What you'll do
In this tutorial, we will learn how to use TAZI. You will be able to create a Business Model and predict CUSTOMER_CHURN to have insights whether customer will leave in a certain period of time.
Problem: Predict customer churn within a certain #months time horizon. Need to minimise the “un committed” customer contracts and increase customer services quality and satisfaction. Customer behaviour changes all the time, as economy and other factors change.
Result: Decrease Customer Churn by 17% by taking proactive/preventive actions.
- Increase Customer loyalty (focused services)
- Product differentiation (eliminate high risk products)
- Sales performance management
The required data source is already in the system and can be used in the Business Models that you will create.
What you'll learn
- How to solve a business problem in an Automated Machine Learning platform easily.
- How to create multiple business models with different AI algorithms and configurations
- How to evaluate a machine learning model
- How to understand a machine learning model
- How to take appropriate actions in time with the help of Automated Machine Learning
- How to report the business model results
If you want to request a demo, please contact us Request for Demo
Recommended Browsers
We recommend that you use the latest version of Chrome for the best experience.
After you login to the TAZI Deploy interface, you will be welcomed to the Solutions page. Solution means a use case that you want to solve.
Each Business Model that is generated is linked to a specific Solution (use case) in the system. When you click to any solution from the Solutions page, you can see its own Business Models.
You can click on any Solution name from the Solutions panel and create new Business Model for that use case by clicking on + Business Model button. When you click Business Model button, it will create a custom business model by default which you can configure more parameter. Another option is creating a quick model. You can see these options by hovering your mouse over Business Model button.
Multiple Business Models can be created for a specific use case. Model configurations or data sources may differ in each Business Model.
You can have a new Business Model after you complete the following steps;
1. Connect Data Source
Choose customer_churn_100k data source from the list and click on
Next button.
2. Configure Business Model
Next, let’s set-up your model configuration. TAZI does model configuration automatically. You should only define your target and business KPI features for TAZI to produce outputs.
With our data source defined we can get into configuring our Business Model. In this section you need to enter target feature, KPI and also you need to choose problem type. Click
SUBMIT button after you are done.
In next step we can start profiler in order to have recommendations for our features. Simply click
Start Profiler button to continue.
You can have detailed information about TAZI Profiler here.
After the profiler finishes you will see the statistics for your features. In order to see TAZI profiler recommendations you can click
Show All Recommendations button.
You can individually select the recommendations and easily apply the recommendations by clicking on
Apply button below. In this example we will select all recommendations.
Click on
Next button to continue.
3. Connect Output Destinations
In this section you can configure output destionation. Continue by clicking
Next button.
4. Overview
This screen shows a summary about all the information you filled in. If you confirm this step, you can click on
Next button to finalize new Business Model creation.
Click on
Submit button and you will see your new Business Model in the following screen.
When you click
Start button, you will be able to start your business model.
After starting your model when you click on it, it will open its run page like the one above screen. Here you can view various configs and results. You can go deep dive into performance and details of your runs through various modules of TAZI.
In this section, you will see how you can evaluate model performance on the TAZI Live interface.
When you click on TAZI LIVE button from the following screen, you will be redirected to TAZI LIVE interface to see performance details of that Business Model.
In TAZI Live screens, there are various tabs you that can check different metrics and monitor your model's performance.
Also, TAZI Live screens slightly change based on your problem type. For detailed information you can view TAZI Help here.
TAZI uses an explanation algorithm to provide insights on the Business Models that are continuously trained on the streaming data. This section explains the use of an online decision tree and its sunburst representation to better understand the model behavior.
To start, click on the Model & Insights button to see the model selection page.
In this screen, there are two sections which are ‘Insights’, which provides evaluations and key finding of the model and ‘Explore’, which allows to explore and understand the model so that you can take appropriate actions on time.
In the Insights tab, Benefit Overview section shows the accuracy and the benefit value of the model for each class. For each class, the number of instances found from that class and the business value of the instances are shown using donut figures.
When you click on Explore tab, you will see the model representation.
- The model can be explored by hovering over nodes. When you mouse-over any node, you can see the pattern for that node on the top. A pattern refers a group of instances that have the same behaviours. Those patterns are shown as explanation to derive insights for that group of instances.
- Explanation is represented with the colors of the Target feature. These colors are set automatically by the system but you can change as you want in the Step 2: Model Configuration --> Target tab.
- Important Features are listed on the left side. These features are determined by the system based on their usage in the patterns of the model.
- Feature histograms, instances and trends belong to the pattern that can be investigated by clicking over any pattern. Take Actions will be explained in Section 5.
By double-clicking on any node, it is possible to take a closer look to the certain patterns.
When you evaluate the model performance and understand the model outputs, you can decide to take further actions. If you want to take an action for any pattern, you can select the action from the action menu for that pattern on the model.
To be able to evaluate model outputs easily and just focus on the most valuable patterns, you can open the Insights tab and review Most Coverage Nodes section.
Most coverage nodes demonstrate the most significant patterns containing many instances and having high confidences in the model.
- The instances of the patterns can be seen by clicking on the first button (1) near the related row below.
- Pattern can be investigated by clicking on the second button (2) below.
When you click the second button on the most coverage nodes table you can see that pattern on the Business Model.
If you think that the explanation of a particular pattern is valuable you can take actions based on that pattern. You can click on the pattern and choose
Take Action. You will see the defined list of actions.
The patterns that you take actions will disappear from the model representation. So you can focus on the others that you did not take any actions.
You can click on the Business Dashboard button as seen on below.
There are different dashboard widgets listed here to give some insights to you to evaluate the overall results. You can also customize this page as you want. Customization will be explained at the end of this section.
Let’s now look at the existing widgets and understand what we have as default. You can see the record counts of predicted results categorized as Churn and Remain. You can also see the total record count near this section.
When you scroll down, you will see a filtering section. If you choose any value from the filtering fields, the following charts are updated based on your selection.
In this section, you can see the whole data or just a part of it with predicted labels and save your search criteria to use later.
Below this saved search section, you can see the model expalantions that are listed for each pattern in the model.
You can also have some widgets to tag your data according to different feature values and you can see the majority of values for that feature.
Customize Advanced Reporting Page
You can click on the Edit button from the top menu of the screen and start to edit this reporting page.
In Edit mode, settings icon will appear in the top right corner of each widget. You can click Edit Visualization button below and start to customize that section.
In the following example, you can add new filtering fields by clicking on the Add button. When you finalize your customization you can click on the Save button from the top left and all your changes will be applied in the report.
