Show IH: I stared at my MRR dropping for 3 weeks. Correlations told me why in 5 seconds.
Quick backstory. I've been building KPILIO for six months. Along the way I built a demo workspace called Nimbus Analytics to showcase the product. I seeded it with realistic SaaS data: MRR, churn rate, conversion rate, support tickets, 10 KPIs in total.
Then I started using it to test the product. And I found something I didn't plan for.
Between early February and mid-March, Nimbus's MRR dropped from $92K to $72K. Churn climbed from 3.2% to nearly 8%. I watched the numbers shift for three weeks without putting the pieces together.
Not because I wasn't looking. Because I was looking at them as separate charts.
Here's what I learned about correlations and why every founder tracking more than two metrics should be running this analysis weekly.
What a correlation actually tells you
A correlation is a number between -1 and +1 that describes how two metrics move together.
+1.0 means they move in perfect sync. When one goes up, the other goes up the same amount.
0 means no relationship. They're independent.
-1.0 means they move in perfect opposition. When one goes up, the other goes down the same amount.
Most correlations in real business data fall somewhere in between. Values above +0.7 or below -0.7 are considered strong. Values between -0.3 and +0.3 are usually noise.
It's not a percentage. It's not a prediction. It's a description of how two lines have moved relative to each other over the window you're measuring.
The 60-day window that clicked
When I ran Pearson correlation on Nimbus's MRR and Churn Rate across the full year, I got -0.18. Basically nothing. Weak inverse relationship, not enough to act on.
When I ran it across the last 180 days, I got -0.19. Still nothing.
When I zoomed into the 60 days where the MRR crash happened, the correlation was -0.93.
The signal was always there. It just got averaged away at longer time windows.
This is the thing I didn't understand before: correlations depend entirely on the window you look through. A business that's been stable for 11 months and volatile for 1 month will show weak annual correlations and strong monthly ones. Looking at only one window lies to you. You have to slice.
Why this matters for solo founders
The reason we don't think about correlations is that running them is annoying. You have to:
Pull both metrics' data out of wherever they live (Stripe, GA4, your database)
Align the dates so you're comparing apples to apples
Calculate the Pearson coefficient (or load everything into a spreadsheet and use CORREL)
Repeat for every pair of metrics you care about
Do it again every time you want a different time window
Nobody does this weekly. I don't even do this monthly. But when I start running them regularly, I expect the insights to be worth the time.
The practical takeaways
If you're a founder tracking metrics and want to start using correlations:
Start with pairs you suspect are related. Revenue and churn. Conversion and traffic. Support tickets and retention. Don't run every-vs-every. Start with hypotheses.
Run the same correlation across three time windows. Most recent 30 days, 90 days, 180 days. If the correlation is wildly different across windows, that IS the signal. Something changed between those periods.
Don't over-interpret correlation as causation. Just because MRR and Churn Rate move inversely doesn't mean lowering churn directly raises MRR. Both might be symptoms of a third factor you haven't identified yet. But a correlation tells you where to look.
Run them regularly. Once a quarter if you're early stage. Once a month if you have traction. The patterns shift over time, and the shifts themselves are useful information.
The KPILIO connection
I built this into KPILIO because I got tired of running correlations in spreadsheets. The Correlation Explorer lets you pick any two KPIs in your workspace, slice across any time window, and see the coefficient instantly.
Matrix view for multiple KPIs at once. Lag analysis for leading indicators.
The Nimbus case I described above is live on the KPILIO demo. You can see the -0.93 correlation across that 60-day window yourself.
We're launching on Product Hunt next Tuesday (April 28). I'd love your feedback before then. If you track KPIs and you've ever found out about a problem too late, there's a good chance this is for you.
kpilio.com | free to start, 3 KPIs forever, no credit card.
Question for IH
For fellow founders: what's the most surprising correlation you've ever found in your business data? I'm genuinely collecting these. The stories are usually better than the numbers.
This is super interesting and definitely something I've heard from others. I actually know a few solo founders and indie hackers who are actively tracking business metrics and they would probably be happy to answer your questions about surprising correlations they've found.
That'd be great, yeah. Send them my way, or just point them at this thread, whatever's easier.
I'm finding the stories are almost always better than the coefficient itself. The "wait, those two move together?" moments are the ones I'm collecting. Happy to chat with anyone who's living in their metrics. Half the reason I'm building this in the open is to hear that stuff.
The interesting part isn’t the correlation.
It’s that the problem only became obvious once the relationship was framed in a way you could read instantly.
That’s the real product.
Most analytics tools don’t fail because they lack data.
They fail because they surface numbers without surfacing meaning.
Same reason most dashboards get ignored:
the metrics are visible,
but the narrative isn’t.
That’s usually the difference between an interesting analytics tool and something founders actually build around.
This is exactly the philosophy KPILIO is built around. The detection part is the easy part now. The hard part is turning a number into a sentence a founder can act on.
A few of the things we've shipped that try to do that:
→ Every anomaly gets a plain-English explanation written by an LLM, not a z-score readout
→ Correlated alerts roll up into a single causal narrative instead of 5 disconnected pings
→ Recommended Actions suggest the next step, not just the next metric to look at
The framing layer is the product. The numbers are just the input.
Appreciate the read. If you have time before Tuesday, I'd genuinely love your feedback on whether KPILIO's framing actually clears that bar.
Exactly — that’s the bar I’m trying to hold.
If the founder still has to interpret the metric, the product stopped one step too early.
The useful threshold isn’t:
“did we detect something unusual?”
It’s:
“did we reduce the time between signal and decision?”
That’s the real test for whether the framing layer is doing its job.
A dashboard becomes infrastructure the moment the founder stops reading it as analytics
and starts using it as operating judgment.
That’s the line I’m trying to get KPILIO across.