outbound
The Reply Is the Metric, Not the Open
Open rates stopped measuring anything useful when mail clients began prefetching images. Replies and positive replies are the two numbers left.
Open tracking works by embedding a small image with a unique URL and counting the request. That mechanism has three problems, and together they have taken the open rate from a useful proxy to a number that mostly measures infrastructure.
Why the open rate stopped working
Images are prefetched. Several mail clients fetch remote content on receipt, through a proxy, before a human has looked at anything. The request arrives, the open is counted, and nobody read the message. Depending on the mix of clients in your audience, a large share of recorded opens are machines.
Images are blocked. Other clients and most corporate gateways block remote images by default. Those recipients can read the message carefully, reply, and never be recorded as having opened it.
Security scanners click. Gateways follow links to check them, which inflates click counts too, and produces the characteristic pattern of every link in a message being clicked within a second of delivery.
The result is a metric that is inflated by one population and deflated by another, with the ratio varying by audience. It is not useless; it is unanchored. A change in open rate between two campaigns may be a change in audience client mix rather than a change in anything you did.
What to measure instead
Delivery. Not opens, but whether the message was accepted and not bounced. This is the authentication and reputation question, and it is the one with a clean answer, since RFC 9110 status semantics and SMTP responses give an unambiguous accept or reject.
Reply rate. A human typed something. It cannot be faked by a proxy and it is the first point in the funnel that proves a person saw the message and had a reaction.
Positive reply rate. Replies classified by whether they move anything forward. Ten replies of which one is interested is a different campaign from ten of which six are, and a bare reply rate hides it.
Meetings or jobs booked, attributed back to the sequence. The only number anybody actually wants, and it is the one that takes the most work to attribute honestly.
Complaint rate. The number that decides whether you can keep sending at all.
Replies need a classification, and it can be crude
Classifying replies sounds like work and it collapses into four buckets that one person can apply in a few minutes a day.
Interested. Not now, with a date. Not me, here is the right person. Stop.
The third bucket is the most underused asset in outbound. A reply that routes you to the right person is a success that most teams record as a failure because it is not a yes.
The fourth is an instruction, and it belongs in suppression immediately, not at the end of the week.
Measure the sequence, not the campaign
Averaged over a sequence, every touch looks mediocre. Broken out by touch, the shape is usually sharp and surprising.
Common patterns worth being able to see in your own data: the second touch underperforming because it says nothing new, a channel switch outperforming the touch before and after it, and the final "closing the file" message producing the second-highest reply rate in the sequence.
None of that is visible without per-touch numbers, and per-touch numbers are a reporting decision made when the tool is set up.
Attribute conservatively
The temptation is to credit the sequence with everything that follows it. The discipline that keeps the number useful is to credit it with what it caused.
A prospect who replied to touch three and booked a meeting is attributable. A prospect who never replied, and arrived through a referral two months later, is not, however satisfying it would be to count.
An honest attribution rule, written down and applied consistently, is worth more than a generous one, because the entire point of the number is to decide whether to keep doing this. A number tuned to justify the activity cannot answer the question it exists for.
And know who else is sending as you
One measurement that is not about performance at all.
DMARC aggregate reports, described in RFC 7489, tell you which sources are sending mail claiming to be your domain and whether they pass authentication. Reading them occasionally is the only routine way to discover a vendor nobody told you about, a forgotten server still sending, or somebody spoofing you.
They arrive as XML and are tedious by hand. Any of the parsers will do. Ten minutes a month is enough, and it is the difference between knowing your sending surface and assuming it.