The Email Agent

This page explains how the agent actually decides things: who it picks, how many it sends, when it sends, and how to read what it did.

Everything here lives under Email Campaigns → Email Agent, which has four tabs: Overview, Queue, Activity, and Settings.

Engagement Tiers​

Before the agent can choose who gets an email, it needs to know who is worth emailing. It sorts every contact into one of seven tiers, and it re-checks them regularly.

Tier	What it means	Example
Hot	Clicked, replied, or opened more than once in the last 30 days. Your most engaged people.	Clicked your pricing link last Tuesday.
Engaged	Opened, clicked, or replied in the last 3 months.	Opened your June newsletter, nothing since.
New	Added in the last 30 days and sent fewer than 3 emails. Too soon to judge.	Signed up on Monday.
Cooling	Nothing opened or clicked for 3 to 6 months. Due for a win-back.	Used to open everything, went quiet in spring.
Dormant	Nothing for over 6 months, or 10 emails with no opens at all.	On the list since last year, never once opened.
Sunset	The agent stopped trying after the win-back attempts ran out. Any open or click brings them back.	Rested after three win-back tries got no response.
Unrated	No emails sent to them yet, so there is nothing to go on.	Imported this morning.

Only real people count​

Mailboxes are full of robots. Security scanners open every link in every email before the human sees it. Some inbox apps pre-load images. If those counted, half your "engaged" contacts would be software.

Bots and email scanners do not count. Only real opens and clicks change a contact's tier. Every number the agent shows you is people.

Two tiers the daily send never touches​

Dormant and Sunset contacts are reached only through the win-back budget — never through the normal daily send. That is on purpose: they are the contacts most likely to hurt your inbox placement, so they get a small, controlled trickle instead of the full flow.

Each contact also carries an engagement score from 0 to 100. Recent actions count for more than old ones, so the score fades if someone goes quiet. It is a quick way to sort contacts by who is warmest right now.

The Queue​

The queue is your running order. The campaign at the top is the one the agent is working on.

To add a campaign: open it, choose Email Agent under Delivery, and click Hand to the email agent. To reorder, drag a row on the Queue tab or use the arrow buttons.

Queue rows carry badges so you can see at a glance what will happen:

• Evergreen — the content stays true over time, so it can be reused for win-backs later.
• Expires — followed by a date. The campaign stops then, no matter how far it got.
• Expired — that moment has passed. It will be skipped.
• Not reused — you turned off automatic re-sends for this one.

One campaign, many days​

This is the single most important behaviour to understand.

The email at the top of the queue stays at the top across days. Every day it goes to more of the people who have not received it yet.

Say you queue a campaign for 50,000 contacts and today's safe volume is 5,000. The agent sends 5,000 today. Tomorrow it sends the next batch. It keeps going until one of two things happens:

• Everyone has had it. The campaign is marked sent and drops out of the queue. The next campaign moves up — possibly the same day.
• The content expired. The agent stops and records how far it got, for example "Stopped: content expired. Reached 7,400 of about 50,000."

Nobody ever receives the same campaign twice. The agent checks what already went out before it plans each day.

Progress shows on the row: "12,300 of about 50,000 sent · continues tomorrow".

When the agent skips something​

A campaign at the top of the queue can be passed over. It is always removed from the queue and always named in the receipt — nothing ever disappears quietly. There are four reasons:

Reason shown	What happened
the content had expired	The expiry date passed.
its sender was not the built-in email sender	The sender was changed or deleted after the campaign was queued.
it targeted no subscriber list	The campaign lost its lists.
everyone it could reach had already had it	Finished. The next campaign takes over immediately.

The Daily Check-In​

Once a day the agent checks in and writes one plan. The Overview tab shows when: "Next check-in: 4:30 AM". You can also force one early with Run check-in now.

Shortly before each check-in, every contact's engagement tier is recalculated. So the plan is always built on today's picture of who is engaged, not last week's.

In Suggest mode the plan waits for you. In Auto mode it runs straight away.

Waves​

The agent does not send a day's mail in one blast. It splits it into waves, and your most engaged contacts go first.

With Send in waves on (the default), a normal day looks like this:

1. Wave 1 — your Hot contacts, on their own.
2. Wave 2 — the rest of your chosen tiers together.
3. Then one wave per win-back email, if there are any.

Why that order matters: mailbox providers watch the first reactions to a send. If the first wave is your Hot contacts, you get opens and clicks quickly, and the provider reads the rest of the send as wanted mail. Start with your least engaged people and you teach it the opposite.

Each wave waits for the one before it to finish. The last one files the receipt. A wave with nobody due is skipped and noted.

Turn Send in waves off and the whole daily audience goes out as one wave instead. There is rarely a good reason to.

Win-Back​

Some of your contacts used to open your email and stopped. Win-back is the agent's attempt to bring them back, and it is deliberately small.

Here is how it works:

• A budget, not a blast. Each day the agent spends a small share of that day's volume on win-backs. The default is 10%.
• A cap inside the cap. Dormant contacts may only take part of that budget. The default is 5%.
• Proven emails only. The agent does not write anything. It picks emails you already sent that earned real clicks and replies, and re-sends the best of them.
• A waiting period. A cooling contact is not tried again for 7 days by default; a dormant one waits 14.
• A limit on trying. After 3 attempts, a dormant contact who still never responds is rested (moved to Sunset) and left alone.
• At most three different emails a day. More than a handful stops being one decision you can review.

