Multi-step sequences, each step with variants. Once a week the variants are compared and the one that is losing is paused so a person can look at it. That is the single scheduled job in the product permitted to write anything, and pausing is all it can do.
What the optimizer did: paused B and left a note. What it did not do: edit B, write a C, or move A’s budget. You decide what happens next.
A reply rate that counts every inbound message counts every out-of-office and counts one keen person three times. It reads about twice the truth. This product counts one distinct person with at least one genuine reply, and shows the breakdown beside it.
| Number | How it is counted | When it is a dash |
|---|---|---|
| People contacted | Distinct leads with at least one send. | Never; sends are logged locally. |
| People replied | Distinct leads with a genuine inbound: not out-of-office, not unsubscribe. | When the inbound sync has not run for that campaign. |
| Automated | Out-of-office and similar, shown as their own count. | Same as above. |
| Reply rate | People replied over people contacted, per campaign. | When either side is unknown. It is not computed from a partial pair. |
| Open rate | Only when tracking is on. | Absent on purpose when tracking is off. An empty column beats a made-up one. |
| Bounces | Not shown as a message count. | Always: a bounce is a delivery report, not an inbound message, so that count would always have been a fabricated zero. |
Launch, pause and resume are clicks. Lanes derived from a campaign you launched — the resume after an out-of-office, a copy addressed to a referred contact — run inside that launch, and only while five guards hold: the parent is still active, the operation is supported, the lane is one the product itself created and recorded, the per-lead guards are unchanged, and the workspace has not opted out. Read more on the security page.
Variants, windows, limits and a weekly pause you can overrule.