The best marketing campaigns send the right message to the right user at the right time.
It’s a tried and tested method that creates an ideal experience for the end user, and maximizes results for the marketing team. The end user sees campaigns that are relevant to them–reducing spam and improving likelihood to convert–while the marketing team sees increased efficiency in ad spend and email sends.
Figuring out how to do this has never been easy though. Personalization on this scale—the type of personalization that drives your Netflix and Amazon recommendations—is prohibitively expensive. It requires access to high-quality first-party data about your customers, along with the machine learning systems to translate that data into predictive insights at a user level.
With the launch of Predictive Cohorts by Amplitude, that’s all about to change.
Predictive Cohorts enables you to group users based on future behavior, not past behavior. Powered by Nova AutoML, the feature automates all the steps of a machine learning model and democratizes access to personalization at scale for any company. Set any desired outcome—activation, retention, lifetime value (LTV)—and within minutes Amplitude will build cohorts of users most likely to achieve that outcome.
This is a monumental shift in how you build audiences and find the right users to target.
Gone is the guesswork of identifying the best users for a campaign based on a dozen different rules. Instead, marketers can now build audiences by a single campaign objective: personalized ads for the users most likely to buy; emails to the users least likely to retain; and in-app offers for the users with the highest predicted LTV.
Predict Any Outcome
To personalize your digital experiences with Predictive Cohorts, you must first predict the desired outcome you want the cohort to achieve. Amplitude makes this process as simple as point and click.
Inside Amplitude Cohorts, specify any event, user property and desired timeframe. Construct outcomes like:
- purchase in the next seven days;
- upgrade from starter to enterprise in the next 30 days;
- or reach LTV of $500 in the next six months.
Amplitude’s Nova AutoML system will then automatically build a machine learning model to create these predictive cohorts within minutes.
The machine learning model analyzes users who did versus did not achieve the desired outcome in a previous period (e.g. users who had greater than versus less than $500 LTV in the last six months), and weighs how hundreds of past behaviors impacted the outcome. The output is a probabilistic score for every user, recomputed every hour, identifying their individual likelihood in the desired time period.
This extensibility to predict any future outcome is profound in its implications. Instead of bucketing users by recency-based criteria (like “last item viewed”), you can now identify users by forward-looking criteria (like “month-one retention”). So rather than maximizing for short-term results — like incentivizing a user to purchase a recently viewed item, which can come across as click bait — you optimize for the long-term interest of the user. In this way, you tailor more personalized experiences to your best users, at exactly the moment when they are ready to receive it.
Identify Your Best Users
With a probabilistic score assigned to every one of your users, Amplitude makes it easy to identify who is the right or wrong user for your desired outcome.
In Predictive Cohorts, you can see a distribution of all your users, ranked by their likelihood to perform the desired outcome in the future. You can select any cohort—such as top 5%, median 10% or bottom 20%—and immediately see the count of users and their predicted conversion rate in that cohort. In the example above, we selected the top 20% of users by predicted LTV.
The power of this segmentation is impressive. Instead of manually grouping your users by a handful of behavioral signals, you’re ranking them by a predicted likelihood, which is based on hundreds of behavioral signals.
But this predicted likelihood is not meant to be a black box. Trust and integrity are key, especially in making ethical decisions about when to use a generated predicted model. We accordingly provide transparency into the underlying model’s accuracy, as well as insight into which behavioral signals were most important to it.
Behavioral signals are ranked by an importance ratio: the ratio of users in the selected cohort who performed an event or event property relative to those not in the cohort. In this example, we see that users in the top 20% of predicted LTV were 14.3x more likely to add a friend in the last 90 days, and 9.8x more likely to contribute to community content.
With sufficient confidence in the model, you can then save the desired set of users as a Predictive Cohort. The cohort will contain all users in the selected probability range, and adjust cohort membership every hour as your users’ behaviors and probabilities change.
Personalize Your Digital Experiences
The fun doesn’t stop at simply identifying your best users. Amplitude enables you to customize digital experiences to those users immediately as they qualify.
Using Amplitude Engage, you can sync a Predictive Cohort directly to your email, ads, and feature flagging platforms. That means as soon as a user enters (or leaves) a Predictive Cohort, they will be automatically synced to the right digital experience.
You can personalize your Braze emails by predicting which users have the highest affinity to churn next week and offer a higher discount to them to retain. In Facebook Ads, you can start to exclude users who have the highest affinity to buy in the next 30 days, as they are going to buy anyways. And using LaunchDarkly, you can now show different product flags to users who have the highest affinity for one product over another.
This gives true personalization power at your fingertips.
Get Access to Predictive Cohorts
Predictive Cohorts is now available as part of Amplitude Engage and includes pre-built integrations with your growth stack, like Amazon S3, Appcues, Braze, Iterable, Facebook, Google Ads, and LaunchDarkly.
With this technology, you’ll be set to anticipate opportunities for activation, retention, upsells, and lifetime value across your business. Download our Predictive Segmentation Playbook and schedule a demo with our team today.
Note: Amplitude recognizes that enabling self-serve predictive modeling comes with an ethical responsibility to ensure fairness and minimal bias. Read more about our planned efforts in Inclusive ML here.