Amplitude

Amplitude is a product analytics company that helps teams understand user behavior and improve digital products. Its AI products have evolved from generating charts from questions to conversational analysis, background monitoring, and connecting behavioral data to external AI tools. By August 2026, it had also introduced Agent Analytics, connecting AI-agent performance to business outcomes, and Wave, which proposes prioritized product opportunities and execution plans from analytics, session replays, feedback, and experiments.[1][6][9]

Amplitude’s analytics business predates these AI products. It grew out of tools built for Sonalight, a voice-based messaging application—an earlier product that explains how the founders encountered the analytics problem.[2][1]

Spenser Skates and Curtis Liu brought Sonalight into Y Combinator’s Winter 2012 batch. In his 2021 retrospective, YC’s Geoff Ralston recalled that they were searching for another product by spring 2012 and settled on mobile analytics; Jeffrey Wang joined them as a cofounder around that transition. The team’s internally developed analytics tools had attracted requests from other YC developers. Amplitude made its public debut on February 21, 2014, as a mobile analytics service.[3][2]

Wang later recalled that making Sonalight successful required advances in voice recognition beyond the founders’ software-engineering background. Analytics offered a different technical direction: distributed systems and understanding user behavior, the areas in which he contributed. He also recalled early customers from gaming companies and startups founded by former gaming employees, who brought experience measuring engagement and refining product loops.[1]

On June 9, 2021, Amplitude announced a $150 million Series F, led by Sequoia Capital with participation from GIC, Battery Ventures, and IVP. It reported total funding of $336 million and a $4 billion pre-money valuation. The announcement followed the launch of its Digital Optimization System. Amplitude went public on September 28, 2021; subsequent company announcements identify it as Amplitude, Inc., Nasdaq: AMPL.[4][3][5]

In a June 2023 interview, Wang, then chief architect, and Joe Reeve, who ran AI R&D efforts, described Amplitude’s first LLM-based feature: question-to-chart. It had begun rolling out to AI design partners. Customers could ask a question about their data and receive a chart. Existing predictive ML products already helped customers anticipate behavior, such as purchasing or churning; traditional ML also narrowed the information supplied to the new LLM workflow.[1]

Reeve described splitting chart generation into seven component questions after encountering hallucinations and failures to recall context. Components selected events, chart types, metrics, and names rather than asking a model to produce an entire chart definition at once. The workflow generated output for Amplitude’s custom query engine, rather than SQL, and then used GPT-3.5. A small internal TypeScript wrapper grouped calls into an inference transaction, monitoring both the overall request and its individual inferences so failures could be traced to particular components.[1]

Wang described testing event-intent matching with a general model, then adopting an embedding approach that he said was faster and cheaper while retaining reasonable quality. Reeve and Wang described Postgres and pgvector for searching taxonomy embeddings, such as event names for browsing, checkout, and purchase. This was a limited rollout with a different workload from ingestion: it stored taxonomy vectors, rather than vectors for every user or behavioral event.[1]

Amplitude used its own analytics to track inference failures, causes, and retries. Its instrumented chart editor also linked users’ changes to the original generation: removing an event or correcting a metric supplied feedback about the specific choice that needed revision. Reeve proposed comparing prompt variants through A/B tests and measuring whether users used or shared generated content; prompt A/B testing was still a next step. Wang separately described an internal prototype combining LLM-generated copy with multi-armed-bandit selection, an experiment toward his vision of self-improving products.[1]

Wang argued that natural language made analysis easier to begin, especially for people unfamiliar with their data, while precise calculations still required explicit semantics. Reeve emphasized that users should be able to understand, edit, stop, and retry model actions. These design concerns accompanied the effort to make analytics more accessible without hiding the choices behind an answer.[1]

On October 15, 2024, Amplitude announced that it had acquired Command AI, formerly CommandBar. It planned to integrate personalized user assistance through nudges, product tours, onboarding guides, surveys, and checklists. The acquisition extended its intended role from showing where users encountered difficulty to helping teams intervene within their products. Command AI’s team would join Amplitude and continue serving existing customers during integration.[5]

On October 21, 2025, Amplitude released its Model Context Protocol server to all customers, making behavioral data accessible from tools such as Claude and Cursor. Separately, its AI Agents open beta began rolling out that week, following a limited June beta. Dashboard Agent analyzed metric changes and supported follow-up questions; Session Replay Agent surfaced patterns and friction, with playlists of relevant replays. Both could deliver insights proactively.[6]

Amplitude’s 2026 agent account describes Global Agent as its primary conversational interface. It searches existing dashboards, notebooks, replays, and AI feedback before creating new analysis when needed, returning insights supported by an explorable live dashboard. Specialized agents monitor dashboards, review session replays, summarize customer feedback, and identify conversion drop-offs. Global Agent can also create behavioral cohorts and draft and deploy in-app guides to them.[7]

Amplitude describes agents assembling business context, project data descriptions, page context, chat history, and uploaded files before acting, with context organized at company, project, and agent levels. Event names, chart titles, experiment results, and replays supply a semantic foundation for analysis. Different agents use models from OpenAI, Anthropic, and Google through an internal orchestration layer. The company reports zero-data-retention agreements with major model providers and says customer data is not used to train external models.[7]

On May 5, 2026, CEO and cofounder Skates announced a strategic partnership with Statsig. Amplitude would take on Statsig’s brand and customers, maintain and develop its cloud and data-warehouse platform, and build a more integrated roadmap. It would work with the Statsig team at OpenAI during the transition. Skates framed the arrangement around evaluating software, measuring its impact after release, and deciding what to roll back or build next as AI accelerated code generation.[8]

For the quarter ended June 30, 2026, Amplitude reported revenue of $100.9 million, up 21% year over year, and ARR of $410 million, up 22%. CFO Andrew Casey attributed $17 million of quarterly ARR additions to Statsig integration and $19 million to core Amplitude execution. ARR is a subscription operating metric, separate from recognized revenue and a revenue forecast.[9]

Amplitude’s investor overview reports approximately 5,200 customers as of June 30, 2026, and 27% of Fortune 100 companies as customers as of December 31, 2025. Its Q2 results reported 824 customers with at least $100,000 in ARR. The unaudited results also showed a $35.2 million GAAP operating loss and $25.6 million in operating cash flow for the quarter.[10][9]