Everything Marketers Need To Know About Engagement-Based Spam Filtering
Could your promotional emails end up in the "spam" folder, even if the subscriber opted in? Columnist Tom Sather explains how email providers use subscriber behavior signals to determine email placement.
Recently, there has been a lot of discussion about the use of subscriber engagement as a signal to filter promotional emails to either the inbox or the spam folder.
There has also been a lot of confusion regarding what the top email providers actually measure when making these filtering decisions, and for good reason. Not all of the email providers use engagement filtering in the same way, and how engagement is defined is different for marketers and email providers.
Many Shades Of Graymail
About five years ago, Hotmail announced that it was beginning to use subscriber behavior as a signal to separate graymail — promotional emails that a subscriber opted into, but are now opposed to receiving — and spam.
Since technically the subscriber knowingly signed up to receive the email, but later found that the emails were not relevant, were sent too frequently, or just weren’t interesting to them anymore, defining it as “spam” wasn’t black and white, but many shades of gray.
Graymail made it more difficult to separate the spam signal from the noise. Rather than unsubscribe from graymail messages, many subscribers would mark the opt-in promotional emails as spam. This unfairly punished brands and businesses sending permission-based emails, since spam reports from subscribers are a major signal for spam filters, helping them decide to send all future emails from that IP address or domain to the spam folder.
On the flipside, outright spammers and senders with questionable practices quickly learned that subscriber complaint issues could be mitigated by keeping inactive email addresses on their file. Inactive email accounts never report an email as spam, and by increasing the number of these inactive email addresses, they could quickly and easily reduce their complaint rate.
To address issues of permission-based promotional emails landing in spam and spammers gaming the system by padding their files with inactive email addresses, email providers looked for subscriber behavior signals to determine where an email should be delivered.
For most permission-based senders, engagement means personalized deliverability. Since the definition of spam is in the eye of the beholder, email providers started filtering emails from brands or businesses based on how a user interacted with them, meaning if a person read every email sent from a brand religiously, future emails would be delivered to the inbox.
If another person received the same emails, but interacted with them in what is considered a negative way, those same emails for would go to that recipient’s spam folder.
The email providers look at hundreds of signals to determine if an email is wanted or not, but for spam filtering based on engagement, the following metrics are some of the most important to track:
- Messages read – a positive indicator that the individual wants to receive your emails
- Messages replied to – a positive indicator that the message is likely personal in nature, and desired
- Messages marked as “not spam” – a very strong positive signal that email providers and ISPs use to train their spam filters
- Messages marked as spam – a very strong negative signal that your email is spam and does not belong in the inbox
- Messages moved to other folders – an indication that the recipient wants your email, but also wants to keep and organize it
- Senders/domains added to address book – a positive signal indicating that future messages should be delivered to the inbox
- Messages forwarded – a positive indicator that the recipient desired the message and that others may want it, too
How Click-Through Rates Affect Engagement-Based Spam Filters
There has been some confusion around the use of clickthrough rates in engagement based spam filtering. In a nearly five year-old presentation (slide 8), I had mentioned that “ISPs don’t care how many clicks or conversions you get,” and that still rings true today.
Email providers and ISPs still don’t measure click activity in their engagement algorithms. If that’s true, then why have so many deliverability experts advised email marketers to look at clicks?
The advice to look at clickthrough rates to get a handle on engagement-based spam filtering is primarily due to the fact that marketers often don’t have access to the same metrics as the email providers and ISPs. Most marketers today can easily see the following the metrics in their own email reports:
- Messages read or opened
- Messages marked as spam
That means that marketers are flying blind when it comes to knowing how many of their messages that were marked as “not spam,” messages deleted without being read, messages forwarded, messages moved, and how many subscribers added them to their personal address book.
Until recently, these metrics weren’t available to anyone. As a result, marketers and deliverability consultants had to look at the closest proxy – opens, clicks and conversions.
What a marketer looked at to determine if they had issues with engagement-based spam filtering was determined by how much data they were collecting. Looking at data beyond email analytics, like past purchases, downloads and website activity, will always be a stronger gauge of engagement filtering issues than opens and clicks will ever be.
In my last column, I looked at engagement with welcome messages and purchases. I found that if a subscribers read three emails in a welcome series, they were much more likely to read a majority of future messages, and also to purchase more.
However, even those that didn’t read a single welcome message, or any messages following, were still valuable and were still purchasing. Therefore, removing inactive subscribers who have never opened or clicked a message probably isn’t wise and will result in removing valuable customers.
Removing Inactives: When And Where It’s Okay
Removing inactive subscribers is something that many experts advise doing to alleviate deliverability issues caused by engagement-based spam filters. However, even that advice is a little bit gray.
For Outlook.com, engagement filtering happens at a personal level only. Because of this, removing inactives to improve inbox placement rates caused by lack of engagement is somewhat unnecessary. However, the majority of spam filtering at Outlook.com occurs not because of a lack of engagement, but because of reputation-based spam filters.
These filters look at many things including spam traps – email addresses that never sign up for anything and are used to catch “spam.” Spam traps typically do not open or click an email, so eliminating accounts that never open, click, or convert can be a valid way to clean your list of these accounts and improve deliverability caused by reputation filtering – just not engagement-based spam filters.
Gmail, on the other hand, uses subscriber behavior signals slightly differently. They have advised marketers sending from new IP addresses to prime them by sending emails to only the most active and engaged users at first before sending to the rest.
Sending to your most engaged “fans” can also help boost reputation metrics if you’re careful. One retailer who was experiencing low inbox placement rates leading up to the holidays, slowly stopped sending to inactive email addresses, and their inbox placement rates went from 0% to 100% in a short period of time.
By focusing on only engaged users using past click behavior, and website and conversion data, this retailer improved placement by betting that their active subscribers would mark them as “not spam” and avoid marking their promotional emails that landed in the inbox as spam. This tactic also reduced negative engagement metrics like “deleted, not read.”
What Marketers Should Really Focus On
While it’s okay to use clicks, conversions and other data as a proxy, engagement-based metrics like “deleted, unread,” “this is not spam,” and “add to address book” are available today through subscriber panel data providers. It’s a much better way to measure personalized inbox filtering decisions and determine if engagement filtering is the actual culprit, or if it’s due to reputation.
Reputation is still the primary reason that marketers see inbox failure en masse. Complaints, spam traps and unknown users are three of the most important measures that email providers look at to determine if a messages should be delivered to the inbox or the spam folder.
Resolving reputation issues often requires optimizing your list acquisition practices, and then cleaning your list. Though many tout cleaning inactives as a technique to combat low engagement filtering, oftentimes the more valid reason to remove inactives is because they’re a sign of old, dirty data.
Focusing on acquiring clean data, cleaning up existing bad data, and providing a great subscriber experience will help reach the inbox more often than focusing and worrying about engagement filtering.