Spenser Skates, co-founder and CEO of Amplitude, recounts the company’s journey from a failed voice-to-text startup to a $5.6B IPO and its current wholesale reorganization around AI, offering a detailed case study in product pivots, category creation, and the discipline required to rebuild a mature company for a new technological paradigm.
The origin story: Sonalight’s failure seeded Amplitude’s core insight
Spenser and co-founder Curtis Liu built Sonalight, a voice-to-text texting app for Android, in 2011–2012, riding early speech recognition improvements but hitting a retention wall: users tried it once or twice but did not come back.
They realized they lacked visibility into why users churned — existing analytics tools (Mixpanel, Google Analytics) answered “how many” but not “which behaviors correlate with retention.”
They built an internal analytics engine to answer that question, discovered that users who sent 3+ messages in their first week retained, and used that insight to improve onboarding; the internal tool became more valuable than the app itself.
Investors (including Y Combinator) pushed them to pivot to the analytics tool; they shut down Sonalight in 2012 and started Amplitude, applying the lesson that behavioral analytics tied to outcomes was the unsolved problem.
Category creation: defining “product analytics” against entrenched incumbents
Amplitude entered a market dominated by Mixpanel (event-based) and Google Analytics (page-view-based), neither of which handled complex, multi-step user journeys or cohort retention analysis well.
Spenser deliberately coined and evangelized the term “product analytics” to carve a new category: analytics built for product teams, not marketers, centered on user behavior over time rather than page views or isolated events.
The strategic bet: product teams were becoming the primary buyers of analytics as software ate the world, and they needed a tool that spoke their language (funnels, retention, paths, cohorts) without SQL.
Amplitude’s early differentiation was the “behavioral graph” — a proprietary data model that joined events across sessions and devices into a unified user timeline, enabling retroactive analysis without pre-defining schemas.
Go-to-market evolution: from bottom-up adoption to enterprise sales motion
Initial traction came from a freemium model and a generous free tier (10M events/month) that let individual PMs and growth teams adopt without procurement; viral spread inside companies created expansion revenue.
Around 2017–2018, as deals grew past $100K ARR, they layered on a direct enterprise sales force, hiring reps who could navigate security reviews, legal, and multi-year contracts — a transition Spenser describes as painful but necessary.
They invested heavily in customer success and professional services to ensure implementation quality, recognizing that behavioral analytics fails if instrumentation is poor; this became a moat and a source of net revenue retention >120%.
Pricing evolved from event-volume tiers to a platform model (core analytics + experimentation + CDP + session replay), increasing average contract value and locking in multi-product adoption.
The IPO: timing, narrative, and the public-market reality check
Amplitude went public via direct listing in September 2021 at a ~$7B market cap (peaking near $10B), choosing a direct listing to avoid dilution and let existing shareholders sell; Spenser viewed it as a financing event, not an exit.
The public narrative centered on “digital optimization” — every company becoming a digital product company — and Amplitude as the system of record for product intelligence.
Post-IPO, the stock declined sharply in 2022–2023 alongside the broader SaaS correction; Spenser describes the discipline of ignoring daily stock price while hitting quarterly targets, and the importance of a long-term shareholder base (they did no buybacks, no guidance manipulation).
The experience reinforced that public markets reward durable growth and free cash flow conversion over narrative; Amplitude reached FCF positivity in 2023, which stabilized the stock.
The AI pivot: why a profitable public company chose to rebuild everything
In late 2023, Spenser concluded that LLMs fundamentally change how users interact with software (natural language > dashboards) and how analytics is consumed (answers > charts); the existing product architecture — built for point-and-click UI — was a liability.
He announced an “AI-first” reorganization: every team rewrites its roadmap around AI-native workflows, hiring is frozen for non-AI roles, and 800 employees are being retrained or redeployed; he compares it to the mobile pivot of 2012 but with higher stakes.
The new product vision: “Amplitude AI” — a conversational agent that answers product questions (“why did retention drop last week?”) by writing and executing SQL, generating charts, and proposing experiments, effectively collapsing the analyst-PM loop.
Technical strategy: build a semantic layer on top of the behavioral graph that maps business concepts to raw events, enabling LLMs to reason reliably; invest in eval frameworks and guardrails to prevent hallucinations in high-stakes product decisions.
Organizational mechanics of the AI transformation
Spenser created a centralized “AI Platform” team that builds shared infrastructure (prompt orchestration, eval harness, RAG over product docs, model routing) so product teams don’t each reinvent the stack.
He instituted a mandatory “AI literacy” program: every engineer and PM completes a 4-week internal course covering prompting, eval, RAG, and agent patterns; performance reviews now weight AI contribution.
Resource allocation shifted: 40% of engineering capacity redirected to AI initiatives; non-AI feature work is deprioritized unless it directly supports the AI agent (e.g., better session replay ingestion for context).
Cultural signal: Spenser writes the weekly “AI update” email himself, highlighting shipping teams, failed experiments, and lessons; he personally reviews every AI feature spec to maintain quality bar.
Lessons on product strategy, competition, and endurance
Category creation requires naming the problem, not just the solution — “product analytics” gave buyers a budget line and a mental model; Amplitude owned the definition for years.
Retention is the only metric that compounds — every product decision at Amplitude traces back to “does this increase the probability the customer is still here in 12 months?”; this aligns sales, product, and engineering.
Direct listing forced financial discipline early — no lockup expiration overhang, no banker-driven pricing; the team had to operate like a public company before the listing.
Incumbents rarely cannibalize themselves — Mixpanel and GA did not build behavioral graphs because it would break their data models; Amplitude’s willingness to rebuild its own foundation (now for AI) is the rare counterexample.
Founder control enables long bets — Spenser and Curtis retained dual-class voting shares, allowing the AI pivot without activist pressure; they view this as essential for multi-decade horizon.
The current state and what comes next
As of the conversation, Amplitude is ~$250M ARR, FCF positive, ~800 people, with the AI agent in private beta for 50 design partners; early signals show 10x faster time-to-insight for PMs who adopt it.
Spenser’s near-term focus: prove the AI agent drives measurable retention improvement for customers (not just engagement), then scale the motion; he treats this as a “second founding.”
He acknowledges the risk: if the agent hallucinates or the semantic layer leaks, trust evaporates; the eval framework and human-in-the-loop design are the primary mitigations.
Long-term vision: Amplitude becomes the “operating system for product decisions” — every feature launch, experiment, and strategy review runs through the platform, with AI as the default interface.
