Self-learningConstantly improving and learning.
Once a week, the last 7 days of replies are read back and turned into additional guidance, such as which types of companies work and what tone is landing. A second model then reviews every proposal before it takes effect.
Learning is switched off. Lead selection and email writing are running on their base prompts only.
These are always additional rules. Everything below is appended to the existing rules, inside a block that states those rules take precedent. The cycle cannot quote any existing emails, dictate what a particular email says, touch recipients or sending volume, or grant itself an exception, as those are stripped in code before any model is asked for its' input. This way, there are no feedback loops.
Controls
Run cycles or disable guidance here.
Test the weekly run does exactly what would happen at Sunday, 6:00 AM: it reads the emails for that week, runs the analyst, applies the guardrails, and sends the result to the reviewer, before showing you what would change, without changing how things are currently done. It makes real model calls over a week of emails, so it usually takes one to two minutes to see the testing results. Make sure to stay on this tab to see the test's results.
In force now
Nothing active yet
No guidance has been accepted yet. The cycle needs at least 3 positive replies in a week before it will propose anything, so, nothing's changed compared to the system's usual approach.
History
Every cycle that has run.
No cycles have run yet.