Every spam filter makes mistakes. The question is what happens when it does.

A filter with two answers, deliver or delete, has to guess on every borderline message. Guess one way and spam reaches your inbox. Guess the other and a customer’s quote request disappears without a trace. You never see the second kind of mistake, which is exactly why it’s the expensive one.

The third answer

Cayve has three answers: allowed, held and blocked.

  • Allowed messages are delivered as normal.
  • Held messages wait for you. Nothing is lost; you decide.
  • Blocked messages are kept as spam for 30 days, in case one needs to come back.

Blocking needs two things: a high score and confidence. Each site has a hold score (50 unless you change it) and an act score (85 unless you change it). Only at the act score, and only when Cayve is sure, does the action you chose for that kind of message apply. A message set to be blocked that Cayve isn’t confident about is held instead.

What “held” looks like

For a form, the visitor sees your normal thank-you message and the entry is saved by your form plugin as usual. Only the notification email is kept back. In WordPress, under Cayve → Messages, a held message shows its reasons and a Release and send button. Press it and the original email goes out as though nothing had happened.

For comments, Cayve uses WordPress’s own moderation queue, so held comments appear where you already look.

Why the visitor never knows

Whatever Cayve decides, the person who submitted sees the same success message. That matters in both directions. A real customer is never told “your message looks like spam”, and a spammer testing your form learns nothing about what got through.

The cost of one lost lead

It’s tempting to judge a spam filter by how much junk it stops. We think the better measure is how many real messages it loses, because each of those might have been a customer. Holding the uncertain ones costs you a few seconds of review. Deleting them costs you the customer.

If you’d like to see where your current setup draws the line, run Cayve in shadow mode alongside it for a week and compare.