Automate the known. Escalate the consequential.
Known events with approved actions run on their own: renewals, auth drift, bounces, opt-outs, scheduling. Ambiguous ones arrive as a brief a person decides.
Jared Hawkins-Young · Founder, Lost Star TechnologyOctober 8, 20265 min readPart 13 of 14Key takeaways
- A known event with an approved action runs on its own: domain renewals, authentication drift, sender health guardrails, bounces, opt-outs, CRM updates and routine scheduling.
- An ambiguous event goes to a person with an evidence packet: rate drops, rising complaints, an underperforming cell, a new data source, and any sensitive or unclear reply.
- Every escalation arrives as the same brief: what moved, what changed, likely causes ranked by evidence, and a recommended next step.
- Each decision a person makes is logged as a rule, a test result or a revised guardrail, so the list of known events grows while relationship judgment stays with people.
Automate an outbound task when the event that triggers it is known and the right action was approved in advance; escalate it when the event could have more than one cause or carries risk to a relationship. In practice, renewals, authentication checks, bounces, opt-outs and routine scheduling run on their own, while rate drops, complaints, new data sources and sensitive replies reach you as a short brief with a recommendation.
How do you decide what to automate in outbound?
Ask two questions about every event the system watches.
- Is the event unambiguous? A hard bounce means one thing. A falling reply rate can mean seven, as part 12 lays out.
- Is the action approved in advance, and safe to take without a second look? Suppressing an address that opted out is required and reversible. Retiring a campaign is neither.
Two yeses and the task runs on its own, with a log entry. Any no, and the system's job changes from acting to assembling evidence for the person who decides.
- 01Domain renewal window opensRenew the approved domain, check DNS and billing, log it.
- 02Authentication driftsCompare SPF, DKIM and DMARC with the approved records; restore them or pause the sender.
- 03A sender fails a hard health guardrailPause it, move the next batch to a warmed standby, start the replacement.
- 04Hard bounce or invalid contactSuppress the address, re-verify or enrich, write the status to the CRM.
- 05Opt-out or do-not-contact replySuppress across every active sequence, log it, stop further steps.
- 06Qualified reply or booked meetingUpdate the CRM with cell, source, signal and thread; route to the rep.
- 07Routine scheduling or an approved FAQAnswer only from approved material, offer the rep's open times, book it.
- 01Delivery or bounces leave the control bandBrief compares senders, providers, authentication, volume and list checks. You choose the fix.
- 02Reply or positive-reply rate fallsBrief separates source, cell, signal age, timing and component changes. You choose the next test.
- 03Complaints or opt-outs riseBrief names the cells and message patterns involved. You pause, revise or retire the play.
- 04One cell underperformsBrief compares it with its own baseline and similar cells. You decide: fit, context or message.
- 05A new data source changes a cellBrief measures verification, fit, duplicates and response. You approve or reject it for scale.
- 06Sensitive, pricing, complex or unclear replyThread and context kept, routing proposed. The rep owns the answer.
What should run on its own?
Opt-outs are the clearest case. The event is unambiguous, the action is required, and delay only creates risk. Google's sender guidelines FAQ recommends fulfilling unsubscribe requests within 48 hours, Yahoo requires bulk senders to honor them within 2 days, and the CAN-SPAM Act sets a legal limit of 10 business days. A rep working through an inbox on Friday will miss the tightest of those. Suppression across every active sequence, the moment the reply lands, will not.
Sender infrastructure is the second group. A domain that lapses takes its sending reputation with it. An SPF record that breaks because someone added a new tool to DNS can get mail rejected outright: since May 2025, Microsoft rejects mail from high-volume domains that fail authentication. The approved action is narrow: compare the records with the known-good version, restore them or pause the sender, and log what happened. Pausing is always allowed, because it is reversible and it protects the domain.
Bounces, CRM updates for a booked meeting, and routine scheduling answered from approved material round out the list. Each has one meaning and one right response, written down before the first send. The CRM update matters more than it looks: when a meeting is booked, the record gets the cell, the source, the signal and the thread. If the CRM never learns which cell and signal produced a meeting, the portfolio view from part 8 has nothing to roll up.
