Advisor

The Advisor checks your workspace for the things that quietly ruin cold outreach: complaints climbing, a domain missing DMARC, a new mailbox at full volume, a list that is mostly shared inboxes. It says what is wrong, why it matters, and what to do, and where the fix is a setting, it offers to make the change.

Advice appears on the page where the problem is, not in a separate inbox you have to check.

A step's copy problem flags its campaign, since a step has no row. When there is nothing to say, nothing renders.

If one check fires on many things, they collapse into a single card ("19 mailboxes are capped above the safe band") that expands to list each affected row with its own Fix, plus one Fix all.

Detection is not AI. Every check is an explicit rule with an explicit threshold run over your own data, so the same numbers always produce the same result. AI only rewrites the explanation to read more naturally; it cannot create a finding, change its severity, or change what a fix does. The Advisor therefore never spends AI credits.

Sample floors

A check stays quiet until it can be right. Two bounces in thirty sends is not a 7% bounce rate.

Severity

Only the first two badge a nav tab.

Deliverability: complaint rate per mailbox (warns 0.03%, critical 0.10%, matching Google's bulk-sender ceiling); bounce rate (warns 3%, critical 5%, where SES opens a review); spam placement on warmup mail at Google, Microsoft and Yahoo (warns 10%, escalates 20%, critical 50%; other mail hosts are not counted, see how spam placement is judged); SPF and DMARC per sending domain (critical while sending cold, since Google requires them; an unchecked domain is never called failing, and DKIM is never reported missing because its selector is not discoverable from DNS); and whether a mailbox has its own verified tracking domain.

Mailbox configuration: daily cap above the 50/day safe band (gently with clean numbers, firmly without); a mailbox under 30 days old sending above 20/day; a send gap under 5 minutes; unresolved errors on a mailbox campaigns still use; too few mailboxes carrying total volume; an inactive mailbox attached to a running campaign.

Warmup: warmup off while sending cold (critical on a new mailbox); warmup left paused; a ceiling too low to balance cold volume (roughly half the cold cap); reply rate under 20%, making warmup one-way; temporarily throttled sending, reported separately from quarantine or a pool block; none of the pool's warmup mail to the mailbox arriving, which stops the pool sending to it (see unreachable mailboxes). Missing or stale Cloud status is reported separately from a real reputation ban, with steps to refresh it without disconnecting mailboxes.

A throttled mailbox remains in the warmup pool. The advisor explains the automatic reduced sending pace rather than asking you to remove it from campaigns, wait a fixed week or manually re-enter the pool. Keep warmup running and do not increase its cap to bypass the slowdown. For placement throttles, half volume steps down to watch below 15% spam placement, and watch lifts below 7.5%; separate ramp limits still apply. Re-entry probation and other cooldowns follow the reason shown on the mailbox. A throttle alone does not mean the connection has failed or the provider has blocked the mailbox.

The advisor's placement sample covers today and the preceding six UTC dates, matching the Deliverability display and placement-health decisions. After an upgrade, the next advisor evaluation replaces outdated throttle/exclusion guidance automatically. A campaign hold you already applied remains yours to release after checking any other findings.

Campaigns: no follow-ups; reply rate under 1% at volume, escalating under 0.3%; one step replying below a quarter of the campaign's best (compared only against siblings, which share an audience); one-click unsubscribe off; a daily limit above what its mailboxes can send; no active sender; follow-up spacing outside 2 to 14 days; under 7 days of leads left; no A/B test at high volume; a sending window too narrow for the volume; contacts enrolled in several running campaigns.

Copy (conservative, and drafts are skipped): a merge variable that will not resolve, shipping a literal {{token}}; two or more bulk-mail phrases (one is never flagged); over 200 words; a subject over 60 characters; more than two links in a first touch; a shouting subject.

Two more copy findings run only when the operator has configured TypeSafe (TYPESAFE_API_KEY), a calibrated reader model rather than a language model: a step that reads as bulk mail to its recipient, or makes a claim a spam filter objects to (guaranteed results, free money, a prize, pressure to act now); and a step with no clear ask, either asking for nothing or for several things. Each step's subject and body are sent to TypeSafe for this, once per version of the copy; the verdict is cached and an unchanged step is never sent again. At most 40 steps are judged per evaluation, and the rest wait for the next run.

