You can use the Zia prediction builder to forecast growth by year, costs, taxes, salary expenditures, the likelihood of winning or losing a deal, the likelihood of a user buying a product, and more. 

Profile permission to configure Zia Prediction Builder

1. Manage configuration: Only those who have the manage configuration enabled for their profile can create, edit, view, enable, disable or delete a rule. 
   
2. View Results: Users with this permission can only view the prediction results. 
   

Note:

By default, view results will be enabled for the profiles when prediction permission is toggled on. Manage configuration must be enabled manually. 

1. Go to Setup > Users and Control > Security Control.
   
2. Select a profile. 
   
3. Go to Setup Permissions >click Zia > toggle Prediction.
   
4. Check Manage Configuration or View Result. 
   

Business scenario

Predicting likelihood of insurance purchase: Zylker offers insurance solutions at competitive prices to both local and international customers. Their customer base is made up of small and medium-sized businesses from different industries. Since every customer's requirements are different, Zylker's reps are encouraged to have face-to-face discussions, visit customers' offices, share new policies and plans, and conduct surveys to learn more about their needs. For the company to reach their potential, they want to be able to predict: 

• The different types of insurance a particular customer is likely to buy
  
• The likelihood of a particular insurance policy being bought by a specific customer
  
• The average revenue generated from different customers based on the type of insurance they purchase
  
• Any variations in year-on-year growth
  

These predictions can help them make better offers to their customers, modify existing sales and marketing strategies, and tweak current processes to increase the overall yield. 

1. The number of leads that will be converted.
   
2. The number of deals that will be closed won or closed lost.
   
3. The approximate time of deal closure.
   
   

Specifications of Zia Prediction Builder

Supported data types

• Date/time
  
• Date
  
• Percentage
  
• Decimal
  
• Number
  
• Currency
  
• Boolean / Checkbox
  
• Single Picklist 
  

Supported modules

The prediction builder is supported in both standard and custom modules.

Data limit

The prediction algorithm will only work if you have at least 200 records that match the  criteria that you have used to train Zia. 

Points to consider

1. Use a unique name for each prediction for ease of identification. The name you choose will be used as a field label to represent the prediction output in the module. The configuration name cannot be edited once the prediction is saved.
   
2. If you are predicting data for a picklist field, each value must have at least 75 records for the prediction model to work accurately. If one of the picklist values has more records than others, then there are chances of getting a skewed outcome because the data will be biased towards one type of value. Zia can make predictions for a maximum of 10 picklist values. 
   
3. If all the existing records have the same value for a number field, Zia will not be able to make predictions. For example, if a custom field - "number of children", has 2 as the value for most of the records, then Zia will not have any data to draw comparisons, and cannot make a prediction. 
   
4. If you make a time-based prediction, Zia will consider fields like created time a as reference points to calculate future time points.
   
5. Predictions cannot be made for fields that are generated via integration. Those fields will however, be considered as contributing factors for making predictions.
   
6. Zia will take 24 hrs from the time of configuration to make predictions. The scheduler will follow the organization's timezone. 
   
7. The model will be retrained once every fortnight. 
   

When to use which field type

You can plan and modify (if required) your data types in CRM based on the information you want to be able to predict:

• If you want to predict salary, expected growth or revenue, number of deals closed by a rep or insurance claim amount you will need to have this data in numerical or currency fields.
  
• If you want to predict whether a customer will buy a product, the likelihood of winning a deal, or if a subscription will be renewed, the data will need to be in a picklist field. 
  
• If you want to make time-based predictions, like when a deal will be closed, or a product delivered, probable month insurance claims is likely to be made, the data will need to be in a date/time field. 
  
• If you want to predict the likelihood for binary true/false or yes/no data (these values can be in picklist or check box fields), then the prediction result will be represented in terms of a percentage ranging between 0 to 100.
  

Where can you see the prediction outcome?

In the above image, you can see that

1. The predicted outcome is Health (for the Insurance type field)
   
   
2. There is an 80% chance for the prediction to be true, and it's trending upwards.
   
   
3. The Any past hospitalisation and Other medical insurance fields need to be filled to improve the quality of the prediction.
   
   

When you click Learn More, you can see the list of fields that positively and negatively contribute to this prediction.

Note

By default, the term "prediction" will be added to the custom field, so we recommend not to enter the term "prediction" in the prediction name. 

Designing the prediction model

Once you have decided what you want to predict, the field that represents the value, and the module that contains the field, you are ready to build the prediction model.

The Zia prediction builder consists of the following components:

1. Prediction name: Enter a unique name for the prediction builder for easy identification across your organization. The name you enter here will be used as a field label for the prediction results.
   
