Lookalike audience selection models

Algorithm name : Lookalike audience selection

Applies to: B2B (contact, account), B2C (customer)

The Lookalike audience selection model empowers marketers to identify potential customers who share characteristics with their existing successful customers.

What is lookalike audience selection model?

The Lookalike audience selection model empowers marketers to identify potential customers who share characteristics with their existing successful customers. This facilitates efficient targeting of prospects, customers, or accounts that exhibit similarities to the best-performing segments, thereby optimizing the audience selection process using profile and behavioral information to enhance conversion rates and ROI.

Some of the key use cases are listed below:

  • Targeting similar prospects: Identify prospects that resemble your best customers or accounts.

  • Exclusion of unresponsive prospects: Exclude non-responsive prospects or customers from your campaigns.

  • Enhanced audience selection: Continuously refine your audience selection process based on ongoing customer engagement.

Benefits of leveraging Lookalike audiences:

  • Data-driven audience selection: Utilize data to make informed decisions about audience targeting.

  • Improved conversion rates and ROI: Increase the efficiency of your marketing efforts by targeting high-potential prospects.

  • Enhanced targeting precision: Achieve more precise targeting with lookalike audiences, improving overall campaign performance.

Understanding the segment inputs to the model

Lookalike audience selection model requires you to set up two segments as inputs – a seed and a target segment.

  • Seed segment - The Lookalike audience modeling process begins with the identification of a reference or seed set. This set comprises customers or accounts that have performed the specific action you are targeting for new lookalikes. For instance, as a marketer, you might select a seed segment of customers who have purchased a particular product or service. The model will then search for other customers who exhibit similar characteristics and are likely to buy the same product or service in the future.

  • Target segment - The target segment refers to the audience of customers or prospects within which you want the model to identify and score potential lookalikes. For example, a target segment could be defined as ‘youth aged between 20-25 years in North America’. This helps the model focus on finding lookalikes within a specific demographic, rather than searching the entire customer base.

Generating the Seed and Target segments

The seed and target segments need to be generated based on one of the two base objects: Mastercustomer or Masteraccount. The choice between these base objects depends on your specific lookalike use case.

  • Mastercustomer segments: Use these segments if you are planning to build a customer or contact-based lookalike model. This is ideal for scenarios where individual customer behavior and characteristics are the focus.

  • Masteraccount segments: Use these segments if you are aiming to create account-based lookalikes. This is suitable for B2B use cases where the target is entire accounts rather than individual customers. You can opt to exclude the seed segment as a criterion in your target segment to ensure your seed audience does not appear in your lookalikes. This helps maintain the distinction between your existing successful customers and the new potential lookalikes.

By selecting the appropriate base object, you ensure that the lookalike model is tailored to your specific needs, whether it's targeting individual customers, contacts or entire accounts.

Important considerations for segment sizes

  • Seed Segment Size: Ensure your seed segment is large enough to produce robust model outcomes. For instance, if your goal is to obtain 100K lookalike customers, it is recommended to have at least 50K customers in your seed segment.

  • Target Segment Size: Make sure your target segment is larger than the required lookalike count. In the example above, for 100K lookalike customers, the target segment should exceed this number.

  • Data Records Limit: To maintain model efficiency, limit the overall data records (including behavioral data) to 5 million.

Note: The actual profile count in the model may be lower than the count observed on the Segmentation canvas. This discrepancy occurs because the model query excludes customers who do not have sufficient data points for the model to effectively learn from.

Configuring model parameters

The lookalike model can be tailored for different outcomes using the following parameters:

  • Lookalike Type: Select the appropriate type for your use case – 'B2C Customer', 'B2B Contact', or 'B2B Account'. This is important to determine the granularity of the data inputs.

  • Lookback Window: Set between 30 to 720 days to define the historical behavioral data needed for training the model.

  • Requested Lookalike Size: Specify up to (and not more than) three times the size of your seed segment. Ensure this number is less than or equal to the target segment size. This helps define the number of lookalikes expected from the model

  • Score Thresholds: Define the similarity score threshold for lookalikes. Higher values yield lookalikes with greater similarity to the seed segment, while lower values increase the number of lookalikes.

