How to Create a Fit Model in Lattice Predictive Insights (LPI)

Context 

Lattice Predictive Insights (LPI) models are used score your records in real-time that your sellers can use to prioritize the right leads or accounts at the right time. This article covers how to create a LPI scoring model. 

Pre-Requisites: 

• You must have the Lattice Predictive Insights platform included in your contract. 
• You must have created a training file already to use to create the model. 

Steps to Create a Model

Step 1: Click on create a new model 

Step 2: Model Settings 

• Load your training file 
• Choose Model Type 
  • Account: If you are looking to score accounts for your sales reps to prioritize choose this option. An example uses cases for this type of model are: 
    • Scoring accounts in Salesforce that your sellers are prospecting on. 
  • Lead: If you are looking to score leads to include in information you are sending to your sellers choose this option. Example use cases for this type of model are: 
    • Scoring lead records that are created in your Marketo or Eloqua instance and using the scoring in your lead scoring model that determines which leads are sent to sellers. 
• Choose Advanced Settings: 
  • One Lead Per Account: When this setting is checked, the model reduces multiple leads from the same account that are present in the training file to one lead. This reduces overfitting in the model when they are multiple leads for some accounts, but not for other accounts. It is recommended that this setting be always checked. 
  • Include Personal Email Domains: When this setting is checked, the model will use records that have a personal email domain such as gmail or yahoo to create the model. It is recommended that this setting be always checked. 
  • Use D&B Data Cloud Attributes: When this setting is check your model will use all attributes within the D&B data cloud that are available in a Rev.Up tenant within the model build. It is recommended that this setting be always checked. 
  • Enable Transformations: When this setting is checked the model will use spam indicator attributes to score records. The purpose of these attributes to handle records that appear to be spam. This is often the case for companies who have content that users want to download and use, but do not want to be engaged by a sales person or marketing. Records that are considered spam often have fake or incorrect company names, websites and other company information and are indicators that a user only wanted a content download. It is recommended that this setting be checked with you have a large amount of leads that you need to score that may be spam. 
  • Some Examples of attributes that you would see in your model using this flag are: 
    • Domain_Length: The length of the value that was mapped to the domain field in the user's training file. When values for this field are not typical this is an indicator that the someone put in information simply to download the content. 
    • Email Length: The length of the value that was mapped to the email field in the user's training file. When values for this field are not typical this is an indicator that the someone put in information simply to download the content. 
    • Email Prefix Length: The length of the value before the email in the user's training file. When values for this field are not typical this is an indicator that the someone put in information simply to download the content. 
    • Invalid Email: A check on whether it is a valid email format. When it is not valid, it is an indicator that someone just wants to download content. 
    • CompanyName_Length & CompanyName_Entrophy:  A check on the name of the company name and whether it matches the email provided. When values for this field are not typical this is an indicator that the someone put in information simply to download the content. 

Step 3: Field Mapping 

Map fields from the training file to fields that will be used by the model to match records in the training file to the D&B data cloud. The more information that can be provided the better the match to the data cloud will be. 

Required Fields for Account models are: 

• Event: A column in the file that indicates if the record is considered a success event or not. 
• Id: A unique id for the record in the file. This will be used to removed duplicate records from the model build. 
• You can provide the website and company name or just the DUNS alone. 

Required Fields for Lead models are: 

• Event: A column in the file that indicates if the record is considered a success event or not. 
• Id: A unique id for the record in the file. This will be used to removed duplicate records from the model build. 
• You can provide the email and company name or just the DUNS alone. 

Step 4: Build Model 

The model will build in 2 - 4 hours. After the model has been built you can come and review the model results. 

Next: 

• How To Review A Lattice Predictive Insights Model (LPI)