What you'll be able to do after this guide: build the one agent allowed to create outreach tasks, so every prospect who can't be reached another way gets exactly one email task, at the right time, with the draft inside.
Email is the slowest channel for a small team: each one needs reading, checking and sending by a person. So at Fernhill the social channel goes first whenever a prospect has a profile, and email is the fallback. The email task agent's job is to spot who has stalled and hand a person a ready-to-send email.
The agent creates a task in three situations:
| Situation | How the agent spots it | When it acts |
|---|---|---|
| A. Request not accepted | Connection status is Request Sent, and connection_requested_at is 7 or more days ago | After 7 days |
| B. Routed but never actioned | Contact has a profile URL, connection status is blank, research finished 7 or more days ago | After 7 days |
| C. Can't be reached on the first channel | No contact at the company has a profile URL, or the segment isn't handled by the first-touch stage (partners, for example) | Straight away |
These are not stalls: a contact who has accepted (the message stage owns them), any company not at Review, and any company whose research isn't finished.
At 08:35 the agent finds:
Merve Ltd: request sent 9 days ago, still pending. Situation A. No open task, no earlier fallback, confirmed email. Task created.
Daves Farmshop: request sent 8 days ago. There's already an open "Contact" task on the company. Skipped, and listed as already covered.
Dell: a partner, so not handled by the first-touch stage. Situation C. Task created with the partnership title.
Anthropic: request sent 10 days ago, but no email address on record. No task. Listed under "Stalled with no email: needs an address or a decision".
Leeds IT: connection_requested_at is blank. Skipped and named, because a missing clock is never guessed.
Open task guard. If the company already has an open outreach task, skip it. Housekeeping tasks such as "Classify segment" don't count.
Once-ever guard. If this contact has ever had an email fallback (the agent writes a marker note each time), skip it. One email fallback per contact, full stop.
One task per company, for the primary contact.
No email address, no task. List these separately for a person to resolve. They don't count towards the cap.
Oldest first, across all three situations combined.
Daily cap, for example 10.
Checking open tasks is slow. On its first run our agent spent its whole time checking tasks for only a fraction of the stalled companies and could only report a guess at the total. The fix:
Do a fast count using field filters only, and report the true total.
Run the slower task checks on only the oldest 20 or so, enough to fill the cap with room for skips.
The summary then says: how many are stalled in total, how many tasks were created, how many remain, and how long the oldest has been waiting.
Title: "Contact [Name] at [Company]", or "Partnership intro: [Name] at [Company]" for partners.
Assigned to the person who sends, due in 48 hours, linked to both the company and the person.
First line: why the task exists, for example "Connection request sent 9 days ago, not accepted. Falling back to email."
The draft, copied from the routing note written at sourcing time. If no draft exists, the agent writes one and says so.
One line each for the buying signal, revenue (if evidenced), backup contacts, and where the full research is.
Last line: "Once sent or skipped, update the prospect status on this record."
After creating the task, the agent writes a marker note on the company recording the contact, the situation, the date the wait started, and the task. That note is what the once-ever guard checks.
The email task agent creates tasks and marker notes. Nothing else. It doesn't change connection status, touch dates or prospect status. A task isn't a conversation; the company becomes Contacted only when a person actually sends the email, which is why every task ends with the reminder to update the prospect status.
Run it after the acceptance check on weekday mornings, so anyone who accepted overnight is already marked Connected and doesn't get an unnecessary email. Don't run it at weekends if the acceptance check doesn't run at weekends, or weekend acceptances will look like stalls.
If the cap is reached every day and the remainder grows for two weeks, your sourcing is producing more prospects than one person can email. Either raise the cap or reduce sourcing volume. Both are fine; a growing invisible backlog isn't.
Next: Write First Emails That Get Replies.
