What are practical applications of AI we can use to engage our customers?
Tune in (or scroll down) to see practical steps and methods you can use to engage your customers in more meaningful and personalized ways:
The world may be abuzz with ChatGPT, but AI ain't new
Just look at the credit card companies. They've been using AI for years.
...not to do silly things like write ads, but to do big, complex, and important things like learning patterns and uncovering anomalies in users' spending behaviors (That's literally how fraud detection works).
The problem is that lifecycle marketers—who have incredible amounts of data and the most potential to drive significant revenue for their companies without pouring all of their money into ads—haven't had these tools.
Instead, we've been given flowcharts:
Oversimplistic, 'one-size-fits-none' strawmen that we call "user journeys."
And, because of the limits our tools put on us, we spend our time doing silly things like A/B testing whether we should send an abandoned cart message after 20 minutes or one hour or trying to see if "Start my free trial" works better than "Start your free trial" when we know we could be doing much more useful things.
Because also, because working with these tools is so radically different from actually experiencing one of these "flowchart user journeys," we can often create experiences that are frustrating for our users, like receiving the same (or very similar) message copy at regular intervals.
This both annoys our users (who can pick up on the patterns, if only subconsciously) and actually causes them to ignore our messages (It's called the habituation effect. Check it out.)
...but we're not this gullible.
We know our users' lives are much more complex than "One message per day at 9:00am."
AI finally gives us the chance to personalize at scale
...and that capability is built on these little things called "propensities."
A propensity is data-backed prediction of whether or not a user will do something in the future (e.g., purchase, subscribe, churn, etc.)
...not whether they have done it, but how likely are they to do it. And this radically changes everything.
Let me demonstrate...with discounts.
Example: When should you send a discount?
It's a tricky question, right?
Should we send discounts when the business unit says to? When we're overstocked? When our user hasn't bought anything in a while?
Propensities give us another option:
You should only send a discount when your user is 60-80% likely to buy.
Giving a discount to a user who is already 90%+ likely to buy is like throwing your precious margin out the window, and giving a discount to someone who's only 20-40% likely to buy isn't going to be effective. You'd be much better served sending a product recommendation to a customer who isn't likely to buy.
...and even if these percentages aren't that accurate, they're much more accurate than just blasting a discount to an entire segment of users or sending them out after some predetermined period of "inactivity."
There are many other uses of AI-driven propensities as well
- Recommender systems (which are incredibly effective) are powered by propensities.
- You can use propensities to understand which of your app events tend to lead to conversion or churn and do things like churn prediction based on which users are doing which events.
- You can also see huge improvements in your KPIs just by applying propensities to your message timing (Not looking at when a customer is typically active, but looking at when they're most likely to be active next.)
Here's a super quick example from a gaming app:
A quick propensity analysis found that new users who played Gin Rummy were only 12-15% likely to be retained to the next month:
By offering a different game to new users, they boosted their user retention by 4x.
Here's another one from an eComm app:
They found that, if a user used a gift card for their first purchase, they were 70%+ likely to churn.
By routing these users into a loyalty program, they saw a jump in user retention as well.
(...and, mind you, these results will be different for every app.)
AI-driven user journeys look different
Instead of one static flowchart, they look like this:
...and even like this:
Some users get five or six messages. Others get dozens of messages.
Each individual user gets their own, individualized experience based on their activity level and demonstrated preferences, and we see incredible results:
Isn't that how it's supposed to be?
So, how can you start using AI to build your user journeys?
If you have an app, the first part is easy:
Gather your data
All apps have event stream data.
This data is an amazing record of every single one of your users taps, swipes, purchases, and more. Most of the time this data lives in a data warehouse like Amplitude or Segment, only to be used when you want to ask a specific query.
We feed this data into a propensity model, which gives us a ton of rich insights, like the ones we shared above.
Act on those propensities
After we get this data, we allow you to use it, either by making triggers based on propensities, or by letting our AI do the work for you.
Oh, and you don't need to change your current stack to implement this. We can sit on top of whatever tools you're currently using and bring you this functionality with a few simple API calls.
Most apps who use us are up and running and seeing results in two weeks or less.
Propensities > Events
Here's the breakdown:
...but you really should watch the video. The gifs are funnier in real-time.