Schaun Wheeler
September 6, 2022
5 Min Read
Data Science

How do you know which messages to write next?

We're making it easy to know which messages you need more of, you can keep your users engaged.

We’ve written a lot lately about the importance of message timing. We’ve shown that even a modest-sized user base can exhibit a ridiculously high number of distinct timing preferences. We’ve shown that just getting timing right can result in a 300% increase in ROI - not in click-through rates, but in actual customers-put-money-in-your-pocket. We’ve even gone so far as to say that the timing of the message is clearly more important than the copy you put in your message. 

We stand by that last statement, but of course that doesn’t mean copy is unimportant. 

Aampe lets you optimize your timing and your copy at the same time, and the vehicle we use to optimize copy is ‘labels’. 

Using personalization labels to understand user preferences

You can check out our tutorial if you want to learn about labels in more depth, but briefly: when you write a message, you can leave a space for, say, a value proposition. Your customers don’t have only one reason to engage with your product—Different value propositions (and different offerings and even different greetings or emojis) speak to different users, and you want to be able to reach everyone in words that speak to them personally. 

So you write your message, you write one value proposition in that message, and then you label that value proposition as a value proposition. Now our system understands what that is, and, once it’s labeled, you can write lots of other value propositions that speak to all your different customers. 

Aampe is able to use the original message as a template, and use the labeled portions to mix and match different text variations to quickly take you from 1 to 1000 messages, and then beyond.

Keeping the conversation going

So you have all those messages, but some subset of the total messages simply won’t appeal to certain users, so Aampe will stop sending those messages to those users. Heck, even 1,000 messages will become stale after a while! Our goal is to allow you to build conversational relationships with your customers at scale, which means you’re going to talk to them again and again over a long period of time. 

Pretty soon, you’re going to need new material, and at that point, you have to ask yourself what new material you need. You wrote a bunch of different kinds of messages—do you have to write more of every single kind, or can you double down on certain themes?

The answer: You should totally double down on the messages you need, and Aampe shows you precisely where to do that. 

How to prioritize message creation

There are several ways you could think about prioritizing copy-writing, but one way is to think about the size of the opportunity: It’s possible that you have a few niche offerings that are really important to a subset of your customers, but it’s a small subset. Writing more copy for those niches will help you grow the engagement of those customers, but it won’t necessarily help you grow a greater number of engaged customers.

Aampe’s learning system works on the basis of personalization scores: numerical representations of every user’s preference for every messaging choice you can make—So if you send a subset of users messages about, say, convenient delivery, our AI will use the results of that experiment to assign every single user a personalization score for “convenient delivery”. This allows us to estimate the size of the opportunity for each label you use in your messages (You can learn more about personalization scores more here - that’s a non-technical version - and here - that’s the technical version.).

When estimating the size of an opportunity, you have to take several things into account at once:

  • How strongly each user prefers a particular option.
  • How likely each user is to return to the app.
  • How likely each user is to respond to messaging in general.
  • How to balance all three of the above pieces of information across all users.

(The reason we need to incorporate information about likelihood to return and likelihood to respond is because not all users are equally valuable. If we just averaged personalization scores across all users, we’d be treating the user who buys something every single day the same as the user who visited the app once two months ago and hasn’t been seen since. A simple average would potentially drastically misrepresent the situation.)

When it comes to aggregating the preferences themselves, we also have to take into account negative opportunity: if 30% of your users have a strong preference for a particular offering, and 30% have a strong lack of preference—they really would rather look at any number of other things—then that still leaves 40% of your users who are persuadable. That’s very different from a situation where 30% have a strong preference and 60% have a strong lack of preference. In that second situation, you just don’t have much room to grow.

We won’t go into any more of the mathy details here. The upshot of all of this is that we aggregate use preference, activity, and responsiveness information into an Opportunity Index that tells you how likely you are to benefit from creating more content for any particular label. Here’s how it looks for one of our customers:

The Opportunity Index

Every dot is a campaign - multiple dots per row indicates labels used across campaigns. Along the left, you can see the different kinds of labels they’ve created: calls to action, greetings, incentives, and value propositions. We’ve replaced the labels with numbers in order to anonymize the data - for our customer, each of those numbers is actually human-readable text.

The Opportunity Index runs from 0.0 to 1.0.  (We like percentages because they’re easy for people to think about…and there’s some cool math stuff you can do with percentages that you can’t do as easily with whole numbers…we like cool math stuff). The higher the value on the index, the more room you have to grow. A score of 0.5 essentially means it’s a coin toss whether writing more content for that particular label is worth your time. The further the label indexes below 0.5, the more you should probably spend your time building out copy for some other label.

It’s not just about more messages. It’s about more of the right messages.

To communicate with your customers individually, continuously, and at scale, you need to write a lot of messages. We make it easy to write those messages. Now we’re making it easy to know which messages you need more of, so you can keep your users engaged and clicking.