How I Found a 40% Drop-off (and What the Funnel Didn't Tell Me)

The funnel is honest but shallow. It will tell you where people leave with brutal precision, and tell you why — never.

On an AI writing SaaS, a cohort analysis in Mixpanel surfaced a 40% drop-off at onboarding Step 3. That's the easy part. The hard part is resisting the urge to immediately "fix Step 3" before you understand it.

Step 1 — Let the data pick the fight

I didn't start with opinions. I started with the funnel, then a cohort split: new vs returning, self-serve vs sales-assisted, device, plan tier. The drop-off concentrated in new self-serve accounts on their first session. That single cut turned a vague problem into a specific one.

Step 2 — Watch people fail

Numbers say 40% leave. Hotjar session replays showed me how. Users hit Step 3, hovered, scrolled up and down, and left. They weren't confused about the button — they were confused about why the step existed at all. We were asking for information the product hadn't yet earned the right to ask for.

Step 3 — Interview until it's boring

I ran fortnightly research with 30+ customers. The insight repeated until it was boring, which is exactly when you know it's true:

"I just wanted to see if it could write the thing. You asked me to configure my whole workspace first."

Step 4 — Test the fix, don't ship the opinion

We rebuilt the flow across two sprints and ran an A/B test across 1,200 users rather than shipping on conviction. The redesign delivered a 32% activation lift and earned approval for full rollout.

The takeaway

Quantitative data earns you the right question. Qualitative data earns you the right answer. Shipping either one alone is how good teams confidently build the wrong thing.