.svg)
How to Set Up Amplitude AI: A Practical Guide
Amplitude AI is a suite of features that lets product and growth teams ask questions about their data, generate charts and dashboards automatically, surface anomalies, and produce written summaries of performance without manual analysis. Setting it up correctly requires more than just turning it on. This guide covers every step: AI Context, Dashboard Agents, Chat, and what needs to be in place for Amplitude AI to produce reliable, actionable output.
Amplitude AI is not a plug-and-play feature, it's a layer that sits on top of your existing setup. Feed it clean data, good structure, and rich context, and it produces insights that genuinely accelerate decisions. Feed it messy events and blank context fields, and it produces confident-sounding noise.
This guide covers how to set it up properly from the start.
What Amplitude AI actually includes
Before getting into setup, it helps to understand what you're actually configuring. Amplitude AI is not a single feature, it's a collection of AI-powered capabilities that work together.
AI Context is the foundational layer. It's where you tell Amplitude about your business: who your users are, what your product does, which metrics matter, and what decisions your team is trying to make. Every other AI feature in Amplitude draws on this context to generate relevant output.
The Dashboard Agent reads the charts on any dashboard and automatically generates a written summary of what the data shows: trends, changes, and notable patterns. It's designed for recurring reporting workflows where you need consistent analysis without manual effort.
Amplitude Chat is the conversational interface. You ask questions in plain English "why did conversion drop this week?" or "show me retention by cohort for new users in March" and Amplitude generates charts, runs analyses, and surfaces insights in response.
Anomaly detection runs in the background and flags when a metric moves significantly outside its normal range — useful for catching problems before they become visible in a weekly review.
Step 1: Set up AI Context
This is the most important step and the one most teams skip. Everything else in Amplitude AI depends on it and it's the primary reason AI in Amplitude fails to produce useful output for most teams who try it.
Navigate to your organization settings in Amplitude and find the AI Context section. You'll see fields at both the organizational level and the project level. Fill in both.
At the organizational level, describe your business model, your product, and your target users. Be specific. "We are a B2B SaaS company selling project management software to operations teams at mid-market companies" is useful context. "We are a software company" is not.
At the project level, describe what that specific Amplitude project tracks: which product, which user segment, which part of the funnel. If you have separate projects for your web app and mobile app, each one needs its own context that reflects what it measures.
Also fill in your key metrics. Tell Amplitude which events represent activation, conversion, retention, and revenue for your product. This is what allows the AI to connect data points to business outcomes rather than just reporting raw numbers.
The more specific your AI Context, the more specific every AI output will be. This is worth spending an hour on properly.
Step 2: Structure your dashboards for AI
Amplitude AI works best when your dashboards are organized around specific business questions rather than collections of loosely related charts.
Before activating any AI features, audit your existing dashboards. Each dashboard should answer one clear question: weekly conversion performance, onboarding funnel health, feature adoption by cohort, retention by acquisition channel. If a dashboard has twenty charts covering different topics, the AI will produce an unfocused summary.
Restructure dashboards around questions your team actually asks weekly. Add titles and descriptions to each chart that explain what it measures and why it matters. These annotations feed directly into AI analysis, a chart labeled "Add to Cart → Purchase conversion rate, 7-day rolling" produces a much more specific AI summary than an untitled chart with the same data.
Step 3: Activate the Dashboard Agent
With AI Context filled in and dashboards structured, activating the Dashboard Agent is straightforward. For a full step-by-step walkthrough of how the Dashboard Agent works and how to get the best output from it, see our complete guide to Amplitude AI Dashboard Agents.
Open the dashboard you want to automate. Find the AI settings and activate the agent. It will immediately read all the charts on the dashboard and generate its first summary. Review this output carefully, it tells you a lot about the quality of your context and chart structure.
If the summary is generic, go back to your AI Context and add more specificity. If it mentions things that aren't accurate, check whether your event tracking is consistent and whether your chart descriptions are clear.
Once you're satisfied with the output quality, the agent refreshes automatically each time you open the dashboard or trigger a new analysis. For weekly reporting, the workflow becomes: open dashboard, review the AI summary, approve and share. The analysis happens automatically.
Step 4: Configure Amplitude Chat
Amplitude Chat works out of the box once AI Context is in place, but there are a few things that make it significantly more useful.
