Email deliverability metrics are only useful if they answer the question you actually have: are my messages reaching people, and if not, why? Many dashboards show a large delivered number, a comforting open rate and little else. That can hide a spam-folder problem for weeks. This guide explains which numbers describe deliverability honestly, which ones mislead, and how to put them together in a view that points to action.
Four layers between sending and a result
It helps to think of a message passing through four layers. Each has its own metrics, and a failure at one layer is invisible to metrics from the layers below it.
| Layer | Question | Main metrics |
|---|---|---|
| 1. Accepted | Did the receiving server take the message? | Accepted, bounced (hard and soft), deferred, blocked |
| 2. Placed | Where did it land? | Inbox, tabs, spam, other; reputation indicators |
| 3. Seen | Did the person notice it? | Opens (weak), replies, time to read |
| 4. Acted on | Did they do what you wanted? | Clicks, sign-ins, purchases, unsubscribes, complaints |
Your sending platform reports most of layer 1 directly, because it sees the replies from receiving servers. Layer 2 is harder, since receivers rarely say where they put a message. You infer it from the other layers and from tools such as Postmaster Tools.
Layer 1: acceptance, bounces and deferrals
What to track
- Accepted rate. Messages the receiving server accepted, divided by messages sent. Often called delivery rate. It is necessary and not sufficient.
- Hard bounce rate. Permanent failures, such as an address that does not exist. Each address should bounce hard only once, because it should be suppressed afterwards. See SMTP bounce codes explained.
- Soft bounce and deferral rate. Temporary failures. A few are normal. A rise toward one provider can mean throttling.
- Blocks. Rejections that cite reputation, blocklists or policy. These need a response, not a retry. Our guide to blocklist removal covers next steps.
- Delivery latency. The time from send to acceptance, which matters most for one-time codes and alerts.
How to read them
| Pattern | Possible meaning | Check |
|---|---|---|
| Hard bounces spike after a campaign | Old or poor-quality list | List source, age and validation |
| Deferrals rise at one provider only | Throttling or a reputation dip at that provider | Volume ramp, complaint data, provider postmaster tools |
| Blocks cite a blocklist | You or your IP address is listed | Which list, why, and the delisting process |
| Latency grows for all mail | Queueing on your side or the platform side | Rate limits, retries and queue age |
| Authentication failures in bounce text | SPF, DKIM or DMARC problem | Domain setup and alignment |
Layer 2: placement, the hard part
Receiving servers accept a message and then decide where to put it. They almost never tell you. Three sources of evidence help, and none is perfect.
- Postmaster Tools and similar dashboards. Google Postmaster Tools reports domain and IP reputation categories, a user-reported spam rate and authentication results for Gmail. Our guide to Google Postmaster Tools setup shows how to read it. Other providers offer their own programmes, with different coverage.
- Seed tests. You send to a set of test mailboxes and see where they land. Useful as a spot check. Weak as a measurement, because test accounts have no history with your mail and do not behave as real people.
- Inference from engagement. If clicks and replies from one provider fall sharply while acceptance stays high, the mail may be going to spam there.
Treat placement tools as indicators. None of these gives a precise inbox percentage. Combine them, look for agreement between them and avoid reacting to a single noisy number.
Layers 3 and 4: engagement, with caution
Why open rate misleads
An open is usually recorded when a tiny image in the message is loaded. Some mail apps now fetch images automatically, on behalf of the user, whether or not the person reads the message. That inflates opens and makes them unreliable, particularly for the share of your audience on those apps. Opens can still show a rough direction when compared with your own past results, but they should not be used to prove inbox placement or to decide who is inactive.
Better signals
| Signal | Why it is better | Limit |
|---|---|---|
| Clicks | Require a deliberate action | Security scanners can click links automatically; filter known scanner behaviour where you can |
| Replies | Strong sign of a real reader and good for reputation | Only suits mail where replies are expected |
| Conversions and sign-ins | Tie mail to the result you care about | Need tracking set up on your side |
| Unsubscribes | Direct, honest feedback | Rise with volume; read as a rate |
| Complaints | The signal providers weigh most | Often hidden; use Postmaster Tools and feedback loops |
For a closer look at complaints, including thresholds and feedback loops, read spam complaint rate and feedback loops.
Slice the data
Averages hide problems. The same numbers, cut three ways, tell you much more.