  • Score Bins: Organize lookalikes based on scores (3-bins - high, medium, low) or population distribution (4-bins - quartiles 1, 2, 3, and 4).

Ensuring adequate data inputs to the model

The lookalike model analyses both the profile and behavioral data to identify lookalikes. Hence, it is important to ensure you have greater data coverage across the below mentioned attributes for robust outcomes. The attributes required for the model are listed here.

Creating lookalike audience selection model

Please note: Before creating the Lookalike Audience Selection model, it is crucial to create the model output object in your tenant.

Follow the steps below to create the lookalike audience selection model

  • Navigate to the Oracle Unity home page and access the ‘Intelligence Workbench’.

  • Under the ‘Models’ tab, click on ‘Create model’ to start the creation of the Lookalike audience model

  • Provide a name and descriptive summary for the model in the ‘Details’ section

  • In the ‘Algorithm’ section, choose the ‘Lookalike Audience Selection’ algorithm. This will display the parameter configuration section.

  • Select the appropriate parameters as outlined in the ‘Configuring Model Parameters’ section. Ensure that the parameters align with your specific use case for optimal performance.

  • Under the ‘Segments’ section, choose the appropriate seed and target segments to input into the model.

  • Before saving the model, go to the ‘Schedule’ tab and select the appropriate cadence for rescoring new lookalikes. This ensures that the model updates periodically to reflect the latest data.

  • Once all configurations are complete, save the model to finalize the creation process.

The model requires a publish before it can be trained and scored for the first time.

Accessing lookalike model outputs

Lookalike model outputs are integrated into the Unity data model and can be accessed through specific data objects:

  • Customer/Contact Lookalikes: Use ‘CustomerLookalike’ or ‘Lookalike_audience’ to refer to lookalike audiences based on individual customers or contacts.

  • Account Lookalikes: Use ‘AccountLookalike’ or ‘Lookalike_account’ to refer to lookalike audiences based on entire accounts.

These outputs can be utilized within the segmentation canvas, allowing for precise audience segmentation based on lookalike attributes.

‘CustomerLookalike’ data object

Attribute ID

Attribute Name

Attribute Description

Data type

SourceCustomerLookalikeID

Source CustomerLookalike ID

This attribute contains the unique ID for the object.

STRING

LookalikeScore

Lookalike Score

This attribute represents the lookalike score to the seed audience.

FLOAT

LookalikeBin

Lookalike Bin

This attribute represents the bin on the look alike score.

STRING

ModelName

Model Name

This attribute represents the lookalike model name.

STRING

SeedSegment

Seed Segment

This attribute represents the Segment used as a seed to the lookalike model.

STRING

TargetSegment

Target Segment

This attribute represents the segment used as a target universe to find lookalikes in.

STRING

MasterCustomerID

Master Customer ID

This attribute contains the foreign key to the MasterCustomerID.

STRING

SourceMasterCustomerID

Source Master Customer ID

This attribute contains the original form of the MasterCustomer ID  from the source data system.

STRING

‘AccountLookalike’ data object

Attribute ID

Attribute Name

Attribute Description

Data type

SourceAccountLookalikeID

Source AccountLookalike ID

This attribute contains the unique ID for the object.

STRING

LookalikeScore

Lookalike Score

This attribute represents the lookalike score to the seed audience.

FLOAT

LookalikeBin

Lookalike Bin

This attribute represents the bin on the look alike score.

STRING

ModelName

Model Name

This attribute represents the lookalike model name.

STRING

SeedSegment

Seed Segment

This attribute represents the engagement score of the customer.

STRING

TargetSegment

Target Segment

This attribute represents the segment used as a target universe to find lookalikes in.

STRING

MasterAccountID

Master Account ID

This attribute contains the foreign key to the MasterAccountID.

STRING

SourceMasterAccountID

Source Master Account ID

This attribute contains the original form of the Master Account ID  from the source data system.

