LTV and Churn
Last updated
Last updated
Learn how to use the predicted Customer Lifetime Value (LTV) and churn attributes to understand and predict each of your customers’ purchasing behaviours over time.
Understanding predicted LTV and churn can offer valuable insights into customer buying habits. This information can help identify potential future purchases and opportunities for cross-selling and up-selling. You can also utilise these attributes to personalise the customer buying journey, especially during important times of the year or events that are significant to specific customer segments.
This feature predicts 2 different attributes:
Future Customer Lifetime Value (LTV): this indicator computes the expected revenue that a customer will generate in a given future period (1, 3, 6, 9, 12, 24 or 36 months). It allows you to segment a customer base, based on their future monetary value. It unlocks several use cases, such as targeting expected high revenue customers.
Churn likelihood: this indicator gives the probability that a customer will leave (not buy anymore or in a few months in the future). This allows you to target customers that are likely to leave at the right time and keep them engaged with actions that are revenue driving, in turn, increasing LTV.
Each time a consumer makes a new action (purchase, engagement, etc.), the predictions are adjusted accordingly.
DinMo's predictive analytics are invaluable for optimising marketing budgets and tailoring customer communications. While exact predictions for individuals aren't possible, these tools shine by averaging data across a broad customer base. Some customers may surpass their predicted lifetime value (LTV), others may not reach it, yet these differences generally even out overall.
Go to the Predictions section and choose Customer Value & Churn Likelohood to predict LTV and churn predictions.
Once on the LTV and Churn configuration page, select the following fields:
Model containing the Customers' data: This user model must contain all the customer data you wish to use for training. You must have a clearly resolved customer list. Ensure there is a relationship with an events model containing all transaction data.
Model containing the Transactions' data: This event model must contain all the transactions associated to the customers in the model selected above.
Once done, give a name to your new intelligent Engine (and optionally, a description). Click on Save to save this new training.
Once the training configuration stage has been completed, you can create predictions corresponding to the different scores you want to use in DinMo (e.g. churn and Customer Value at 1, 6, 12 months).
To do so, go to the Predictions tan and click on the "+ New Prediction" button.
Configure the prediction:
Select the score you wish to compute, such as churn likelihood or Future LTV.
If you select Customer Value, choose the time window for the prediction. For example, selecting a three-month window will compute the future customer value within the next three months.
Name the prediction and give a description (optional)
Pick your schedule type to determine when to update your prediction (manually or on a regular basis)
Click on Save.
Once you've created your training and predictions, it will be available on the page listing all your predictions. You can get more details about one specific training by clicking on it. Check the Predictions tab if you want to learn more about all the scores you compute.
By clicking on a specific Intelligent Insight, you will find several informations in the Training tab:
Prediction quality: When you configure a training on DinMo, the data used as a training set is split into two data sets, one for model training and another for model evaluation. After training the models, we compute predictions on the holdout data to compare it to the ground truth we previously stored. This gives you an idea of the accuracy of the model.
Intelligence engine parameters: All the set-ups you did in the previous flow. You can modify them via the “Set up” tab.
Scores are automatically transferred back to the Users model used for training. As a result, you have direct access to churn and LTV information in the user model you use on a daily basis in DinMo.
For each prediction, an associated predictions model is also automatically created, pointing to a predictions table in your data warehouse. You are free to use the results stored in your data warehouse, particularly for BI purposes.
In this prediction model, several fields are populated and refreshed automatically:
primary_key
: It corresponds to the primary key of your model. This field is not intended for use in segments/activations.
user_unique_id
: This is the unique identifier (primary key) of your user model. It can be a customer_id, an email, etc.
last_update_date
: This date corresponds to the last time a user (identified by a user_unique_id) was updated, i.e. his predictive attributes were updated.
This may be the case for a number of reasons: the customer has performed an action that has an impact on predictions (e.g. a transaction), you have forced a new prediction, etc.
If you predict a customer's future value:
expected_additional_ltv
: This indicator computes the expected revenue that a customer will generate in the future, at a specific prediction window (1, 3, 6, 9, 12, 24 or 36 months)
prediction_window_months
: This corresponds to the attributes prediction window, which you configured during the set up. You can have several windows predictions (1, 3, 6, 9, 12, 24 or 36 months) in the same prediction.
If you predict the churn likelihood:
churn_score
: This indicator gives the probability that a customer will leave (not buy anymore or in the given time window).
Again, exact predictions for each customer are not possible. Some customers may surpass their predicted lifetime value (LTV), others may not reach it, yet these differences generally even out overall. It's exactly the same for churn, or the number of predicted purchases.
To apply AI attributes as filters for your segments:
Create a new segment and select the user model you used for predictions.
The score will appear in the filters of the users model, like all the attributes specific to the customer. For example, you can choose customers with a churn likelihood above 90% and a LTV lesser than 10 in the coming month.
The most frequently encountered customer cases are listed in the illustration below.
To set up these use cases, you'll need a few basic segments.
LTV-Based Segments (Time window can be adjusted)
Low LTV: Bottom 20%
Regular Customers: LTV between 20% and 80% percentile
Top Customers (based on future LTV): Top 20%
Churn-Based Segments (Recommended to use a short time window, e.g., 1 month, to identify at-risk customers)
Loyal Customers: Churn probability in the bottom 20% percentile
Moderate Risk: Churn probability between 20% and 70% percentile
High-Risk Customers: Churn probability between 70% and 90% percentile
Lost Customers: Churn probability above 90%
It is of course possible to use other segments, combining several attributes (predictive and non-predictive) to implement more advances use cases.
Top Customers (Advanced): Customers ranked in the top 20% based on both current LTV and predicted future LTV
Future Whales: New customers whose buying behavior closely resembles that of top customers
Priority Win-Backs: High churn probability combined with historically high LTV
Low Purchase Propensity: High churn probability combined with low predicted LTV and a history of infrequent purchases.
Transaction Value Field: This field must be available in the events model you've just chosen. This represents the amount a customer spends per transaction. This field should be in a unique currency. Don't mix euros and dollars, for example.