Nobody gets a win-back email they have already received. Within each group, the agent starts with the people who engaged most recently — they are the likeliest to come back.

How the "best" email is chosen​

The agent ranks your past emails on how many clicks and replies they earned per send. Clicks are counted within 48 hours of each person's own copy arriving, and replies within 7 days, so an email from last year and one from last week compare fairly. Replies count double — a reply is a much stronger signal than a click.

The ranking is deliberately pessimistic about small samples. An email needs at least 200 sends to the comparison group before it is ranked at all, and one sent to 40 people that did well cannot outrank one sent to 40,000 that did nearly as well.

If no email has enough history yet, the agent simply does no win-back that day and says so: "No win-back today: no evergreen email has enough proven history to re-send yet."

Warmup​

If you send a lot of email from a domain that has never sent much, providers get suspicious. Warmup is the fix: start small, prove you are fine, then grow.

The agent turns warmup on by itself when your account does not yet have much engagement history to lean on.

How the ladder works:

1. Day one starts at a small volume. The default is 500 emails.
2. The level holds for a couple of days, so there is something to measure.
3. Every day the agent takes three readings from the last 72 hours of your mail.
4. All three healthy → the volume doubles. Any reading bad → it holds or halves instead.

The three readings, called gates:

Reading	Default limit	What it tells you
Bounce rate	at most 2%	How many addresses do not exist or refused the mail. High means a dirty list.
Complaint rate	at most 0.1%	How many people hit "report spam". Even tiny numbers matter here.
Open rate	at least 5%	Whether anyone is reading. Low means you are already being filtered.

What happens on a bad reading:

• Bounces or complaints over the limit — the daily volume is cut in half. It never drops below your first-day volume.
• Open rate below the target — the volume holds where it is instead of doubling.
• Not enough mail in the window to judge — the volume holds. The agent will not ramp on guesswork.
• Sending paused by the bounce-rate safety brake — the volume is halved too.

A day you skipped teaches the ladder nothing. It picks up from the last day that actually sent.

Every reading and every decision goes into the receipt, so "why did we only send 500 today?" always has a written answer.

Send-Time Optimization​

Two people on the same list read email at different times. One clears their inbox at 7am, the other at 9pm.

With send-time optimization on, the mail system holds each copy and delivers it at the hour that person usually engages. The default holding window is 24 hours, and you can extend it.

• A shorter window reports results sooner and keeps the send feeling like "today".
• A longer window gives more room to find a good hour for someone who only ever reads on Mondays.

One deliberate exception: the first wave of a normal send is never held. Delaying your Hot contacts by up to three days would throw away the early engagement that the wave order exists to create. Every later wave, and every win-back wave, is optimized.

The Overview Tab​

Everything the agent is doing right now fits on one screen.

Status​

At the top: which mode you are in, stated in a sentence — "Your email agent is running on auto" — with the mode switch beside it and the next check-in time below.

If the agent is off, you get a short pitch instead, with a Turn on the email agent button that starts you in Suggest mode.

Needs your approval​

Only appears in Suggest mode, when a plan is waiting. It says in plain words what the agent wants to do:

Send "August product update" to 4,120 contacts in your selected tiers today. Plus win back 380 contacts using "The 5-minute setup guide".

Three controls:

• Approve and send — go ahead.
• Skip today — nothing goes out. The queue is untouched and tomorrow is a fresh decision.
• See details — open the full plan (see below).

Up next​

The campaign at the top of the queue, with how many people it will reach — "Will go to about 12,400 contacts in your selected tiers" — or its progress if it has already started.

If the queue is empty, this becomes an invitation to write something.

Recent activity​

The last week of days, one line each, in plain English:

• "Sent 'August product update' to 4,120 contacts (16,540 of about 50,000 so far)."
• "Won back 380 contacts."
• "Rested 42 unresponsive contacts."
• "Warmup: ladder level 4,000."
• "Nothing was ready to send."

Click any line to open that day's full receipt.

Reading a Plan​

See details opens what the agent intends to do, before it does it.

• Audience — three numbers: how many are In the lists, how many are In today's tiers, and how many are Planned to send. Plus the hours it will send in.
• Send order — the waves in order, with how many are planned for each.
• Win-back — the budget in emails and as a percentage, how many cooling and dormant contacts are due, and which past emails will be reused.
• Warmup — whether you are warming up or at full volume, today's ladder level and cap, the three gate readings, and the decision they produced.

Every section carries a sentence explaining itself, for example: "4,120 contacts in the hot, engaged, new tiers, capped to 4,000 by today's warmup level."

Reading a Receipt​

Receipts live on the Activity tab. One row per day, oldest at the bottom, with a Day, Status, Sent, and Emails reused column. Expand a row to see the whole thing.

Inside each receipt:

• Sent / Failed / Skipped / Contacts rested — four counts summarising the day.
• Sends by tier — how many emails went to Hot, Engaged, and so on.
• Waves — each wave, how many were planned, how many actually went. A wave with nobody due reads "Nobody was due for this wave."
• Emails reused — which past emails were used for win-backs, and to how many people.
• Emails passed over — any queued campaign that was skipped, with the reason.
• Deliverability readings — the day's bounce, complaint, and open figures.

---

Next Steps​

• Email Agent Settings — every lever and its default
• Reply actions — let the agent handle what comes back
• Campaigns — hand a campaign to the queue