All-AI List management system: Chapter 3 - Frequency

How do you manage a list of users? That is what I specialize in. I extract the most value possible from a given list of leads and/or customers. Many, many factors go into optimizing the return on such a list, and a good portion of these factors can / should be improved by AI. In fact, I am here to say that, at long last, it is now possible to set up an entirely automated list management system, providing a personalized experience to every single person on the list, thanks to a series of layers of decisions made by AI.

Throughout this series, I break down each of these layers, and explain how to best leverage today’s AI technologies to maximize the engagement / revenue you can get from your list of users, without ever discarding tried and true human-driven tactics. The winning formula for list management will always be a combination of AI and human-made decisions. What will change is the delineations of tasks assigned to AI and those reserved for human judgment.

Today’s layer: Frequency

The first layer of decision a list management system must make is to decide the frequency of communications to send? to each user. The entire point of selecting the right message to send is defeated if you are not sending at the correct frequency.?

Take a brand you have limited interest in, but in the back of your mind, you are not unsubscribing from because you are still holding out for the day where the circumstances align for you to engage with it and “do the thing” that helps its bottom line. If the algorithm that picks the right content for you does its job correctly, but sends you an email every day, you are going to click that unsubscribe button sooner or later, probably way sooner than later.?

Conversely, if you are passionate about a particular brand and are eagerly awaiting its next newsletter or promo in your inbox, but only receive one or two emails a month from them, you are going to feel disengaged from them and eventually lose interest. I don’t know for you, but when I don’t hear from a brand in my inbox after a while, I assume they went out of business or something bad happened to them. After all, I never unsubscribed. They just…stopped messaging me.

Most brands I know do not give this frequency layer any thought at all. Every user gets either the same amount of emails, or gets a random amount of “targeted campaigns”.

A case study

At Change.org, which is a billion-email sender, probably hunting down the trillionth email sent if not already well past this milestone, this cookie-cutter frequency was gospel. Except in the United States, which already in 2015 used AI to hit their list up to multiple times a day, everybody else around the world needed to send one email per week to each user. No more, no less.?

I challenged this conventional wisdom. As I detailed in a similar post back in 2016 when I left Change.org, I strongly intuited that some of our users would be very happy with multiple emails a week, while others probably felt like? one a month was plenty. Long story short, the worldwide org pushed back hard and helped me design and put in place a strict longitudinal test to measure retention, fatigue, and everything else under the sky that could possibly go wrong if we had the audacity to send more than one email a week to our most engaged users.

Well, guess what? Not only did every single engagement metric go up, the unsubscribe rate for the heavily engaged group basically collapsed to a value asymptotic to zero. There was an initial spurt of unsubscribes, and then everyone else stayed on, clicked on emails, signed petitions, gave money to the campaigns… After 18 months of using my system, Change.org’s active user list doubled in Canada, from 1.2 million to 2.4 million, while revenue from email went 3X. All key metrics were way up and to the right. Except the churn rate, which went the other way, and that’s obviously good.

Think about who you are thinking about

You have to keep in mind that when you’re concerned that you may be sending too many emails, you are de facto giving more importance to your least engaged segments. You are prioritizing not losing a user which is a few emails too many away from cutting ties with you, over a user who can’t get enough from your brand. How does that make sense??

Luckily, all that headache is a thing of the past. You can let AI select the best cadence for each user. Remember: AI is great at finding patterns in data. Way better than you or me. So let it.?

How I did it

Iterable has this hilariously mislabeled feature called “brand affinity”. Reading the name, I assumed it was some sort of social listening tool, taking the words used by a given user on their social network account while speaking of the brand and mapping them back to their profile in whatever platform you are hosting your data in. And, how cool would that be? As always, I seem to live in the future. It’s only a matter of time before we get that. But no, all “brand affinity” means to Iterable is how likely a user is to interact with your emails. It’s a good consolation prize, I’ll take it.?

So the first journey (formerly known as workflow, called “canvas” in Braze) you want to build is a simple daily list pull of your entire mailable list (you do have such a list, right? All cleaned up and develirable to? Right?). This daily pull will filter between the 3 tiers of engagement and send users accordingly to their respective delay nodes. A user will wait, say, 3 days if they are highly engaged, 7 days if they are averagely engaged, and 14 days if they are disengaged.?

You should definitely AB test the best frequency for each group, maybe your active users can take a daily email, maybe your disengaged users can't even put up with a bi-weekly email. Measure engagement, unsubscribes and churn if you have a system sophisticated enough to measure implicit churn. Check for RPU if you can and it’s relevant to you. Also, make sure to prevent users from entering that same frequency-controlling journey multiple times at once, or else you’ll end up with everyone getting an email every day.

All delay nodes, upon completion of the waiting period, send users to the (same) next journey in the system, which will be the topic of the next edition of “The Future is Now”.

Other considerations

You could tag your users with the latest status they had when they were sorted in their respective delay node (tag “average” affinity users with an ‘“average affinity” tag), but at least in Iterable, that engagement status is a user property that updates itself in real time, so the tag would likely be not only unnecessary, but a lagging indicator which could prevent you from taking the next best action on a user. For example, if a user is in long delay node because at the time they were triaged they were in the lowest engagement segment, but while in the delay node they self-motivated to go back to older emails and engage with them, then at the time of sending the email they should get content that reflects this latest spurt of activity, not the old news that they were disengaged two weeks ago.?

I also must bring out the inner Dela Quist in me and mention that engagement is not limited to a user’s interaction with your outbound commercial communications. A user that ignores all your emails, but comes back to your site regularly, shouldn’t be treated as though they were “disengaged” from your brand. And, spoiler alert, Iterable’s “brand affinity” feature doesn’t control for that, so you need to cover for that. Add a node after the “brand affinity” split to the “low” affinity path, and send users who performed relatively recent site visits or app logins to the “medium” engagement delay node.

I know this is technical stuff, but I’m telling you exactly how you can implement this system to boost, if not optimize, your list management, RPU, engagement… In short, this is how you get the most value out of your list, and perhaps even more importantly, offer your users the best and most personalized experience possible.?

In the next edition, we’ll talk about the next layer of personalization. Stay tuned!