Why Your Open Rate Went Up and It Means Nothing
Your open rate went up eleven points last quarter. Nobody touched the subject lines. Nobody moved the send time. We see this pattern constantly on our own sending fleet, and the explanation is almost never that your audience suddenly started caring. Open rate stopped being a behavioural metric years ago. Most reporting dashboards never got the memo. Knowing what an open actually records - and what it quietly leaves out - is the difference between a campaign report you can act on and a chart that flatters you into repeating a mistake.
What Actually Happens When an Open Is Counted
An open is one thing: a request for a tracking pixel. A 1×1 image loads, your platform logs a hit, that hit becomes a row in a report. The recipient may never have looked at the message.
Plenty of systems fire that pixel with no human anywhere near it. Mail Privacy Protection prefetches images on Apple Mail whether or not the message gets read. Gmail proxies images through its own cache. Security gateways and link scanners request every asset in a message before it lands in the inbox, so the pixel fires during inspection.
The reverse happens too. Plain-text readers and clients with images off register nothing, even when someone reads the whole thing and acts on it.
So what have you got? A number that measures image loading behaviour across a mixed population of software. Human attention is not in there.
The Three Boring Reasons Your Open Rate Climbed
When a client asks us why their opens jumped, the answer is usually one of three unglamorous mechanics.
- Audience shift. More Apple Mail addresses landed in the segment. Prefetching does the rest. No behaviour changed, only the mix of mailbox providers.
- List shrinkage. You suppressed inactive contacts. Denominator dropped, percentage rose, total reach fell. A better ratio over a smaller audience is not growth.
- Filtering change. A corporate gateway switched prefetching on. Overnight, a whole domain looks engaged.
And then the send-time experiment and the subject-line rewrite collect credit for movement they had nothing to do with. The team draws a conclusion, bakes it into the next campaign, and the real driver stays invisible. Sending more often rarely fixes this either - more emails and more engagement , and volume just adds noise to an already unreliable number.
Tip: test the hypothesis before you celebrate. Segment the same campaign by mailbox provider and compare open rate across them. If the lift lives in one provider, it is infrastructure, not copy.
What We Look At Instead on Our Own Sending Fleet
We run our own outbound servers, so delivery outcome comes first in every report. Accepted, deferred, hard bounce, soft bounce - plus the SMTP response text behind each one, kept verbatim.
Bounce classification carries more information than bounce count. A 5.1.1 means the mailbox does not exist and the address should go. A 5.7.1 means the receiving system refused you on policy grounds, which is a reputation problem, not a list hygiene problem. Mash both into one “bounces” column and you have thrown away the distinction that tells you what to fix.
Complaint rate comes from feedback loops, watched per receiving domain rather than blended into one figure. A blended complaint rate hides the domain that is about to block you.
Then unique clicks, and what happens after the click, with link-scanner user agents filtered out. Reply rate and unsubscribe rate close the set. Both are hard to fake. Both need an actual person. We write about this side of measurement regularly in our campaign analytics articles.
Engagement Signals the Big Filters Actually Use
Receiving systems never see your pixel. They see their own users acting inside their own client, which is a completely different dataset.
Positive signals: moving a message out of spam, replying, adding the sender to contacts, time spent reading before closing. Negative: deleting without opening, and marking as spam. Those push reputation in opposite directions and not one of them shows up in your platform. If you want to understand how subscribers really behave with your mail, this is the dataset that matters.
Which is why a rising open rate can sit right next to falling inbox placement. Prefetching inflates one number while user behaviour quietly degrades the other. They measure different things, so they can drift apart forever.
None of it matters without authentication in place first. SPF, DKIM with proper alignment, a DMARC policy, a correct Return-Path and a matching PTR record are the entry ticket. Fail those and your engagement signals never get a chance to count for you.
Building a Reporting Setup That Does Not Lie to You
Five changes turn campaign reporting from decoration into a diagnostic tool.
- Split every report by mailbox provider. One blended number averages away every mechanic worth seeing.
- Tag pixel-only opens separately. An open with no click and no reply belongs in its own bucket.
- Filter known scanner user agents from both opens and clicks, and keep that filter list current.
- Track click-to-delivered, not click-to-open. Click-to-open inherits every distortion sitting in the open denominator.
- Keep a seed panel across the providers your list actually uses, for real inbox placement rather than inferred placement.
Tip: never compare open rates across a period when your list composition changed. Different population, different number, no conclusion.
Tip: treat any open landing within a second of delivery as machine traffic. Humans are slower than that.
Tip: retain raw SMTP logs long enough to explain a drop three weeks later. Our campaign reporting and delivery data is built around exactly that retention window.
Where Consent Fits Into All of This
A list built on real consent produces engagement signals that survive filter scrutiny. People who asked for your mail open it, reply to it, and rarely report it. That is the whole mechanism.
RODO and article 398 of the Prawo komunikacji elektronicznej require consent for commercial email in Poland, B2B included. No exemption for writing to a company address instead of a personal one. None.
Proof of consent means a stored timestamp, the source, and the scope of what was agreed to. A vague appeal to legitimate interest is not proof, and it will not help you when a provider asks why your complaint rate looks the way it does.
Purchased and scraped lists announce themselves in the data long before a regulator does: complaint spikes, spam-trap hits, sudden deferrals from one provider. Compliance and dostarczalność are the same work, done once.
What We Have, What We Do Not Have Yet
We operate our own sending fleet with per-IP and per-domain reputation control. Warmup is handled in-house, on our schedule, not outsourced to a shared pool where someone else’s list quality becomes your problem.
We have been blocklisted at large filters and worked our way back off. That process is slower and far more procedural than vendor marketing suggests: find the exact source, fix it, document the fix, request delisting, then sit there while traffic patterns rebuild trust. No shortcut. We do not sell one either.
Some of what this article describes is industry practice rather than a button in our panel. Provider-split reporting and scanner filtering are on the roadmap, not shipped. The features that are live today are documented as they actually work. We would rather publish a lower, truthful open rate than a flattering one, because the flattering number costs you a decision later.
The Metric to Judge a Campaign By
Open rate is a diagnostic hint. Read next to delivery outcomes and complaint data, it can point you toward a problem. Read alone, it points nowhere.
Start from intent instead. Decide what the campaign was supposed to cause - a reply, a booking, a renewal, a download - then measure that thing directly, and treat everything upstream as plumbing.
One practical next step: rebuild last quarter’s reporting split by mailbox provider. Most teams find that a single provider drove the change they had credited to creative work.
Measure outcomes. Keep the authentication clean. Keep the consent record. The rest is instrumentation, and instrumentation that flatters you is worse than none at all.