List quality: shared inboxes (info@, sales@) above 15%; consumer domains above 40% of a B2B list; suppressed contacts above 10%; missing first names above 10% when copy personalizes on them; unsubscribed contacts still enrolled.

Opening a recommendation walks three screens: Why (with the measured numbers), What changes (the exact before and after of every field), and Done (with Undo where revertible).

A finding that asks you to stop cold sending from a mailbox, for severe spam placement (50% or more on at least 20 major-provider deliveries), quarantine or a pool block, applies Hold from campaigns rather than switching the mailbox off, and leaves its warmup as it is. A recoverable placement slowdown below that severe band does not offer a campaign hold on its own; complaints, bounces and other confirmed problems are still checked separately. The hold is offered only while campaigns can still pick the mailbox, so a mailbox that is already held, resting or switched off gets no fix, and Undo never releases a hold someone else set.

Nothing is applied until you have seen the second screen. The change runs as you, with your permissions, and is written to the audit log. If you can see a finding but lack permission for its fix, applying it is refused rather than silently escalated.

Findings needing a judgement call have no settings preview. Their second screen is the ordered set of manual steps plus a link to the right screen. A missing DMARC record tells you where the _dmarc host goes and why the first record should start at p=none. Where the remedy is genuinely one sentence, it stays one sentence.

Fix with agent

Findings whose fix edits something Zexr owns (copy, sequence shape, list membership) offer Fix with agent: it reads the current campaign, step, or list, makes the smallest change that resolves the problem, and reports back.

It runs as you, with your permissions, and only gets the tools that finding's category needs. A copy problem gets read access to the campaign and edit access to that step, nothing more. No tool set can send mail, delete a campaign or contact, or touch team, billing, or API keys. Each loop costs one AI credit.

When it finishes you get its account plus the list of tools it actually called, which you can check against the audit log. An agent that concludes nothing needs changing says so, and the finding stays open rather than being quietly closed.

Findings whose fix lives outside Zexr never offer this, because no agent can reach a DNS record at your registrar.

Records to paste

Where the fix is a DNS record, it is on screen with one copy button per field, laid out as type, host, value.

• SPF arrives with your provider's include filled in. An unrecognized provider leaves an obvious placeholder rather than a guess, because a wrong include looks finished and fails silently.
• DMARC is scoped to your domain and starts at p=none, monitoring without affecting delivery. Tighten to p=quarantine once reports show legitimate mail passing.
• DKIM is never called missing, since Zexr cannot see a selector DNS does not list. When a domain is already flagged for SPF or DMARC and DKIM went unverified too, the card adds the host to publish at and names the console screen that generates the value, which is per-domain and cannot be templated honestly.
• Tracking domain gives the full CNAME to add for the tracking host.

Autopilot

Under Settings > Workspace, off by default. It only takes fixes that are bounded, move in the safe direction, and can be undone:

• lowering a cap above the safe band, or one a new mailbox jumped to early
• cutting the cap on a mailbox drawing complaints
• widening a gap that is too short
• matching a campaign's daily limit to its mailboxes' real capacity
• turning on one-click unsubscribe

It never pauses sending, edits copy, starts or resumes warmup, or acts without an undo. Eligible findings are marked Auto on the card whether or not it is on, so the set is visible before you enable it. It acts as the member who enabled it, with their permissions, logged as them, and stops if that person leaves. At most ten fixes per evaluation.

• Snooze hides it for a week or a month; the check keeps running.
• Stop suggesting this dismisses it until the problem clears and recurs later.
• Useful / not useful feeds threshold tuning.

Under Settings > Workspace you can switch the Advisor off, enable autopilot, mute whole categories, or raise the minimum severity. Changing these needs the manage-settings permission and applies workspace-wide.

Findings refresh in the background and when you open a page whose advice went stale. Adding, changing or removing a mailbox re-checks them right away, and removing a mailbox, campaign or step closes the advice about it in the same action. Updates are live, so a teammate applying a fix clears the card on your screen.