A Working Prototype
CARRavan.
Want to explore it yourself? Skip to the live demoMost SaaS leaders can recite their NDR number. Few can explain why it moved, or whether it's sustainable if it's trending positive. That gap is usually about tooling and data, not talent. Mid-market and turnaround-stage companies rarely have clean CRM, CSP, BI, and NPS integrations, so the data needed to diagnose NDR/GDR sits scattered across exports nobody fully trusts. In many cases there's little hard data, but strong anecdotal evidence that matters just as much to the analysis.
I built CARRavan as a working prototype to prove that approach works: skip the perfect stack, start with simple CSVs, and you can still see exactly where retention is leaking and why.
Screenshot 1
CEO/CFO Dashboard
This is the board-ready view: NDR and GDR calculated the way finance actually argues about it. I built support for both calendar-year-to-date and rolling-twelve-month methodology, because I've sat in enough budget cycles to know a CFO and an operator can each have a legitimate reason to want a different number.
The same dashboard goes deeper below the fold: churn and expansion broken out account-by-account, and the same NDR/GDR numbers sliced by industry, service model, and CSM owner, so a headline retention number never hides which segment is actually driving it.
Screenshot 2
Account Detail / Sterling Logistics
An aggregate NDR number can hide the real story. Sterling Logistics' engagement scores, activity and health, dropped to their lowest point right as a renewal came in contracted. That's not something you'd catch from the dashboard above. It only shows up once you drill into the account's own timeline.
What happened next is the more interesting part: both scores recovered steadily over the following months, and the account expanded by August. The leading indicator was visible months before the financial outcome. That's the whole argument for tracking engagement data alongside revenue instead of waiting for revenue to tell the story on its own.
(Correlation, not causation — the data shows scores and outcomes moving together in sequence. I'm not claiming the score recovery caused the expansion, just that it preceded it and tracked with it.)
Screenshot 3
Settings / Methodology Toggle
Calendar-year and rolling-12-month NDR are both legitimate, and they often disagree. I built support for both, and the system won't even offer rolling-12 as an option until there's a full 12 months of data behind it. No methodology gets used past what the data can actually support.
Where this is headed
CARRavan is a functional prototype, not a finished product yet, but I'm not far off. Building it surfaced the same judgment calls I'd bring to a real engineering org when advocating for a customer-driven enhancement. I had to scope a feature down to what the data could actually support, which is where methodology gating came from. I had to make sure customer value was actually being delivered, not just the feature itself. And I had to figure out what really mattered first, especially once something had to ship in phases. That's the same rigor I bring to retention diagnosis itself: don't trust the first answer, check it against the actual output, and say plainly what's still uncertain instead of papering over it.
This has been one of the best learning experiences of my career, and it's changed how I think about what's possible for a CS organization. I'm now building out other Customer Success AI tools, using that same judgment to decide what's actually worth building next. I built CARRavan using a combination of Claude, Lovable, ChatGPT, and Gemini, and I'm continuing to build with that same stack. If you want to see where it's headed, I'd love to show you.
Live Demo
See it for yourself
This page shows you the highlights. The live demo lets you click through CARRavan directly — read-only, no sign-up required.