Judgement calls
Things left out on purpose, and why. The brief asked for thinking, not a perfect answer, so this page is the thinking behind the cuts.
No fabricated numbers
The funnel dashboard runs on illustrative data. Nothing on the slides claims a SigNoz activation rate, churn rate, or close rate. I don't have that data, and a made-up number presented as theirs is the one mistake that can't be walked back.
No machine-learning model
A supervised model on closed-won versus closed-lost is the right long-term way to set scoring weights. I've built the equivalent for keyword intent before. It's not in this deck because SigNoz doesn't have the deal volume to train it yet. Proposing it now would be proposing a tool for a problem that doesn't exist. The scoring rules say what to do instead: instrument, tag outcomes, let the data set the weights when there's enough of it.
No forecasting layer
Behavioural forecasting of pipeline needs a few hundred closed deals to be worth anything. Same reason as above.
No new tools beyond two
The question named four tools. That's a test of whether I can run what's already paid for. It isn't an invitation to rebuild the stack.
Every tool you add is a bill, a login, and something one more person has to learn. They'd be learning it at the same time they're learning the signals.
So the bar was simple. Name the gap the four tools can't close. Then add the smallest thing that closes it.
Segment passed. Nothing else carries product events into HubSpot. A warehouse passed. The score has to be computed somewhere that can do joins, and HubSpot can't.
Redis didn't pass. Common Room didn't pass. A second CDP didn't pass. Each one solves a problem SigNoz doesn't have yet.
One more reason. The first month is for proving the motion with what's there. A new tool that shows up before the first win looks like spending. After the first win it looks like building.
No weights on day one
Assigning points to signals before there's conversion data is a guess dressed as a model. The rule ships first. Weights come from outcomes.
The score is not a verdict
It's a sort order with its reasons attached. Reps overrule it. When an experienced rep overrules it and wins, that's the most useful data the system will ever get.
Two stages, not four
Q4 allows one or two. Picking two that work by different mechanisms (product-event emails, signal-triggered sales follow-up) says more than covering more of the funnel with the same mechanism.
Q5 kept brief
The question said briefly. The stack wiring page is longer than the slide because this is the appendix. The slide is one diagram and two additions.
What I'd do in week one
Before any of the above: find out what fraction of users are self-hosted, what fraction of closed deals came from a migration signal, and whether docs traffic is tracked at page level. Those three answers decide how much of this plan applies as written.
Source: signoz_assignment_q3.md appendix; signoz_assignment_progress.md session notes, 5 and 6 Sep 2026.