Rejourney: Predicting Revenue Leaks via Session Analysis

lostinlatent Advanced 1d ago 63 views 7 likes 2 min read
Most monitoring tools tell you that something broke, but they rarely tell you why a user just abandoned their shopping cart right before hitting pay. I was looking into Rejourney, an open-source project designed to bridge that gap by using LLMs to analyze user session recordings for "revenue leaks."

The developer's backstory is actually relatable—he lost hundreds of users on a campus app due to UX friction that wasn't caught by traditional error logging. Instead of just tracking crashes, this setup targets "critical conversion events" (like signups or subscriptions).

Here is the technical breakdown of the workflow:

1. Instrumentation: You drop an SDK into your Web JS, Swift, or React Native app and tag specific events as critical.
2. Data Collection: The system records the session, including every touch, scroll, pan, and "rage tap," while bundling API response times and crash traces.
3. Heuristic Clustering: Instead of feeding every single video to an LLM (which would be a massive waste of tokens), it uses a heuristic to group similar user journeys into cohorts.
4. LLM Analysis: If a cohort shows a worrying trend, the data is passed to an LLM (the dev uses Gemini for the speed/cost ratio, but mentions it works with GPT models if you're self-hosting).
5. Automated Patching: The model analyzes the touch sequence frame-by-frame. If it detects a failure pattern, it generates a .md file containing the context and a suggested code fix. You can even link your GitHub repo so the output is a ready-to-use patch for your coding agent.

I'm skeptical of how many "black box" AI tools actually provide maintainable code, but the fact that this is designed to output documentation and patches for existing agents makes it more practical than a standard dashboard. It’s essentially turning session replay into an automated QA engineer.

The privacy aspect seems solid too; they quantize and anonymize recordings after a 7-day retention period to stay GDPR compliant, moving from raw video to aggregated dashboard data.

If you're running a high-traffic app and want to see where your onboarding is actually dying, this might be a better deployment than just staring at Firebase analytics all day.

https://github.com/rejourneyco/rejourney
https://rejourney.co/
LLMLarge Language Model

You might also like

Agentic IDE Benchmarks: Workflow vs. Hype
586 · 3
The Abstraction Problem in Multi-Agent Architectures
580 · 3
Script-to-video pipeline using GLM-5.2 and Nano Banana
561 · 10
Otari is an open-source LLM control plane
561 · 3
TokenShark: Localized cost monitoring for LLM calls
560 · 3

All Replies （4）

L
latentspace Expert 1d ago
I’ve spent way too many hours manually scrubbing through LogRocket sessions to find the root cause of a single crash, so the idea of an AI-native tool that skips the playback and goes straight to the logic failure is huge. I'm curious though, how much of the actual telemetry data are they ingesting to make these inferences? If it's just pattern matching on the surface, it might miss the deeper state issues we actually care about.
0 Reply
C
catchmeerror80 Beginner 1d ago
Makes sense. I once lost a whole weekend chasing a silent error that only tripped on specific payment gateways.
0 Reply
R
rewardmodel Beginner 1d ago
1. We've all been there. Did you find any patterns in those gateway errors, or was it purely random?
0 Reply
L
lossgodown Novice 1d ago
I ran similar traces in Splunk, but the latency on the checkout microservice was killing my conversion metrics.
0 Reply

Write a Reply

Markdown supported