Email delivery rate versus inbox placement

Start with the export definition, not the dashboard colour. Ask which event creates a “delivered” status, which receiving system produced it, whether later failures update it and which messages enter the denominator. If the report does not answer those questions, write the uncertainty beside the number.

SMTP acceptance transfers responsibility to the receiving server; it does not identify the destination folder. Microsoft also documents that its trace label “Delivered” can include messages sent to Junk or quarantine. Those are separate questions about the same message.

This guide proposes a manual reporting convention for making those distinctions explicit. It is a reading aid, not a claim that every reporting system uses these definitions. Map your actual fields before applying the example, and keep their original names available for review.

Use one campaign message addressed to one recipient as the reporting unit. A follow-up is another unit. Retrying the same message to the same recipient does not create another unit. Document how your export identifies that pair so attempt counts do not quietly inflate message counts.

Choose a submission window, timezone and observation cutoff. Include every eligible submitted unit from that window, even if its outcome is unresolved. State any exclusions before calculating the rate. Unsubmitted prospects belong outside this submission cohort; count them separately if they matter to the campaign review.

Name the receiving boundary you intend to measure: the intended recipient’s receiving system. A handoff to an intermediary alone does not establish that boundary. If your records cannot distinguish the two, describe the handoff you can establish and retain the uncertainty.

Save the export version with the report. Otherwise two colleagues can calculate different answers from records that changed between downloads and mistake a timing difference for a calculation error.

For this worksheet, classify the evidence available at the cutoff into the four statuses below. They are mutually exclusive. Keep the underlying event history, but count the message/recipient pair once in the summary. The categories are our reconciliation convention, not a universal vendor formula.

Consider an invented teaching example with 100 submitted units. At the chosen cutoff, 90 are A, four are F, four are P and two are U. These are illustrative figures, not Inbox Leads customer data or a benchmark. The four categories total 100.

Recorded acceptance without later failure is A/N: 90/100 = 90%. Terminal failure is 4%, pending is 4% and unknown is 2%. Subtracting failures alone gives 96%, but that remainder contains six unresolved units. It cannot be relabelled as verified acceptance.

If no messages were submitted, do not calculate a submission rate. If no destinations were observed, report the Inbox share as unavailable, not 0%. Coverage is also unavailable when there are no accepted units eligible for review. Keep the zero counts visible and explain why a percentage cannot be calculated.

Placement evidence needs its own inclusion rule. Identify the exact accepted units inspected, the observation method and the time of inspection. Use mutually exclusive destination categories at that time: Inbox, Junk, quarantine or another identified destination. Quarantine is an administrative holding area, not a mailbox folder.

Decide whether Inbox includes category tabs and record that choice. An observation after a person moved the message does not establish its original location. If initial placement is the question, keep later observations labelled as later observations instead of silently substituting them.

Unobserved means the destination was not established by your chosen method. It does not mean the message disappeared. Eight confirmed Inbox observations do not imply that the remaining 82 accepted units went to spam.

The inspected group may be selected because someone reported a problem, volunteered to help or was easy to contact. Without a justified sampling design, its percentage should remain descriptive of that group. More observations can improve coverage without making the group representative.

A seed test examines selected test accounts under particular conditions. Record the account set, receiving ecosystems, configuration, message version and test time. Its practical value is to reveal an observation worth investigating or checking again; it cannot establish the destination of every prospect’s message.

Keep seed results separate from the real campaign cohort unless they concern exactly the same reporting units. Combining ten test deliveries with 100 campaign submissions changes the population. It does not add missing destination evidence for the 100 campaign messages.

For a comparison, retain the same seed set and method where possible. If the set changes, explain the change before interpreting an improved percentage. Show counts as well as percentages, especially when one account can change the headline substantially.

A useful test note says what was observed, what changed and which question remains. “These selected accounts showed Inbox on this test” is narrower and more useful than a universal placement guarantee. A second favourable test still leaves unobserved campaign recipients unobserved.

An open event or reply does not identify the original destination. A person may find a message through search or move it before responding. No reply leaves several possibilities, including timing, priority, relevance and availability. Neither event should be used to fill an unknown folder field.

Report engagement and business outcomes alongside delivery evidence with their own definitions and denominators. If replies are counted by unique recipient, say so; that is a different unit from message/recipient pairs. A follow-up sequence can contain several messages to one person.

Choose the next action from the unresolved question. Unexplained U records call for better event matching. Pending records call for a later status review. Confirmed Junk observations call for the troubleshooting workflow. Low replies with no destination evidence call for a broader campaign review, not an assumed filtering diagnosis.

Attach this worksheet to the export or review note. Replace every placeholder and leave unavailable evidence marked unknown. Use the same unit and cutoff throughout the reconciliation. Before sharing, check that the status total and the destination subtotal both balance.