Florin Armasu, CEO and founder of Data Innovation and creator of the Sendability platform, has spent fifteen years fixing deliverability problems for enterprise clients including Nestle and Brown-Forman. In a conversation with Sophie Steffen for The Open Rate Club, he shared the kind of war stories that deliverability consultants accumulate the way plumbers accumulate stories about what people flush down toilets – with weary expertise and the faint suggestion that none of this had to happen.
The Five Million Euro Reactivation Campaign
The story begins, as most deliverability disasters do, with good intentions. A major fast-moving consumer goods brand in Barcelona decided to run a reactivation campaign. The email, by all accounts, was well-crafted. The creative agency was Ogilvy. The content was compelling. The strategy, however, was to send it to the entire database – including a massive segment of subscribers who had not engaged with the brand’s emails in months.
The logic was straightforward in the way that most expensive mistakes are straightforward: the email is good, so everyone should see it. The spam filters at Hotmail disagreed with this assessment in the way that spam filters disagree with things – silently, decisively, and without appeal. The brand was blocklisted. Hotmail represented 35% of their database.
Armasu’s estimate of the financial exposure is bracing: a sustained blocklist at Hotmail for one year would have cost the company at least five million euros in lost revenue. They were not sending novelty newsletters. They were driving revenue through a commercial email channel that had, until that morning, been working just fine.
“Deliverability is sometimes treated like a responsibility of – you work with Salesforce, thus you have very good deliverability. And I can tell you that it’s not like that.”
The fix was not a quick DNS change or a settings toggle. It was six years of ongoing deliverability management, because the problems that surface in a blocklist are symptoms of structural issues that take months to address and constant monitoring to prevent from recurring. The relationship between Armasu’s team and the FMCG brand began with a crisis. It continued because the crisis revealed how much was invisible.
The Invisibility Problem
Deliverability failures are not like search ranking drops. If your page falls from position one to position five in Google, you notice. Your traffic chart bends downward. Someone sends a panicked Slack message. A meeting is scheduled. But when your emails start landing in spam at Hotmail while Gmail and Yahoo remain unaffected, nothing obvious happens. Your overall open rate dips slightly. Your overall click rate dips slightly. Nobody panics because the aggregated metrics look merely mediocre rather than catastrophic.
The problem only becomes visible when you break your metrics down by mailbox provider – and most marketers do not do this. They look at campaign-level averages, which is the deliverability equivalent of checking your average body temperature and concluding that everything is fine despite the fact that one of your legs is on fire. Armasu has spoken to companies that had no idea their Outlook performance was degraded, no idea their Gmail reputation was slipping, because nobody had built the dashboard that would have shown them.
The painful lesson, as Armasu frames it after fifteen years of watching this pattern repeat, is that deliverability problems accumulate silently until the day they become acute. You do not get a gradual warning. You get months of quiet erosion followed by a sudden inability to reach a significant portion of your audience. And at that point, the remediation options – changing IPs, changing domains, changing ESPs – take three months minimum, cost serious money, and may not work on the first attempt.
The Bad Neighbor Problem
Shared IP sending is standard practice for small and mid-market senders. The economics are straightforward: a dedicated IP requires enough volume to build and maintain a sending reputation, and if you are sending 10,000 emails a month, you do not have enough volume to warm a dedicated IP. So your ESP puts you on a shared IP pool alongside dozens, hundreds, or in the case Armasu describes, 2,000 other senders.
The problem is that IP reputation is collective. If one sender on the pool buys a list and mails it, the IP reputation suffers for everyone. If another sender triggers a spam trap, the IP reputation suffers for everyone. You can follow every best practice with the diligence of someone who has actually read the M3AAWG guidelines, and your emails still land in spam because sender number 847 out of 2,000 decided that purchasing a list from a broker was a reasonable customer acquisition strategy.
Armasu’s example is a psychologist’s platform – a small, presumably well-intentioned business that was sharing its sending IPs with 2,000 other senders. Their content was fine. Their list was clean. Their deliverability was poor. The diagnosis was not their sending practices. It was everyone else’s.