Give it context about who is using it. If your Amplitude project is used primarily by your marketing team, note that in the project-level AI Context. The Chat feature will adapt its language and frame its responses accordingly, marketing-oriented questions will get marketing-oriented answers rather than engineering-focused ones.
Start with questions you already know the answer to. The best way to validate that Chat is working correctly is to ask it something where you can verify the output against what you'd build manually. "Show me conversion from signup to first purchase over the last 30 days" if that matches what you'd build yourself, the setup is working. If it doesn't, something in your event structure or AI Context needs attention.
Use Chat for exploration, not final reporting. Chat is best for quickly surfacing data, exploring hypotheses, and building one-off analyses. For recurring reports and structured summaries, the Dashboard Agent is the better tool.
Step 5: Clean up your event tracking
This step should ideally come before all the others, but in practice most teams are setting up AI on top of an existing implementation. Either way, event tracking quality is the single biggest factor in AI output quality.
Start with a clean tracking plan before implementing or cleaning up events. This gives you a single source of truth for what should be tracked, how events should be named, and what properties each event needs to carry.
Check that your events have consistent naming conventions. Mixed conventions: some events in snake_case, others in Title Case, others abbreviated, make it harder for AI to understand relationships between events.
Make sure key events have descriptions. In Amplitude's event taxonomy, you can add plain-language descriptions to each event. These descriptions feed directly into AI analysis. An event called "btn_clk_cta_hero" with no description is opaque. The same event with a description "User clicked the primary CTA button on the homepage hero section", gives the AI the context it needs to use that event correctly in analysis.
Resolve any identity issues. If your user IDs aren't consistent, anonymous users not being merged with identified users after login, for example: AI analysis will produce fragmented user journeys that don't reflect reality. The most common Amplitude SDK issues that affect AI output quality almost always trace back to identity resolution problems or inconsistent event firing.
What to expect after setup
With AI Context filled in, dashboards structured, and event tracking clean, Amplitude AI produces output that's qualitatively different from what most teams experience when they first turn it on.
Dashboard summaries become specific to your business rather than generic descriptions of chart shapes. Chat responses reference your actual metrics and user segments rather than generic analytics concepts. Anomaly detection flags things that are genuinely unusual rather than surfacing noise.
The ongoing maintenance is light. Review AI Context quarterly to make sure it still reflects your business accurately. Add event descriptions as new events are implemented. Restructure dashboards as your reporting needs evolve.
Learn how to automate reporting with AI Agents
Stop spending hours on analytics reports, learn how to make Amplitude's AI do it for you in 30 minutes.
Is your Amplitude setup AI-ready?
If you're not sure whether your events, context, and dashboards are configured correctly for AI, the AI Readiness package covers exactly that.
Explore the AI Readiness package →
Want to talk through your setup?
Book a 30-minute call with Gregor →
FAQ
What is Amplitude AI?
Amplitude AI is a suite of AI-powered features built into Amplitude that includes AI Context, Dashboard Agents, Chat, and anomaly detection. Together these features allow product and growth teams to automate analysis, ask questions about their data in plain English, and generate reports without manual effort.
Do I need to pay extra for Amplitude AI?
Amplitude AI features are available on paid plans. Specific feature availability varies by contract tier. Check your current Amplitude plan or contact Amplitude directly to confirm which AI features are included.
Why is Amplitude AI giving me generic answers?
Generic output almost always comes from missing or incomplete AI Context. Navigate to your organization and project settings in Amplitude and fill in the AI Context fields — describing your business model, users, key metrics, and decision framework. This is what makes AI output specific and actionable rather than generic.
How long does it take to set up Amplitude AI properly?
The technical setup — activating Dashboard Agents and Chat — takes under an hour. The more time-intensive part is filling in AI Context properly and restructuring dashboards around clear business questions. Expect to invest two to three hours for a thorough initial setup, with lighter ongoing maintenance after that.
What event tracking quality do I need for Amplitude AI to work well?
Your events should have consistent naming conventions, plain-language descriptions in Amplitude's event taxonomy, and correct user identity resolution. You don't need perfect tracking — but significant inconsistencies in naming, missing descriptions, or fragmented user identities will noticeably reduce AI output quality.
