. Global SaaS vendors play a smaller role than at most Korean peers.

Baemin — Amplitude and AppsFlyer are central. In 2024 the Woowa Brothers tech blog disclosed RudderStack adoption and a self-hosted PostHog experiment. All events route through RudderStack for data governance.

24. Japan - Mercari, SmartHR, LINE's Choices

Mercari — Japan's largest secondhand marketplace app. A 2024 Mercari Engineering Blog post discloses the stack as Amplitude (product) + Adjust (MMP) + BigQuery (warehouse) + Looker (BI). RevenueCat is not used; Mercari runs its own payments.

SmartHR — Japan's HR SaaS leader. Given its B2B nature, Mixpanel is the main tool because of strong Group Analytics, with Google Analytics 4 as a supporting marketing-side option. A 2025 SmartHR Tech Blog post revealed an evaluation of self-hosted PostHog.

LINE — LINE relies overwhelmingly on its in-house data platform (Data Platform 2.0), with little use of Amplitude or Mixpanel. Internal users analyze on top of an in-house OLAP stack (Apache Druid + Trino). External SaaS use is largely confined to MMP (Adjust + F.O.X).

Cygames / KLab / Mixi (Gaming) — Adjust plus Singular plus in-house analytics. Game ROAS makes or breaks the business, so both MMP precision and in-house analytical depth are needed.

Japan's distinctive trait is that in-house build ratios are much higher than in Korea. LINE, Mercari, Rakuten, and ZOZO all have very strong BI teams and treat SaaS analytics as secondary supporting tooling. Japanese engineering salaries are relatively lower than in the US or even Korea, which improves the ROI of building rather than buying.

25. Practical Application - Five Scenario Playbooks

To close, here are recommended stacks by company stage.

Seed stage (under 10K MAU, 1-3 person team):

• Product analytics: Firebase Analytics + GA4 (free)
• Session replay: Microsoft Clarity (free) or PostHog Free
• A/B: Firebase Remote Config or PostHog
• MMP: not needed if you are not running paid ads yet. If you are, Branch Free tier.
• Total cost: 0 USD per year.

Series A (about 100K MAU, 10-30 person team):

• Product analytics: Amplitude Free or PostHog Cloud
• Session replay: PostHog or LogRocket
• A/B: Statsig (free)
• MMP: AppsFlyer or Airbridge
• Subscriptions: RevenueCat (free up to 2,500 USD in revenue)
• Total cost: 10K-30K USD per year.

Series B (about 1M MAU, 50-150 person team):

• Product analytics: Amplitude Growth plus BigQuery raw export
• Session replay: FullStory or LogRocket
• A/B: Statsig Pro or GrowthBook self-host
• MMP: AppsFlyer + Airbridge dual
• Subscriptions: RevenueCat
• CDP: Segment or RudderStack
• Total cost: 300K-800K USD per year.

Series C and beyond (about 10M MAU, 300+ person team):

• Product analytics: in-house OLAP (ClickHouse, Druid, Pinot) plus partial Amplitude
• Session replay: FullStory Enterprise
• A/B: in-house or Eppo
• MMP: AppsFlyer + Airbridge + in-house SKAN processing
• Subscriptions: RevenueCat or in-house
• CDP: Segment Enterprise or self-built plus Snowplow
• Total cost: 1M-5M USD per year.

Enterprise:

• Adobe Analytics + Adobe Experience Platform standard
• MMP: AppsFlyer Enterprise
• In-house BI team of 50-200
• Total cost: 5M-20M USD per year.

As the company stage advances, in-house build ratio grows and SaaS dependency shrinks. What does not change is that at every stage, MMP and subscription analytics remain overwhelmingly more efficient as external SaaS than as in-house builds.

26. References

• Firebase Analytics documentation — https://firebase.google.com/docs/analytics
• Google Analytics 4 documentation — https://support.google.com/analytics/answer/10089681
• Amplitude documentation — https://www.docs.developers.amplitude.com/
• Mixpanel documentation — https://docs.mixpanel.com/
• PostHog documentation — https://posthog.com/docs
• Heap documentation — https://developers.heap.io/
• Adobe Analytics documentation — https://experienceleague.adobe.com/docs/analytics.html
• AppsFlyer documentation — https://dev.appsflyer.com/
• Adjust documentation — https://help.adjust.com/
• Branch documentation — https://help.branch.io/
• Singular documentation — https://support.singular.net/
• Kochava documentation — https://support.kochava.com/
• Airbridge documentation — https://help.airbridge.io/
• Adbrix documentation — https://help.adbrix.io/
• RevenueCat documentation — https://www.revenuecat.com/docs/
• Adapty documentation — https://docs.adapty.io/
• Qonversion documentation — https://documentation.qonversion.io/
• Segment documentation — https://segment.com/docs/
• RudderStack documentation — https://www.rudderstack.com/docs/
• Hightouch documentation — https://hightouch.com/docs
• LaunchDarkly documentation — https://docs.launchdarkly.com/
• Statsig documentation — https://docs.statsig.com/
• GrowthBook documentation — https://docs.growthbook.io/
• Eppo documentation — https://docs.geteppo.com/
• FullStory documentation — https://developer.fullstory.com/
• LogRocket documentation — https://docs.logrocket.com/
• Apple SKAdNetwork documentation — https://developer.apple.com/documentation/storekit/skadnetwork
• Google Privacy Sandbox for Android — https://privacysandbox.google.com/private-advertising
• Apple App Tracking Transparency — https://developer.apple.com/documentation/apptrackingtransparency
• Standard Webhooks specification — https://www.standardwebhooks.com/
• Mercari Engineering Blog — https://engineering.mercari.com/en/blog/
• SmartHR Tech Blog — https://tech.smarthr.jp/
• Toss tech blog — https://toss.tech/
• Karrot tech blog — https://medium.com/daangn
• Woowa Brothers tech blog — https://techblog.woowahan.com/