What needs a person, and why?
Everything with more than one plausible cause or real relationship risk. A rate drop could be data, delivery, timing or copy. A complaint spike could be one bad cell or one bad sentence, and the stakes are high: Google's FAQ says that from June 2024, bulk senders with a user-reported spam rate above 0.3% became ineligible for mitigation. A cell that underperforms might be the wrong fit or the wrong moment. A new data source might be a gift or a reputation problem. And a reply that is sensitive, about pricing, or simply unclear belongs to the rep, as part 14 describes.
The outreach email itself stays on the person's side of the line in this model: an agent drafts it, a person approves it. Hunter's State of Email Outreach 2026, built on 31 million emails sent in 2025, found manually edited emails drew a 5.2% reply rate against 4.4% for fully automated ones. Approval is not a tax on the system. It is part of why the email works.
What should an escalation look like?
Every escalation arrives as the same brief, so you can decide in minutes rather than reconstruct the problem from dashboards. Here is one, with illustrative numbers, for the drop traced in part 12.
| Part of the brief | Content |
|---|---|
| What moved | Positive replies in the logistics software, 51 to 200, head of sales cell fell from 2.4% to 1.1% over two weeks. |
| What changed | A new contact source entered the cell on Tuesday. Its hard-bounce rate is five times the old source's. No message version changed. |
| Likely causes, ranked | 1. Data quality in the new source. 2. Quarter-end timing, which overlaps the drop. 3. Message: ruled out, no change. |
| Recommended next step | Quarantine the new source, re-verify it, and run the cell on the old source for one week as the control. |
| Your decision | Approve, change or reject. The choice is logged as a rule for the next time this pattern appears. |
Routine work moves immediately. Human judgment enters with context, evidence, and a defined decision to make.
How does the automated list grow?
Every decision you make is logged as a rule, a test result or a revised guardrail. When you approve the same response to the same pattern a few times, such as quarantining a contact source whose bounce rate crosses a set line, it can move to the left column with that exact action and threshold. What never moves is relationship judgment, strategy, and anything that reaches a buyer without your approval. After the meeting, part 13 draws the same line for the work that follows the meeting.
Who this is not for. If you send a few dozen emails a week by hand from your own inbox, most of the left column is a checklist, not a system. Write it down anyway: the renewal date, the DNS records, the suppression list, who answers which kind of reply. The automation can come later. The boundary should exist from the first send.
Next in the series
Part 14, Turn qualified replies into the next right action, applies the boundary to the most visible moment in outbound: a buyer writing back.
Common questions
- Should an AI agent pause a sending domain on its own?
- Yes, when a hard guardrail fails, such as authentication breaking or bounces crossing a threshold you set. Pausing is reversible and protects the domain. Choosing the fix, retiring a domain or changing a campaign is a decision for a person, made from the evidence brief.
- How fast do cold email opt-outs have to be processed?
- Google recommends fulfilling unsubscribe requests within 48 hours, Yahoo requires bulk senders to honor them within 2 days, and the CAN-SPAM Act sets a legal limit of 10 business days. Automatic suppression across every active sequence is the reliable way to stay inside the tightest of those.
- What should never be automated in outbound?
- Decisions with more than one plausible cause or real relationship risk: changing strategy after a rate drop, responding to complaints, approving a new data source, and answering sensitive, pricing, legal or unclear replies. The system prepares the evidence; a person decides.
Sources
- Google, Email sender guidelines FAQ (unsubscribes within 48 hours; spam rate above 0.3% ineligible for mitigation from June 2024)
- Yahoo Sender Hub, Sender best practices (honor unsubscribes within 2 days)
- FTC, CAN-SPAM Act: a compliance guide for business (opt-outs within 10 business days)
- Microsoft, Outlook's new requirements for high-volume senders (enforcement May 5, 2025)
- Hunter, The State of Email Outreach 2026 (31 million emails sent by Hunter users in 2025)
Written by Jared Hawkins-Young from the system he builds and runs. Diagrams are drawn from the same specs as the software. Published October 8, 2026. Illustrative numbers are labelled as such; every other figure carries its source.