STRING

Using lookalike model outputs

Lookalike model outputs are integrated into the Unity data model and can be utilized like any other data attribute within the segmentation canvas. This allows marketers to create precise audience segments based on lookalike attributes. Below is an example illustrating how to use these outputs.

Example: Filtering Audience with Lookalike attributes

  • Seed segment selection: Choose a seed segment, such as 'Seed-FY23Purchaser', which consists of customers who have made purchases in FY23.

  • Similarity score threshold: Apply a filter to select lookalikes that have a similarity score of 60 or higher.

This setup filters the audience to include only those lookalike individuals who are highly similar to the specified seed segment, ensuring targeted and relevant marketing efforts.

Model inputs

For b2c customer/ b2b contact lookalikes, the following attributes are required as input to the model.

Attribute from Unity Data Object

Unity Data Object

Description of the attribute

Data type

Must have?

ID

MasterCustomer

Mastercustomer identifier

String

Yes

Gender

MasterCustomer

Gender of the customer

String

 

Age

MasterCustomer

Age of the customer

INT

 

City

MasterCustomer

City name

String

 

State

MasterCustomer

State name

String

 

Country

MasterCustomer

Country name

String

 

okToEmail

MasterCustomer

Marketing Permissions

String

 

okToText

MasterCustomer

Marketing Permissions

String

 

Customer ID

Customer_MasterCustomer

Customer identifier

String

 

JobTitle

Customer

Job title or role

String

 

JobTitleLevel

Customer

Level of the job title

String

 

JobDepartment

Customer

Department to which the employee belongs to

String

 

Source Customer ID

Event

Customer identifier from the source

String

 

Source Event ID

Event

Event identifier from the source

String

 

EventID

Event

Event identifier

String

Yes

Type

Event

Event type indicating an email click/ web view/ email subscribe, etc. 

String

Yes

Event Timestamp

Event

Date when the event was captured

Timestamp

Yes

Source

Event

Source of the event

String

 

Source ID

Event

Source ID of the event

String

 

Source Category ID

Event

Unique category Id from the source

String

 

Category ID

Event

Unique category Id from the source

String

 

Order Total

Event

Total revenue in the order

Float

 

Medium

Event

Medium/ Channel of the event

String

Yes

Source Order ID

OrderItem

Order id from source

String

 

Source Product ID

OrderItem

Product ID in the order from source

String

 

Product ID

OrderItem

Product ID in the order

String

 

Quantity

OrderItem

Quantity of items purchased

INT

 

Extended price

OrderItem

Price paid for the order

INT

 

 

 

 

 

 

  

For b2b account lookalikes, the following attributes are required as inputs to the model.

Attribute from Unity Data Object

Unity Data Object

Description of the attribute

Data type

Must have?

ID

MasterAccount

Masteraccount identifier

String

Yes

Country

MasterAccount

Account’s country

String

 

ZipCode

MasterAccount

Account’s zipcode

String

 

SourceMasterAccountID

MasterAccount

Masteraccount identifier from the source

String

 

State

MasterAccount

Account’s state

String

 

City

MasterAccount

Account’s city

String

 

DNBLocationType

Account

Location type provided by D&B. For example - Single Location, Headquarters, Branch etc.

String

 

OwnershipType

Account

Account’s ownership type

String

 

LineOfBusiness

Account

Account’s line of business

String

 

Type

Account

Account type

String

 

EmployeeTotal

Account

Total number of employees associated with the account

int

 

AnnualRevenue

Account

Account’s annual revenue

int

 

SourceID

Account

Source ID

String

 

CustomerID

Customer

Customer identifier

String

 

SourceCustomerID

Customer

Customer Id from the source

String

 

JobTitle

Customer

Job title or role

String

 

JobTitleLevel

Customer

Level of the job title

String

 

JobDepartment

Customer

Department to which the employee belongs to

String

 

SourceEventID

Event

Event identifier from the source

String

 

EventID

Event

Event identifier

String

Yes

Type

Event

Event type indicating an email click/ web view/ email subscribe, etc. 

String

Yes

EventTS

Event

Date when the event was captured

Timestamp

Yes

Source

Event

Source of the event

String

 

Medium

Event

Medium/ Channel of the event

String