The solution is straightforward in theory: move to dedicated IPs. In practice, this requires enough sending volume to sustain a reputation, a warming period of weeks to months, and the technical infrastructure to manage IP rotation and throttling across multiple mailbox providers. It is the sort of migration that small businesses find prohibitively complex and mid-market businesses find prohibitively expensive, which is why the bad neighbor problem persists as one of the most common and least discussed causes of deliverability failure.
Apple Mail Privacy Protection: The Metrics Distortion
Since October 2022, Apple Mail Privacy Protection has been systematically destroying the reliability of open rate data. The mechanism is simple: when a user with an Apple device receives an email, Apple’s servers pre-cache the content, triggering a pixel fire that registers as an “open” regardless of whether the recipient actually looked at the message. The email arrives in the inbox, Apple downloads it, and the ESP records an open. The recipient may never have scrolled past the subject line.
The effect on aggregate metrics is dramatic. Open rates on iCloud have ballooned to near 100%. But the distortion is not confined to iCloud – it affects any mailbox provider whose users read email on Apple devices. A Gmail address read on an iPhone triggers an Apple-cached open. A Hotmail address read on an iPad triggers an Apple-cached open. Companies are reporting average open rates of 60% and interpreting this as evidence of engagement excellence rather than what it actually is: Apple’s privacy infrastructure generating phantom signals at industrial scale.
The downstream effects are less immediately visible but more operationally damaging. Send time optimization – the practice of delivering emails when individual recipients are most likely to open them – depends on knowing when people actually open emails. Apple’s pre-caching destroys this signal. Click-through rate – clicks divided by opens – has declined, not because clicks decreased but because the denominator inflated. Engagement-based segmentation becomes unreliable when a significant percentage of your “engaged” segment consists of recipients whose engagement was faked by a proxy server in Cupertino.
The Two-Layer Spam Filter
Modern spam filters, Armasu explains, operate on two layers – and most marketers are only aware of the first. Layer one is the traditional reputation check: does this sender, from this IP, have sufficient reputation to deliver to our users at all? This is the layer that blocklists operate on. It is the layer that IP warming addresses. It is the layer that most deliverability consultants focus on, because it is the layer with the most visible failure modes.
Layer two is newer, more sophisticated, and considerably harder to manage. It asks a per-recipient question: is this specific email, from this specific sender, likely to be relevant to this specific user right now? A sender can pass layer one – their IP is fine, their domain reputation is fine, their authentication is perfect – and still have individual emails routed to spam because the recipient has not engaged with the sender’s emails in months.
This creates a philosophical problem for reactivation campaigns. You want to re-engage inactive subscribers. But the spam filter has learned that these subscribers do not engage with your emails, so it routes your reactivation attempt to spam. The subscriber never sees it. They remain inactive. The spam filter’s assessment is confirmed. You have entered a feedback loop where the filter’s prediction of non-engagement becomes self-fulfilling.
Breaking this loop requires what Armasu calls orchestrated sending – mixing inactive recipients into sends alongside active ones so that the campaign-level engagement signals remain healthy while the individual-level targeting attempts to re-engage dormant subscribers. It is not a setting you toggle. It is a sending strategy that requires volume control, segment mixing, and real-time monitoring, the kind of operational sophistication that most platforms do not support out of the box.
For the full interview breakdown, see our complete Expert Insight with Florin Armasu.
Tools Mentioned in the Interview
The following tools and platforms were referenced during this conversation.
Related Content
Email Marketing Execution Behind the Scenes: Expert Insights with Florin Armasu
Florin Armasu, CEO of Data Innovation and creator of Sendability, unpacks the open source stack, deliverability disasters, and metrics that actually matter after 15 years of sending emails for brands like Nestle and Brown-Forman.
Written by Sophie SteffenEmail Metrics Beyond Open Rates
Open rates are now fiction. Florin Armasu makes the case for active database growth rate as the single most important metric in email marketing -- and explains why almost nobody tracks it.
Written by Sophie SteffenEmail List Hygiene and the Gray List Dilemma
The standard advice on email list cleaning is to purge aggressively. Florin Armasu, with 15 years and enterprise clients behind him, argues that discarding 80% of your database is not best practice -- it is waste.
Written by Sophie Steffen