Okay, hear me out. I know everyone's obsessed with fancy analytics dashboards (and don't get me wrong, I love a good Amplitude session), but you'd be surprised what you can pull off with just your data warehouse. I was helping a friend at a SaaS company (around 50k MAU, subscription model) who didn't have a complex event pipeline set up yet.
They already had their core user events flowing into BigQuery via Segment. The question was simple: "Are people who use feature X more likely to convert to a paid plan in their first week?" Instead of setting up a whole new tool, we wrote a query that basically rebuilt a funnel.
The trick is using window functions to track each user's journey through key events over time. You can map out sequences like: `signed_up` -> `activated_feature_x` -> `started_trial` -> `paid_subscription`. By counting users at each step and comparing cohorts (those who did vs. didn't hit an intermediate step), you get classic funnel conversion rates. It's not as pretty as a dedicated tool, but it's incredibly flexible and directly answers specific business questions.
The real power? Since it's your raw data, you can slice by any other attribute you've collected—plan type, traffic source, device—without pre-defining those dimensions. It's perfect for one-off, deep-dive questions or for companies that are data-rich but tool-budget-poor. Makes you appreciate having a solid CDP/Segment setup even more, because once the data is clean in BigQuery, the analysis is all yours.
p-value or it didn't happen
You've touched on a crucial, often overlooked principle, data warehouse as a computation layer. This approach eliminates the synchronization latency and potential data loss you'd get piping events to a third-party analytics tool. The ability to join your event sequence directly with your user dimension table for slicing is a killer advantage.
A key caveat with this method in BigQuery is the computational cost of those window functions over large, unbounded event histories for your entire user base. For recurring analysis, materializing the user journey stages as a separate table or view can be more efficient than running the full sequence logic on the fly each time. Also, while you can track "time between steps" easily, attributing drop-offs to specific UI elements becomes harder without the session replay context a tool might provide.
The flexibility is indeed the main benefit. You can seamlessly incorporate backend business logic, like a user's computed credit score or internal segment flag, into the funnel analysis without building a custom integration. That's something most out-of-the-box tools struggle with.