A PM's analysis published on Medium on April 23, 2026, makes a sharp observation: "The quiet winner in B2B product analytics is PostHog." Heap got absorbed by Contentsquare. June shut down in August. Statsig sold to OpenAI for $1.1B. Meanwhile, PostHog crossed 100,000 installs with zero SDRs and landed a $1.4B valuation. The thing behind all of that is Max AI. Type "show me everyone who signed up last week but hasn't activated yet" and it writes the SQL, builds the chart, and hands you a summary. According to PostHog's own data, it also cut their install wizard from 10 minutes to 90 seconds. In an era where AI feature marketing is everywhere, this is one of the cases worth actually pulling apart.
What Makes PostHog Max Different from Other 'AI Analytics'?
Mixpanel Spark AI, Amplitude AI Insights, Statsig AI (pre-acquisition) — every analytics tool has an "AI feature" these days. But one Medium analyst puts it well: "Most AI features are useful but thin. They're just one layer on top of the same analytics engine." So what actually sets Max AI apart?
Max doesn't just generate answers — it autonomously operates the PostHog UI based on your question. Insight creation, dashboard building, session replay summaries, HogQL generation — wherever you'd normally click through menus, Max clicks first.
PostHog's own September 2025 engineering post, "What we've learned about building AI-powered features," lays out the design principles behind this. The core idea: it has to live inside the main product flow. If you create a separate tab and say "come over here to ask the AI," users won't bother. Instead, Max is invoked directly from inside the Insight view, the Session Replay view, and the Dashboard view.
- Model mixing
Fast responses route to gpt-4.1-mini; deep analysis goes to gpt-4.1. This keeps costs from exploding and keeps response times within what users will actually tolerate. - UI context injection
MaxUIContextinjects the current screen, active filters, and dashboard state with every request. That's why "redraw this chart differently" actually works. - Hard guardrails
Explicit rules like "never generate API keys." In areas where hallucinations destroy business value — auth, billing, data handling — refusal is the default. - Monitoring is the real product
It's not install-and-done. Tracking failed calls, bad routing, and rejected responses to continuously refine those guardrails is the actual ongoing work.
In March 2026, PostHog Clusters rolled out to all LLM Analytics users. It auto-groups incoming LLM traces by behavior — and at launch, the largest cluster was "event tracking analysis and product metric investigation." People using AI to analyze AI behavior: a meta pattern that signals exactly how analysis itself is changing.
Why PostHog Over Mixpanel or Amplitude?
Here's how the tools compare on actual decision-making criteria — drawn from four years of hands-on deployments and replacements across three B2B SaaS companies.
| Criteria | Amplitude / Mixpanel | PostHog (with Max AI) |
|---|---|---|
| Adoption after 6 months | Without a dedicated analyst, only 2–3 people actually use it. Everyone else just asks on Slack. | At a 60-person company, the entire 8-person PM team was running their own queries within a week. Two engineers were self-serving too. |
| What's included | Analytics only. Replays, flags, and experiments are separate tools or add-ons. | Analytics + session replays + feature flags + A/B testing + surveys + error & LLM monitoring — all in one. |
| Cost at 50K MAU | Amplitude $30K–$60K/year, Mixpanel Growth $800–$2K/month | $200–$400/month. Billing caps keep it from spiraling. |
| AI interface depth | Natural language queries + auto-summaries (thin layer) | Max autonomously operates the UI + generates HogQL + summarizes session replays + Plan/Research Mode |
| Learning curve | Frequent complaints that the interface is overwhelming | Developer-first. Rougher for solo PM use than Mixpanel |
Here's the thing — the real conclusion isn't "pick PostHog because the AI is better." It's "the bundle is deep and the automation is baked in, so the whole team actually uses it." Max is the final friction-remover: it lets PMs pull their own insights without SQL. It doesn't replace analysts — it distributes the act of analysis.
From 2020 to mid-2025, PostHog grew to 13,700 active domains, with more than half of the latest YC batch adopting it. 100K installs without a single SDR — engineers install it, like it, and tell a friend at another company. Classic PLG.
① Question the "AI feature marketing" framing. As PostHog itself acknowledges, natural language queries are great for ad-hoc questions, but they don't replace knowing which questions to ask in the first place. ② The developer-first interface is rough for PM-only workflows. Non-technical PM-heavy organizations will find Mixpanel smoother. ③ Pricing is usage-based — LLM inference cost plus a 20% PostHog margin. 1,000 credits = $10. Heavy users should set billing caps upfront.
Getting Started: Adding Max to Your Team
Starting PostHog from scratch takes some time, but turning on Max AI is fast. If you're already on PostHog, you can have it running in under 30 minutes.
- Enable LLM Analytics in PostHog Cloud or self-hosted
Dashboard bottom-left menu → "AI" toggle. Free tier includes 2,000 credits per month. At ~$10 per 1,000 credits, there's no real barrier to getting started. - Call Max from any Insight view
Click "Ask Max" in the top right from any Dashboard, Insight, or Session Replay screen. Screen context gets injected automatically, so short commands like "redraw this graph in 7-day intervals" just work. - Try Plan Mode and Research Mode
Plan is for strategic questions like "give me three growth hypotheses for next quarter." Research is for multi-step digs like "go deep on what caused the activation drop over the last 30 days." Both run separately from the default mode. - Monitor guardrails and usage
Always review the SQL Max generates against your own data. Check your credit usage weekly and set a billing cap — one runaway session can multiply your monthly cost several times over.
One important note: Max AI really shows its value when PostHog's other tools are already running and data is flowing in. Event tracking, session replays, and feature flags need to be active before Max can handle integrated commands like "pull the replays for users who dropped off on this screen." If you're starting fresh, don't lead with Max — start with tracking infrastructure.
There's a bigger pattern here. In 2025, the category got reshuffled — Heap to Contentsquare, June shutting down, Statsig to OpenAI. PostHog is the only one that kept growing, PLG-style, while the independent category shrank around it. Max AI is the final accelerant on that growth curve — not magic. The point isn't automating the analyst's job. It's making PMs and engineers who never ran analysis start running analysis. That's what 100K installs is actually about.
Deep Dive Resources
PostHog AI Official Docs — Full manual for Max, Plan, and Research Primary source for the natural language interface, HogQL generation, and session replay summaries. Includes pricing ($0.01/credit, 2K free per month). posthog.com
What we've learned about building AI-powered features — PostHog Engineering Nine design principles behind Max AI. Includes MaxUIContext code samples, model mixing strategy, guardrail patterns, and the install wizard 10-min → 90-sec case. newsletter.posthog.com
The State of B2B Product Analytics in 2026 — Full category comparison Covers Heap, June, and Statsig acquisitions and shutdowns. Real-world 6-month adoption rates for PostHog vs. Amplitude vs. Mixpanel, plus actual pricing comparisons. medium.com
PostHog Clusters Launch — Auto-grouping LLM traces by user behavior The March 2026 feature that rolled out to all LLM Analytics users. The fact that the largest initial cluster was "event analysis itself" is a meta signal worth reading. createwith.com
.png)
