ESPs and deliverability

There’s an ongoing discussion, one I normally avoid, regarding how much impact an ESP has on deliverability. Overall, my opinion is that as long as you have a half way decent ESP they have no impact on deliverability. Then I started writing an email and realised that my thoughts are more complex than that.

Here are some excerpts from the email, because in other circumstances I would have just written it as a blog post.

I have dealt with ESP clients in the past who had a collection of customers that were so bad everything mentioning that ESP (even tests from my wttw account to my gmail account that simply contained the domain name of the client) went to bulk. That was a few years ago, though, and the gmail filters have improved and are in some ways even more discerning. I still occasionally find some domain reputations so bad it breaks delivery from one account to another. This is unusual, though, and it never happens overnight.

Of course there are things the ESPs can do that affect all of their client mail. Most of those things involve letting customers get away with bad address collection practices in enough volume that all the customers are considered problematic. If the ESP doesn’t make their customers behave and lets them send whatever they want to whomever they want, then yes, the ESP is going to have problems.

The ESPs with good delivery have extensive and active deliverability and compliance desks. One desk catches customers at the early end of problems, where it’s not enough to actually hurt their delivery but they’re clearly on a path to bad delivery. The other deals with customers who have not taken the initial advice and have continued down the path. What they’ve done is unacceptable and they have to either fix it and get back up to snuff or find a new ESP. 

The bulk of my clients right now are ESPs, or SaaS providers that are ESPs but don’t realize they are. They generally come to me because they’ve not been handling deliverability at all and now much of their customer mail is going to bulk. They didn’t see themselves as ESPs so they didn’t pay any attention to what customers were doing and there as enough grey mail to ruin delivery. The thing is, these folks are often using one of the commercial SMTP by API vendors (all the ones I can think of right now start with S) so I know that all of the actual technical stuff is correctly managed. I also know their overall complaint rates and bounce rates and all the surface stuff are within acceptable parameters, otherwise they’d be turfed off their SMTP provider. 

A lot of senders, and even some of the deliverability folks, haven’t really kept up with how the ISPs are tracking and filtering mail today. In the B2C space IP address is almost irrelevant. In the B2C space IP address gets you through the SMTP transaction. After that it’s (almost) irrelevant for inbox delivery.

This has been true for ages – it’s been 7 or 8 years since I had a Return Path certified client showing me data that their content mail and their advertising mail was delivered differently at Yahoo, despite identical authentication, IP and everything technical. ISP technology has only gotten better in that time. The content, history and mailstream reputation drives where the mail gets delivered much more than most technical factors, assuming a marginal competence in setting up the mailserver or using one of the commercial bulk MTAs.

In my own work don’t really look at IP rep any more – the publicly available IP reputations don’t reflect on delivery like they used to. I mean, I gave up joking about folks confused by delivery problems senderscore was 99 years ago because it was so overdone. Even now, good IP reputation gets you in the front door, it doesn’t get you to the inbox. 

Right now, delivery is challenging. The filtering technology we’ve been modelling for the last decade has changed significantly over the last 18 – 24 months in ways that confound those models. I think we’re in for another year or more of fine tuning before the filters themselves are stable enough to create accurate models. It is a gross oversimplification to blame any one factor for poor delivery.