Context
At Census (acquired by Fivetran), our own CS team had the same problem as everyone else's: thousands of product events per account and no time to read any of them. I built this to fix that. The approach - summarizing raw event data into something a CS team can actually use - works for any company with event data in a warehouse.
The idea
Our customer success team had thousands and thousands of rows of event data per account. Nobody was sorting through that. They needed a summary - what's going on with this account, in plain language, updated daily.
The output: a 30-day product usage summary per account with growth signals, areas of concern, key feature usage, and suggested talking points. All of that from raw event data, no manual analysis.
How I built it
Step 1: Prep the data
A simple SQL query that creates one row per account with a JSON array of all events completed in the selected time period. You can add whatever data points are useful for your summary - we kept it simple to start.
Step 2: Generate the summary
I used AI to take that JSON of events and produce a human-readable summary. The prompt is set up to output a few specific things: an overview of the account's activity, growth signals, potential areas of concern, and talking points the CS team can use in their next conversation.
Preview it, make sure it reads well, tweak the prompt until you're happy with the format. We also added some formatting so it looks clean in the end tool.
Step 3: Sync to Salesforce
Set up a daily sync that pushes the usage summary to the account record in Salesforce. The CS team opens an account and sees a fresh summary of what's happening - no dashboards to check, no event data to dig through. For you it could be whatever tool your CS or sales team lives in.