Over the past four months, Verizon Media Group (VMG)  has been rolling out their new MTA platform, named Atlas. Ostensibly, Atlas is simply their newest generation platform and senders should expect identical behavior, but as everyone knows, even the most pedestrian software changes often bring subtle changes, which is the case here. For reputable senders, the differences appear to be negligible, but for sketchy senders, there seems to be better classification and a higher amount of email going to spam.

We analyzed the impact by looking at inboxing rates across our permissioned email panel.  Around the end of January of this year Verizon began rolling out their Atlas platform across a small percentage of inbound traffic, in the inbox this can be seen via their addition of the X-Atlas-Received header.  After an initial, apparently test, rollout a stepped rollout began to occur and you can clearly see that here:

Because the Atlas traffic split was done across all traffic, we can compare inboxing rates on the portion of traffic flowing through the Legacy and Atlas platform.  If we look at the median difference between a given senders inboxing rate on the Legacy Platform and Atlas, we can ascertain whether there’s a difference in behavior across senders.  If the platforms truly performed identically, you would expect the difference to be zero (or some distribution tightly distributed around zero). Instead this is what we see:
As expected by this being a purely operational change, we see a large concentration centered around no-change, but we also see this pretty significant set of senders with extremely serious inboxing impacts. The question then becomes who are these affected senders and what are their common characteristics. The first cut we will investigate is segregating out email service providers from self-hosted senders.

Here we see senders on Email Service Providers experiencing what we would expect from a straight infrastructure change. The distribution is well concentrated around zero. For the outliers seeing negative impacts within these groups, these are disproportionately represented at sending providers that do little to no sender quality checking at signup time (key amongst these are generic cloud providers who offer sending as a simple service). 

Within the self-hosted sender we see a grouping of senders that again experience what we would expect, with minimal inboxing differences under Atlas, but also we see the big peak of senders who have seen a radical decline in inboxing.

Looking at this group, we see some distinct breakdowns by the networks that the senders operate on. Grouping self-hosted traffic by the Autonomous System Number (ASN) of the network it’s being hosted one, and looking at the top 20 represented ASNs, we see a clear breakout of the bimodal curve:

In the low impact category, you see large strategic senders (like Groupon, LinkedIn and Microsoft), while the heavily impacted ASNs are largely hosting providers.

Together, these align with VMGs new platform showing stronger spam-folder placement of mail from poor quality/batch-and-blast senders that are typical on low-cost hosting platforms. At the same time, there’s strong evidence that for reputable senders working with knowledgeable providers, or sophisticated strategic senders operating their own platforms, that the upgrade to Atlas is a non-event.

~ George