PLTV: Setup

Predicted Life-Time Value has several outputs:

• Predictions dump

• Third-party services integration

• Dashboard

Assetario and customers will collaborate to define a reasonable set of intervals for LTV predictions (considering the customer’s apps retention curve, business goals, etc.) to predict LTV growth over time. The LTV predictions can be used in combination with a live operations plan to plot cohort-level LTV curves for up to D180 in the future, which will be updated regularly.

PLTV Configuration

Assetario deploys one machine learning model for each configuration provided by the client. Each configuration must contain the following information:

1. Input interval: How many days of activity to consider when predicting the user’s future spend. The typical default value is three days (72 hours), commonly known as D3.

2. Output interval: How many days in the future should Assetario predict? The typical default value is 30 days, commonly known as D30. If the input interval is three days and the output interval is 30 days, Assetairo will predict the total spend at the end of the 30th day of the user’s activity (since they have logged in for the first time), including the spend during the first three days.

3. Prediction level: What is the resolution at which Assetario should be making the predictions?

   1. user_level: Make predictions for every individual user.

   2. campaign_level: Group users by the UA campaign to which they were attributed and make a prediction for all users who installed the app due to that campaign on a particular day. For example, if 123 users came from campaign facebook-campaign-1 on 2022-02-22, then all these users will be grouped together and the average LTV of this group will be predicted.

   3. cohort_level: Same as campaign_level, but instead of grouping users per campaign, the users are only grouped per day on which they installed the app.

4. Prediction type: Each prediction type corresponds to asking and answering a different question.

   1. binary: Will the user spend more than [cutoff]% of all the other users? In this case, please define the cutoff value. An example cutoff value could be 99% or 95%.

   2. spend_bucket: If I split my users into spend quantiles, which quantile will the user be in? For example, will the user spend more than 25% but less than 50% of the other users or will they spend more than 50% but less than 75%? Or more than 75% of the user base? Please, define the bucket boundaries, which implicitly define the number of buckets as well. An example bucket cutoff value could be 0%, 5%, 35%, 51%, 98%, and 100% and would result in Assetario predicting a value between 0 and 4 included for each user.

   3. raw_spend: How much in $ or other currency of choice will the user spend?

Output Format

Assetario always outputs one file/table for each pLTV configuration. For example, if the client wants a D3 to D30 binary prediction and D3 to D30 raw prediction, Assetario would deploy a machine learning model for each, thus resulting in two outputs.

Assetario will deliver the output PLTV in raw format to a CRM or DB destination specified by the customer daily.

These output tables have a set of fixed columns:

1. identifier: this could be a user_id or a campaign_id or empty in case of cohort-level predictions, but it always represents an identifier of the user or a group of users for whom the prediction was made.

2. predicted_value: the value predicted by the PLTV algorithms.

   1. For user_level predictions, this value represents the actual predicted value for that user.

   2. For campaign_level predictions, this value represents the mean predicted spend for all users who installed on a particular install_date and were attributed to the particular user acquisition campaign.

   3. For cohort_level predictions, this value represents the mean predicted spend for all users who installed on a particular install_date.

3. install_date: date on which the user or group of users installed the application

4. created_at: timestamp of when the prediction was created. It could happen that Assetario updates its machine learning models or reruns predictions. Therefore, when analyzing the data, it is important to always take the most recent value for each user according to the created_at timestamp.

5. Custom Fields: These are defined by the customer and by how the PLTV values are used. For example, if the Assetario PLTV engine is integrated directly with Facebook Value Optimized Campaigns, the custom fields would include the user’s hashed email and hashed phone number to comply with the corresponding Facebook API.

A prediction is uniquely defined by the combination of these three fields: identifier, install_date, created_at

identifier	predicted_value	install_date	created_at	custom_field_1
A	$0	2022-01-01	2022-01-04T00:00:12	custom_value
B	$12.99	2022-01-02	2022-01-06T00:00:12	custom_value

Sample LTV predictions for Users A and B, made using data from the first three days of activity after they have installed the application.

Dashboards

Assetario’s product also includes dashboards for marketing managers to visualize the predicted LTV for all their marketing campaigns and be able to be filtered by any user segmentation. You can read more on the PLTV Dashboard.