- By stream. Receipts, password resets, newsletters and promotions behave differently. Report them separately. See separating transactional and marketing streams.
- By mailbox provider. Gmail, Yahoo, Microsoft and the rest each have their own filters. Show at least the top few by volume.
- By campaign, list source and sending domain. A single bad import or campaign explains many spikes.
- By time. Look at days and weeks, not one send. Small volumes produce noisy percentages.
A simple dashboard that works
| Panel | Shows | Alert idea |
|---|---|---|
| Acceptance and bounces | Accepted, hard bounce and deferral rate per stream | Hard bounce rate rises well above your own norm |
| Provider view | The same, split by the top providers | One provider diverges from the others |
| Complaints | Postmaster spam rate and feedback loop counts | Approaching the 0.10% guidance |
| Authentication | SPF, DKIM and DMARC pass rates from reports | Any fall in pass rate or alignment |
| Engagement | Clicks, replies, conversions, unsubscribes | A sharp fall with no change in volume |
| Latency and queue | Time to accept; age of oldest queued message | Latency above what the message type tolerates |
Choose your own alert levels from your history. We do not publish universal thresholds, because normal values depend on your list, content and audience. The Google spam rate figures are the exception, since Google states them in its sender guidelines.
A worked example: delivered is 99 percent, yet sales are down
A shop sees order-confirmation delivery at 99 percent, but support receives more messages asking where receipts are. The team splits the data by provider. Acceptance is high everywhere, but clicks on the confirmation link fall sharply for one large provider, and that provider shows rising deferrals earlier in the month. A header check on a test message reveals that DKIM now fails for mail from a new subdomain that was added without a record, as described in reading email headers.
They publish the missing DKIM record, confirm that authentication passes and watch the provider view. Within days, deferrals return to normal and link clicks recover. The overall delivery number never moved, which is why it did not warn them. The example is illustrative; the pattern of a healthy average and a broken slice is typical.
Common mistakes
- Equating delivered with inbox. Accepted is only the first layer.
- Averaging all mail together. A good stream can mask a bad one.
- Chasing opens. They can rise without any more people reading.
- Ignoring small senders' noise. With low volume, a single complaint swings the percentage. Look at counts as well as rates.
- Changing many things at once. Then you cannot tell which change caused the result.
- Skipping authentication checks. Start from a verified domain, as in authenticating your sending domain.
Want help reading your delivery data? Share a date range and the stream you are worried about and we will help you interpret it.
Further reading and sources
Frequently asked questions
What is the difference between delivery rate and inbox placement?
Delivery rate counts messages that the receiving server accepted. Inbox placement says where an accepted message ended up: the inbox, a tab, the spam folder or elsewhere. A message can be delivered and still land in spam, so a high delivery rate alone does not prove good deliverability.
What bounce rate is acceptable?
There is no single number that fits every sender. Watch your own trend and investigate sudden rises. Hard bounces should be rare on a clean list and should be suppressed after the first one, so a high or rising hard bounce rate points to list quality problems.
Are open rates still useful?
Only with care. Some mail apps load images automatically, which can record an open the person never made. Opens can still show a rough trend over time, but they should not be your main measure of deliverability or success.
How do I measure spam complaints at Gmail?
Use Google Postmaster Tools, which shows a user-reported spam rate for domains that send enough mail to Gmail. Google asks senders to keep that rate below 0.10 percent and not to reach 0.30 percent.
What is a deferral?
A deferral is a temporary refusal by the receiving server, usually an SMTP reply starting with 4. The sender is expected to try again later. A rising number of deferrals to one provider is an early sign of throttling or reputation trouble.
Do seed tests show my real inbox placement?
They give a rough picture. Seed accounts have no history with your mail, so they do not react as real recipients do. Use them as a spot check and compare against real signals such as clicks, replies and complaints.
How often should I review deliverability metrics?
Check alerts daily or as they fire, review dashboards weekly, and do a deeper review each month or after any change to your sending setup, list or content.
Why should I look at metrics by mailbox provider?
Each provider applies its own filters. A problem at one provider can be invisible in your overall average. Per-provider views show where to look first.
Which metrics matter most for transactional mail?
Acceptance, delivery speed, bounces and deferrals, and whether time-sensitive messages such as sign-in codes arrive in a few seconds. Complaints are usually low, but any rise deserves attention.