◾Best Time Model
Last updated
Last updated
‘Best time’ is an AI-based predictive model that can identify user behavior patterns and send SMS, Email, and push notifications at the best time for every particular user.
Operators can build their best-time models under the Label tab in the BO. When creating a model, you can control the main aspects that will influence model performance:
Activity trigger – you are selecting user activities that will be taken into account when analyzing user behavior. We are commending always to have as minimum “Login” and “Online” activities to build a model for users that didn’t generate any Product-specific activities.
Algorithm – there are 3 algorithms available for your choice, you can see the result on the test user in the right top area.
Day & Hour selection logic – you can instruct mode to either select multiple best hours during each day and multiple best days, or to have only one best hour and one best day.
Sensitivity – is another parameter that controls the threshold in the selection of multiple best hours. A higher value will select more hours during the day. Setting the value to 99, will almost disregard any algorithm-based predictions and will reflect user activity hours.
Smartico user ID for preview - you can use this option to preview the best time model that you’ve created first with 1 Smartico user, before building it for the whole user base.
After setting all the options you can prepare the row activities data for the model.
Once the row activities table is ready, you can build your best time model.
As soon as you are ready with the model, you can build it for all your users by selecting ‘Reset all users & calculate model’.
Depending on the number of users and activities they generated, the initial building of the model may take hours, after that new users and new activities of existing users will be applied to the model with an up to 5 minutes delay (e.g. after registration of new user of after his first actions like bets, deposits, etc, it will take 5 minutes to build ‘Best time’ model).
You can define up to 3 'best time' models, which can be used to test different algorithms and sets of activity triggers. Once the model is created it will be available in the ‘Best Time’ section under the Label tab, where you can also see the active campaigns in the ‘Campaigns: active’ column:
The best-time model includes a dedicated report tab, allowing operators to access campaign-specific information about this model. They can check the ID of the campaign, the Label, the Campaign Name, the Campaign Status, and the creation date. The report can also be exported.
As soon as the model is defined and calculated for all users, you can select it during the building of marketing campaigns. At the Mail, SMS, or Push notification activity level, you can select a specific model as an option in Allowed communication hours.
In addition, you can choose if communication should be sent in the next best hour or the best hour of the best day. For example, If the campaign starts at 8 AM on Monday and targets ‘Next best hour, no matter the day’, it will be sent to this user at 1 PM on Monday.
The second option, ‘Next best day and hour’, will send the communication at 1 PM on Tuesday, because Tuesday and Wednesday are predicted to be the best days for communication.
The Third available option is ‘Next best hour or as soon as possible during current day’. With this setting, if the Best Communication Time for players is not today or has already passed, the resource will be executed immediately as the element becomes active. For example, if you have promotions that are valid only for 1 day and you want to use the BCT models when picking the ‘Next best hour or as soon as possible during current day’ option you are making sure that the communication will be sent to players in their Best time (if it is during the current day) or right after the element becomes active either way, it will be sent during the current day.
As soon as you define the 'Best time' model(s), you can see predictions for each user on the CRM page.
