92 emails in my inbox, 2 left after Jev sorted them
I run 10+ products solo and read the support inboxes myself. After a press release for one of them, the contact address got buried in pitches: webinar invites, free whitepapers, agencies wanting to run my ads. Gmail filters didn't hold. Senders change every time, and "thank you for your inquiry" shows up in both auto-replies and real human replies.
The worst ones borrow a platform. Someone shares a pitch as a Google Doc, so the email comes from Google. Or they sign you up for their Zoom webinar, so the invite comes from Zoom. You can't send Google or Zoom to spam.
Jev is what finally fixed it. It doesn't write anything, it just tells you how likely each yes/no answer is, and you decide where the line goes. Per email I ask five yes/no questions in one call: did a person write this, should we reply, are they selling us something, do they want a meeting, are they clearly asking us to stop. A few rules around it:
Ask twice, act only if both agree. Otherwise I look myself.
I set the thresholds from 26 of my own emails. The 0.9/0.1 defaults left almost everything undecided, probably because my mail is mostly Japanese. Real replies landed 0.61 to 0.93 on "should we reply" and pitches 0.05 to 0.49, so the line went at 0.6.
"Never contact them again" needs 0.7, since you can't take it back.
Bounces and anything about invoices never go to the model.
On October 2 there were 92 emails in the inbox. After the sort, 2 were left: a Google Workspace invoice and one reply I actually needed to answer. It runs every five minutes now.
What are you using Jev for in the products you run? Or where are you thinking of trying it?
Replies
the "ask twice, act only if both agree, otherwise I look myself" rule is the real insight here, not the thresholds. that's you admitting the model's confidence number and its actual reliability aren't the same thing, so you built a disagreement check instead of trusting a single score. curious how often the two calls actually disagree on your mail - is it rare enough that the manual-review pile stays small, or common enough that you're still looking at a meaningful chunk of the 92 yourself most days? also wondering if the senders who borrow a platform (Google Docs, Zoom invites) start adapting their wording once a tool like this gets popular enough, the same way spam adapted to keyword filters.
@galdayan Rare enough that I don't think about it. On October 2, the only 2 left for me out of the 92 were the Google Workspace invoice and the one reply I needed to answer.
On the Google Docs and Zoom ones, probably yes. But however they word it, it's still someone selling something, and that's what I'm asking about.
@hisashispace that "it's still someone selling something regardless of wording" line is actually the whole defense against the arms race - you're asking about intent, not surface pattern, so rewording the pitch doesn't move the needle the way it would against a keyword filter. the failure mode I'd worry about more is the other direction - a genuine reply from an actual customer that happens to sound promotional, like someone thanking you and mentioning they told a friend about your product. has that ever gotten misread as a pitch and almost got buried with the rest?
@galdayan Not that I've caught so far, but it's only been running since September 30, so I wouldn't call that proof. That's the case the "ask twice" rule is there for. If the two answers don't agree, it stays in my inbox and I read it myself instead of it getting archived.