Rejourney——发现应用和网站中的用户“泄漏”

Hi there HN, Ever used a user/session replay tool before?Absolutely phenomenal for immediate feedback (and your dopamine).It causes a “Honeymoon Phase” where you are addicted and constantly checking out how users found your app, the outcome, and if they faced any roadblocks with the UX.After a week, you maybe start marketing your app or website, and the sheer influx of session replays gets you excited.You can’t check all of them so you check the first few, or maybe query for certain user paths to check out success/failure from the session you randomly pick.Whatever provider you use has an “AI replay watching” feature that you hope will save time.

You try it out, and it ends up summarizing a few select sessions, misses things you clearly saw in the random replays you decided to watch.Another week passes, and you no longer check these replays except maybe at the end of every week just to reassure yourself everything is good.Well.It was fun while it lasted.But it’s starting to get expensive and my team can save money by just turning it off.Basic observability ends up doing the same end-thing at this point because we don’t even watch a percent of the replays.Rejourney was initially a basic session replay platform like all the rest.The difference was a focus on performant and smaller package footprints.

But recently, we began working on an idea called “leaks.” The name originates from a user leak.The core concept of what a session replay is meant to do.Basically, a place where users "leak out" of the intended experience such as completing onboarding.Unlike other tools that attempt to do this with AI summarizers, we employ a lot of heuristic-focused math and algorithms to process at scale.

This helps catch prevalent AND non-prevalent issues, ranking in terms of wideness of impact, effect on revenue, or any desired outcomes: Weighted evidence from: Temporal corroboration - checks signals happening together Time-decay weighting - prioritizes recent user problems Exploration sampling - catches unknown failure patterns Evidence grounding - verifies claimed supporting evidence Set intersection - removes unsupported event claims Jaccard similarity - matches similar visual text Linear interpolation - remaps compressed replay timestamps Structural clustering - groups related user failures Prevalence estimation - estimates overall issue impact Has it been proven in the field?Yep.

It’s been helping about 200 apps/websites for two months now.The amount of issues it has been able to find that would otherwise be missed resulted in a few joyful emails sent our way.Furthermore, apps with larger volume of users seem to use the tool more heavily than those with say, only a few hundred active users a day.We’re in the works of an MCP to help faster dump the found leaks into whatever coding tool your team uses.For now, the good old copy/pasting of the .MD issue context will need to be employed.Demo of leaks: https://rejourney.co/demo/leaks General web