.svg)
The Missing Layer: Why Amplitude Needs Your Context
We've been living in Amplitude for years. In that time, we've seen it all: pristine implementations, "track-everything" nightmares, and data graveyards where perfectly good events go to die.
But even in the best setups, there is a recurring ceiling. Amplitude is only as good as the data you send it. It knows your click events and your user properties, but it doesn’t know the stuff that lives in your head: your OKRs, your team's internal vocabulary, or what "conversion" actually looks like for your specific business model this quarter.
Until now, the AI had to guess. Now, with AI Context, you can finally give Amplitude its memory.
What is AI Context?
Think of AI Context as "Custom Instructions" for your analytics. It’s a dedicated space where you provide the business logic, definitions, and preferences that turn generic AI responses into tailored strategic advice.
When you use features like Analyze with AI or ask for an explanation of a metric spike, the AI doesn't just look at the numbers; it looks at the "memory" you’ve provided.
Where it Lives
You don't need a new SDK or a developer ticket. It’s sitting right in your Settings under the AI Context tab. It is split into two critical layers:
- Organization Context: Your "Global Truth." This applies to everything: company mission, high-level goals, and universal definitions.
- Project Context: Specific nuances. If you have a "Marketing Site" project and a "Product App" project, their goals are different. Project context ensures the AI doesn't talk to your PMs like they’re Performance Marketers.
The Three Pillars of Context
Amplitude categorizes your input into three buckets. To get the most out of it, you should fill them in this order:
1. Business Context
This is the most important section. Who are your users? What is your North Star? What does "good" look like right now? If the AI doesn't know you're focusing on retention over acquisition this month, its recommendations will be useless.
2. Data Conventions
Data doesn't speak for itself. Use this section to define:
- Conversion: Which specific event counts?
- Retention: Are you looking at Day 7 or Rolling?
- Glossary: Define internal terms so the AI stops guessing in Slack-speak.
3. Analytics Preferences
This is where you get "bossy" with the output. Tell the AI how you like to work:
- "Prefer weekly views over daily."
- "Always break down by Plan Type."
- "Frame recommendations as experiments to run."
Don't Write It From Scratch (The Workflow)
Most teams freeze when they see an empty text box. You don't need to be a prompt engineer; you just need a better workflow.
- Use an LLM (ChatGPT/Claude): Ask it, "Do you know what Amplitude AI Context is?" (Ensure it understands the specific 10k character limit and structure).
- Force Clarifying Questions: Don't let the AI guess your business. Tell it: "Ask me 5 questions about my business goals and data before generating my context."
- Answer and Paste: Once it has your specific answers, ask for a "copy-paste friendly format" aligned with Amplitude’s documentation.
The "Before & After" Test
Is it actually working, or just taking up space? There’s an easy way to tell.
Ask Amplitude AI a broad question like: "Which pages are users most interested in?" * Without Context: You’ll get a generic list of high-traffic URLs.
- With Context: The AI might say: "Since your goal is Lead Gen for Enterprise users, the Pricing and Demo pages are your most critical 'interest' drivers, despite lower raw traffic than the blog."
That shift: from raw math to business intelligence, is the whole point.
Maintenance: It’s Not "Set and Forget"
Analytics is interpretation. If your ICP changes or you pivot your North Star metric, your AI Context becomes "stale." When the AI starts giving answers that feel "off," it’s usually because your context doesn't match your current reality.
Review your AI Context quarterly. Keep your Organization level lean and use Project levels for the heavy lifting.
AI Context won't fix bad tracking, but it will stop your AI from being a stranger to your business.
.jpg){kind=tracking}
