Split the number before acting on it. Permanent failures mean addresses that no longer exist and must be deleted. Temporary ones mean full mailboxes, busy servers and greylisting, and those addresses are usually fine. Averaging the two together hides which problem you have.
Providers read the permanent share as a proxy for practice. Somebody mailing addresses that returned a hard failure last month is telling Gmail something about how they collect and maintain data, and the inference is not flattering.
The dangerous moment is the first send to old data. A file that sat untouched for eighteen months can bounce a fifth of its addresses in one campaign, and that arrives as a single spike rather than a gentle trend. Verify before that send, not after it.
Watch the figure by source rather than in aggregate. An overall 3% looks acceptable while one lead vendor inside it bounces at 12%, and that vendor is both your problem and your remedy. The average is what hides it.
There is a floor below which no amount of cleaning helps. People leave jobs between your verification run and your send, so a small permanent bounce rate is evidence that your list is made of real, moving humans.
Working the number out from one send report
Say your report shows 24,000 sent and 720 bounced. Divide bounces by sent, not by delivered, and you get 3.0%. That figure alone tells you little, so split it. Suppose 410 were hard bounces and 310 were soft. Your hard-bounce rate is 1.7% and your soft rate is 1.3%.
Now read the codes on a few of the hard ones. A 5.1.1 means the mailbox doesn't exist, and that's a list problem. With 5.7.1 you're looking at a policy block, which points at your sending setup and not at the address. On the soft side, 4.2.2 is a full mailbox.
In this example the list problem is 1.7%, which is under the 2% line. The headline 3.0% looked worse than the list really is, and the fix for the rest lies with your sending, where cleaning won't help.
Three cases where the rate misleads you
Small sends are the first. Three bounces out of 60 messages is 5%, and it means nothing. Don't react to a rate until a few thousand messages sit behind it.
Suppression can flatter you as well. Your platform stops mailing addresses that bounced before, so each campaign's rate covers only the survivors. A steady 0.4% can sit on top of a list where a fifth of all contacts are already suppressed. Look at the suppressed count next to the rate, because together they describe the list.
Mixed streams are the third. Password resets and receipts go to people who typed their address a minute ago, so they almost never bounce. If those share a report with marketing sends to year-old data, the blend will look healthy while the marketing stream is in trouble.
A single campaign can also swing on one bad import. Rolling 30-day figures, split by stream, are the version worth putting on a dashboard.
Related questions
What bounce rate triggers filtering?
Above 5% providers start treating the list as unmaintained. Above 10% some stop accepting your mail entirely.
Do soft bounces count toward the problem?
They are tracked separately and weighted less. A soft bounce that repeats across a retry window is converted to permanent by most platforms.
Can a bounce rate be too low?
A rate of zero across a large send usually means suppression is hiding failures rather than that every address is perfect.
How much does verification reduce bounces?
It removes the addresses that could be proven dead. The catch-all portion stays unproven, so the improvement depends on your domain mix.