FREE

Deliverability Monitoring

You cannot see the inbox. A 250 OK tells you a server accepted the message, not where it landed. This playbook builds the instruments that turn an invisible outcome into a monitorable one.

Facts checked August 2026

The uncomfortable truth under every deliverability dashboard is the measurement gap. Acceptance is not placement. The receiving server says 250 OK and then does whatever it wants: inbox, promotions, spam folder, or a quiet discard. No provider tells you directly where your mail landed. Everything in this playbook is a way of triangulating that answer from signals you can actually observe: your bounce logs, your complaint feeds, your engagement by recipient domain, and the partial telemetry each provider chooses to share.

Each provider's instrument has known blind spots, and honest monitoring starts by writing them down. Google Postmaster Tools covers personal Gmail only, so your traffic to workplace domains is invisible to it. It lags a day or more, goes dark below a privacy volume floor, and its spam rate is measured against inbox-delivered mail, which produces the tool's cruelest trick: as placement collapses and mail routes to spam, the denominator shrinks and the spam rate can print a beautiful 0.0%. An empty or flattering dashboard is not evidence of health. It can be evidence that nothing is reaching an inbox at all. Google has also moved compliance itself into the telemetry: its dashboards and API now return plain-language verdicts on whether one-click unsubscribe is present and whether unsubscribes are actually honored, which means indifference to the rules is now a recorded state, not a private one.

Microsoft is the weakest window of the big three, and in 2026 it got weaker: spam-trap hit data was removed from SNDS entirely, and complaint reports were cut down to header-only form. SNDS still grades your IPs by junk-folder share, but the explicit early warning that a bad list was hitting traps is gone, which moves the burden onto your own hygiene program to catch what Microsoft now punishes silently. Yahoo runs the cleanest feedback loop in the business, keyed to your DKIM signing domain, and offers something genuinely rare: server-side engagement feeds measured from its own interface, beyond the reach of privacy proxies. Almost nobody enrolls. If Yahoo matters to your file, be the sender who did.

Then there are seeds and panels. Seed lists verify infrastructure: authentication passing, content filters clearing, placement at the long tail of providers with no postmaster tools. They cannot predict engagement-based filtering, because synthetic mailboxes have no history with your brand, and a perfect seed score can coexist with your real list rotting in the spam folder. Panels measure real placement but run thin and skewed for any specific sender. I have used seed-based reporting for years as part of the standing deliverability routine, and it earns its place there, as a directional instrument inside a regular cadence. What it never gets to be is ground truth.

My own day-one stack, at national retail scale: Google Postmaster Tools watched weekly and daily during any ramp, across the four surfaces that matter (user-reported spam rate, domain reputation, authentication pass rates, IP reputation); SNDS for the Microsoft side, caveats and all; the ESP's own bounce and complaint reporting; periodic runs through an email verification service as part of standing hygiene; and seed reports on a regular schedule. None of these is exotic. The mechanism is that they are enrolled on day one, owned by name, and read on a rhythm, because every one of them is worthless the week you finally think to check it.

The practical placement instrument, though, is not a provider dashboard at all. It is your own engagement data cut by recipient domain. Track clicks and conversions per domain cohort against that cohort's own baseline, and placement shifts announce themselves: when the Gmail cohort's clicks collapse while Yahoo holds steady, you are watching a placement event no dashboard will show you. Single-day moves are noise. Seventy-two-hour trends against a cohort baseline are signal.

Use each instrument for what it measures, and let no single one of them tell you the program is fine.

SIGNALS IT IS WORKING

Every reputation surface has a named owner and a weekly timestamp.

Alerts fire on concurrence, not single metrics, and the last three incidents were detected by the scorecard rather than a revenue chart.

Empty dashboards are recorded honestly, and someone other than its author can read the scorecard cold.

REPLACES

Discovering a placement collapse from the revenue chart three weeks late, then reverse-engineering what happened from memory.

0 of 4 steps complete

STEP 01Enroll in everything, and record the blind spots

Postmaster Tools with API access, so compliance verdicts and stats can feed your own monitoring rather than a browser tab. SNDS with its IP-level framing and its 2026 limitations written into the runbook. Yahoo's complaint feedback loop enrolled against the exact DKIM d= domain your mail signs with, plus every other feedback loop your volume justifies. Ask about Yahoo's performance feeds if that cohort is material. For each instrument, record at setup what it cannot see, so nobody later mistakes a blind spot for a clean bill.

STEP 02Build the weekly scorecard

One page, per-provider cohort rows, a small set of banded columns: delivery rate, hard bounce rate, complaint rate, unsubscribe trend, and a placement estimate where you have one. Every row gets a verdict: healthy, watch, or act. Two standing rules: empty Postmaster data is recorded as below volume floor, never as zero, and a flattering spam rate during a placement slide is treated as the shrinking-denominator trick until proven otherwise.

| Metric | Healthy | Warning | Critical |
| --- | --- | --- | --- |
| Delivery rate | Above 95% | 90 to 94% | Below 90% |
| Hard bounce rate | Under 0.5% | 0.5 to 1% | Above 1% |
| Complaint rate | Under 0.10% (documented target) | 0.10 to 0.29% | At 0.30% (documented ceiling) |
| Unsubscribe trend | Stable | Sustained rise | Spike alongside complaint movement |
| Placement estimate | Above 85% | 70 to 84% | Below 70% |

STEP 03Watch cohorts, not averages

Blended metrics hide every placement event inside an average. Segment engagement by recipient domain, baseline each cohort against itself, and alert on divergence: one cohort dropping while others hold is a provider-specific event and names your next move. This is also where post-privacy measurement honesty lives: clicks, conversions, and provider-side engagement data are the signals; opens are a machine artifact wherever privacy proxies operate.

STEP 04Design alerts for concurrence, and decide the seed question explicitly

Page-worthy alerts fire on concurrent signals: a cohort engagement drop plus reputation-citing bounce strings plus complaint movement is an incident; any one alone is a watch item. Then make the seed and panel decision deliberately instead of by default: what cadence, which providers, who reads the report, and what it is allowed to prove. Write down that a seed pass does not overrule a cohort collapse. When the concurrent alert fires, you are no longer monitoring; hand off to the incident-response playbook.