2. Source Module: Choose the module that contains the field that you want to predict. It can be a standard or custom module.
   
3. What do you want to predict?: Select the field which has the information relevant to what you want Zia to predict. If Zylker wants to predict the likelihood of health insurance being bought by  IT companies, it will select the picklist field called "Insurance type".
   
4. Select source value manually: This field will be displayed only if a picklist field is selected for the prediction. You can toggle it if you want Zia to predict an outcome for selected values only. If Zylker wants to predict the likelihood of health insurance been purchased, we will select the picklist value "Health".
   You can select multiple picklist values in a source value, for example, "health" and "personal accident". In that case, Zia will create a prediction for the selected values.  
   
   

• Define negative values: Values that are unrelated to the data you want to predict are the negative values. By defining these values, you will set a boundary for what data is relevant to Zia. In our example, you could select "automobile" insurance as a negative value.
  

The option to select a negative value will be displayed only if you have selected one value in the source field. If you select at least two values in the source field then Zia will not prompt for you to define a negative value.

• Which records should Zia consider as learning data: If you want to predict a picklist field, it is not mandatory to define learning data. By default, Zia will take all the existing records into consideration for making a prediction—you should have at least 200 records with data related to what you want to predict. However, you can distinguish the records that have the right values by specifically selecting them.
  
  
  
  For example, if family medical history, annual income, other medical insurance, marital status or age have values that will help Zia make the prediction, then you can select these fields as learning data.  
  

Note

It is mandatory to select learning data for number, date/time, and boolean fields. 

Creating a Prediction Model

You can build a prediction model for both standard and custom modules.

To create a prediction model

1. Go to Setup > General Settings > Zia.
   
2. Under the Predictions tab, click New Prediction.
   
3. Enter the Prediction Name [Opting for health insurance] and select a module from the drop-down list [for example, Insurances].
   
4. In What do you want to predict? select the appropriate field from the drop-down list.
   
5. Toggle Select Source Value Manually to specify the fields. 
   
6. Choose the field(s) [Health].
   Only if you select a picklist field for prediction and choose a single source value, you must mention the negative value too. [Automobile]. 
7. Click All records [Insurances] or Specific records [Industry is IT] from the drop-down list.
   
8. In Which records should Zia use for learning?, select the fields and the values to use for the prediction [marital status, family medical history, income etc.].
   
9. Click Save.
   
10. In the custom field creation Attention popup, click Yes, Create. 
    

Prediction Details Page

1. Summary
   
2. Custom field
   
3. Model accuracy
   
4. Prediction accuracy
5. Contributing factors
   
6. Additional information
   
7. Waiting for data
   

Summary

1. the module for which the prediction is configured.
   
2. the field that is been predicted.
   
3. records that will be predicted. If prediction is configured for specific records, then the criteria will be shown. 
   

 Custom field

Model accuracy

The erstwhile component Learning Pattern Score has been renamed to Model accuracy.

1. Less than 50: Poor prediction.
   
2. 51 – 80: Moderate prediction.
   
3. 81 - 100: Excellent prediction. 
   

1. View version history
   
2. Compare model accuracies
   

View and compare model versions

Note:

1. The factors included and excluded during learning are defined by Zia in the corresponding version and a user cannot change them.However, you can moderate those contributing factors and it will reflect in the upcoming learning schedule.
   
2. The date logs indicates the retrain schedule your prediction rule is going through.

Furthermore, to assess the efficacy of Zia's learning, you can also compare up to three versions with the configuration created.

Prediction Accuracy

Contributing factors

The fields that influence the prediction are the contributing fields. These fields are identified by the system on its own. A user cannot change or modify the fields that appear here. Basically, these fields will show the data that was taken into consideration by Zia to predict. For example, to predict the prospect status Zia may have used the following fields to pattern: prospect source, number of employees, industry and annual revenue.  

This section will help you understand what influences Zia's prediction. If needed you can review these fields and values and make changes as per requirements to see if the prediction differs.  

Additional Info  

This section gives graphical representation of data that is most likely to be predicted by Zia. For example, if health insurance is the most preferred type of insurance in your company, then the likelihood of new applicants choosing health insurance is more as compared to other insurance types.   

Waiting for data

This section will appear only when Zia is un able to predict due to insufficient records (Zia needs a minimum of 200 records to predict). It displays the number of records that are available for Zia to learn. In the following image, Zia is configured to predict the Expected Revenue for Deals module. The waiting for data section shows that there are currently 0 records therefore the prediction cannot be made . 

The user will also receive a notification in Zia Notification Panel that Zia couldn't predict the outcome due to insufficient records.